Warehouse activity profiling

Diagnose the distribution. Design for the observed work.

Foundations

01

The analysis answers four warehouse questions

• Which customers, SKUs, and locations generate the work?
• Which orders require distinct picking and handling flows?
• When does released workload exceed normal capacity?
• Which warehouse changes deserve a quantified operating test?

Activity profiling converts transaction data into distributions that change a warehouse decision. Useful profiles reveal concentration, mixture, scale, timing, or exceptions.

Warehouse activity profiling and data mining is the systematic analysis of item and order activity, seeking to identify and take advantage of patterns of opportunity. Our pattern recognition algorithms highlight the root cause of material and information flow problems, pinpoint major opportunities for process improvements, and provide an objective basis for project-team decision making.

I will start with some of the major motivations and potential roadblocks to successful profiling. Then I will review a full set of example profiles and their interpretations. The examples will serve to teach the principles of profiling and as an outline for the full set of profiles required for warehouse reengineering.

— World-Class Warehousing and Material Handling (Frazelle 2016, 21–22).

Diagnose the workload before choosing equipment

1. Validate transactions. Connect orders, items, inventory, and locations.
2. Describe the workload. Expose concentration, mix, scale, and timing.

3. Locate causes. Trace material-flow and information-flow problems.
4. Test a redesign. Screen process, storage, and MHE options before selecting an experiment.

Design rule

Diagnose the warehouse activity before prescribing equipment, storage, or picking methods.

Imagine that you were sick and went to a doctor for a diagnosis and prescription. When you arrived at the doctor’s office, he already had a prescription waiting for you, without even talking to you, let alone looking at you, examining you, or doing blood work. In effect, he wrote your prescription with his eyes closed and a random prescription generator. Needless to say, you would not be going back to that doctor for treatment.

Unfortunately, the prescriptions for many sick warehouses are written and implemented without much examination either. For lack of knowledge, lack of tools, and/or lack of time, many warehouse projects launch without any understanding of the root cause of the problems and without exploration of the real opportunities for improvement.

Warehouse activity profiling and data mining is the systematic analysis of item and order activity, seeking to identify and take advantage of patterns of opportunity. Our pattern recognition algorithms highlight the root cause of material and information flow problems, pinpoint major opportunities for process improvements, and provide an objective basis for project-team decision making.

I will start with some of the major motivations and potential roadblocks to successful profiling. Then I will review a full set of example profiles and their interpretations. The examples will serve to teach the principles of profiling and as an outline for the full set of profiles required for warehouse reengineering.

— World-Class Warehousing and Material Handling (Frazelle 2016, 21–22).

Nine disciplines keep profiling decision-ready

Understand the work

• Preparation: study the operating pattern
• Participation: validate with warehouse teams
• Pictures: make every figure self-contained

Challenge the design

• Power: eliminate methods that do not fit
• Probity: test preferred equipment against data
• Principle: measure concentration; never assume it

Decide at the right depth

• Purpose: design for distributions and peaks
• Perspectives: connect activity to space
• Pitfall: stop when the next test is clear

Operating principle

A useful profile narrows the warehouse design space and defines the next operating test.

I have been leading, facilitating, researching, and teaching activity profiling and data mining for more than 20 years. I recently summarized our lessons learned as the P’s and P’s of profiling.

The Preparation of Profiling. The first thing I do before writing a new article is to read what others have written about the topic. If I am preparing to teach a class or a seminar, I do the same thing: I review what others have prepared on the topic to stimulate my thinking. Reviewing activity profiles works the same way; it is like reading an illustrated journal of the warehouse activity. As you study the profiles of customer orders, purchase orders, item activity, inventory levels, and so on, the creative juices should begin to flow as your understanding of underlying activity improves.

The Power of Profiling. Done properly, the patterns in profiling quickly reveal a small set of candidate process flows and material handling equipment types for each warehouse activity. Conversely, profiling quickly eliminates process and equipment options that really do not warrant consideration. Many warehouse reengineering projects go awry because we waste valuable resources analyzing concepts that were doomed from the start.

The Participation of Profiling. Profiling involves key stakeholders. During the profiling process, it is natural to ask people from many affected groups to provide data, to verify and rationalize data, and to help interpret results. To the extent those stakeholders have been involved, they have helped with the design process.

The Probity of Profiling. Profiling supports and promotes objective decisions. I worked with one client whose team leader we somewhat affectionately called “Captain Carousels.” No matter what the data said, no matter what the order and profiles looked like, no matter what the company could afford, he was going to have carousels in the new design.

The Purpose of Profiling. You will see myriad complex statistical distributions in our journey through warehouse activity profiling. Why go to all the trouble? Suppose that we are trying to determine the average number of items on an order. Figure 2.1 shows that 50 orders are for one item, 0 are for two items, and 50 are for three items. What is the average number of items per order? It’s two. How often does that happen? It never happens! If we are not careful to plan and design based on distributions as opposed to averages, our plans and designs will be flawed.

The Principal and Principle of Profiling. Vilfredo Pareto was an Italian economist and gardener. Pareto’s law stems from his observation that 80 percent of the land was owned by 20 percent of the population and that 20 percent of the peapods in his garden yielded 80 percent of the peas. Joseph Juran called this principle the “separation of the vital few from the trivial many.” Activity profiling and data mining are essentially the repeated and logical application of Pareto’s law, sometimes referred to as the 80/20 principle or ABC analysis. Examples of Pareto’s law working in warehousing include

• A minority of the items generate a majority of the activity.
• A minority of the items are responsible for a majority of the inventory.
• A minority of the customers generate a majority of the picking and shipping activity.

The Pictures of Profiling. When you see a picture you experience a thousand simultaneous thoughts. We are aiming for the same effect in warehouse activity profiling as we paint a representative picture of warehouse activity. In profiling, we capture the activity of the warehouse in pictorial form so that we can quickly develop wise, data driven, consensus decisions.

The Perspectives of Profiling. If a picture is worth a thousand words, then a video is worth a thousand pictures. Interactive data visualization, simulation, and animation are the closest media we have to video for activity profiling. Interactive data visualization (Figure 2.2) is simply dynamic charting controlled by the user. Simulation (Figure 2.3) is typically a two-dimensional representation of the physical warehouse with the activity profiles moving warehouse objects according to the profiles. Animation (Figure 2.4) is three-dimensional simulation.

The Pitfalls of Profiling. One warning before we begin to profile the warehouse (as an engineer and logistics nerd, I fall into this trap a lot): We can drown in our own profiles. Some people call this paralysis of analysis! If we are not careful, we can get so caught up in profiling itself that we forget to solve the problem.

— World-Class Warehousing and Material Handling (Frazelle 2016, 22–25).

The mean describes no order

Figure 1: No order equals the two-item average in this 100-order sample.

Two operating modes

• 50 orders contain 1 item.
• 50 orders contain 3 items.
• The 2-item mean never occurs.

Design separate handling rules for the two recurring order types.

The Purpose of Profiling. You will see myriad complex statistical distributions in our journey through warehouse activity profiling. Why go to all the trouble? Suppose that we are trying to determine the average number of items on an order. Figure 2.1 shows that 50 orders are for one item, 0 are for two items, and 50 are for three items. What is the average number of items per order? It’s two. How often does that happen? It never happens! If we are not careful to plan and design based on distributions as opposed to averages, our plans and designs will be flawed.

— World-Class Warehousing and Material Handling (Frazelle 2016, 23–24).

The case contains 2.25 million order lines

The analysis uses anonymized 2004 operating records from an office-products distributor:

Table 1: Population used by the activity profiles.

Population	Coverage
Order lines	2,249,630
Customer orders	883,910
Active SKUs in sales	31,423
Customers	1,097
Study period	January–December 2004

The source includes order identifiers, customers, SKUs, suppliers, shipped quantities, timestamps, dimensions, case packs, and monthly inventory (Bartholdi and Hackman 2019).

The dataset was published for educational use with a Georgia Tech warehouse project. Customer and SKU identifiers remain anonymized in the profiles. Customer calendar profiles use the order transaction timestamp as an order- release proxy. Purchase-order profiles group positive supplier-SKU demand by week and round replenishment to complete cases. Dock, pick-start, and ship- confirm events are unavailable. Slides using substitute fields or modeled scenarios state that limitation in the speaker notes and beside the relevant figure.

This is a project to accompany the course and the book Warehouse & Distribution Science, at the Georgia Institute of Technology, by John J. Bartholdi III and Steven T. Hackman. Everyone is welcome to use the book and materials for educational purposes, as long as all copyrights remain intact.

The company data is copyrighted and proprietary. You may use it for the purposes of this course only. If you would like to use it for something else, please contact me to discuss.

— Warehouse & Distribution Science: A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., secs. “A project for S. P. Richards Company” and “Data”).

Overnight service makes peak release the constraint

Business and service

• Customer promise and order cutoff
• Special handling and product restrictions
• Seasonality and SKU introduction rate
• Receipts and shipments per operating day

Work and capacity

• Warehouse area and storage modes
• Pick lines, orders, and handling units per day
• Units, cases, and pallets per pick line
• Pickers, shifts, docks, and material-handling equipment

Service consequence

The distributor accepts late orders and ships overnight. That service promise makes peak-hour picking, route completion, and workload concentration more important than annual averages.

Baseline purpose. Profiles need an operating context. Before charting, state what the business does, who the customers are, the service promise, and any special handling requirements. Record warehouse area, storage modes, and material-handling equipment.

Workload measures. Record the number of SKUs; average pick lines per day; average pieces, cases, and pallets per pick line; orders shipped per day; order pickers; shifts allocated to pallet, case, and broken-case work; receipts per day; new-SKU introduction rate; and seasonality. Use distributions and peak periods after establishing these baseline measures.

Company relevance. The S. P. Richards case is an office-products distribution center with stable customers, late order cutoffs, overnight fulfillment, and cart tours covering more than 100 aisles. A typical cart contains roughly 50 pick lines. These operating conditions make routing, batching, and peak-period capacity central decisions (Bartholdi and Hackman 2019).

Frazelle. The baseline-statistics questions precede the profile families (Frazelle 2016, 28–29).

Orders, lines, and SKUs answer different questions

Table 2: The unit of analysis determines which warehouse question the profile can answer.

Observation	Typical records	Profiles supported
Order header	Customer, release time, service requirement	Order size, mix, and timing
Order line	SKU, quantity, UOM, status, lot or serial	Handling work and increments
SKU	Unique stocked item and packaging data	Popularity, cube, and inventory

The three rows are a course synthesis of database grain. An order header defines the transaction; an order line records one requested SKU within that transaction; and the SKU master supplies the item’s descriptive, packaging, and storage data. Changing grain changes the warehouse question that can be answered.

In the table, release time, unit of measure, status, and packaging are fields documented in the S. P. Richards project. “Service requirement” and “lot or serial” are course examples of fields that may distinguish otherwise similar orders or lines; they are not fields reported in that project.

Item master, with the following fields:

• Unique sku ID (text)
• Short text description (text)
• Vendor (text)
• Length in inches of storage container (2 decimal places)
• Width of storage container (2 decimal places)
• Height of storage container (2 decimal places)
• Selling units per storage container (2 decimal places)
• Total number of requests (customer orders) during study period (1 year?)
• Total number of selling units requested during study period
• Maximum on-hand inventory (either in selling units or in storage units, whichever is more convenient)
• Current location(s): Zone, aisle, section, shelf, position

This will be used to support computation of re-slotting Zones B, C, and F

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Data,” item 1).

An Excel spreadsheet (1.3MB) with status and dimensions of the skus. Dimensions are given for both eaches and for cartons but also, in many cases, for an “inner pack” as well. Recall that this is a packing level between the carton (case) and the each (selling unit). For example, Pens are sold by the each but come packaged in inner packs of 12; and the inner packs come in cases of 12 inner packs (144 eaches). Mostly, SPR does not use the inner pack information but it might be useful to us in those instances when case information is not available. The fields are mostly self-explanatory:

• Stocked: Possible values are Y (yes), N (no and may be ignored, or X (new item that will be stocked but has no sales history)
• OMS status: Possible values are A (active), or D or X (discontinued) or P (purged). All skus marked D, X, or P may be ignored.
• SellingUOM: Selling unit of measure.
• SelLen, SelWid, SelHgt, SelWgt: Length, width, height, weight of the selling unit in inches and pounds
• ShipSelling: Possible values are Y (Selling UOM is a corrugated box that can be labeled and shipped without repacking
• Inner1Qty: The number of SellingUOM’s in an inner pack
• Inner1Len, Inner1Wid, Inner1Hgt, Inner1Wgt: Length, width, height, and weight of the inner pack, in inches and pounds
• Case Qty: The number of SellingUOM’s in a case
• CaseLen, CaseWid, CaseHgt, CaseWgt: Length, width, height, and weight of a case, in inches and pounds

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “status and dimensions of the skus”).

Additional sku data in the form of a text file (1.6MB). This data complements (and, in some ways, duplicates) the previous. New fields include:

• Vendor ID: From whom the item is purchased
• Hazmat Code: Whether item is hazardous and so requires special handling. May be ignored.
• Selling UOM:
• Stocking Status:
• Item Status:
• Selling UOM Length, Width and Height: in inches
• Case UOM Length, Width and Height: in inches
• Selling UOM/Case:
• Quantity Ordered: Measured in selling units
• Lines Ordered: That is, picks, requests
• Max Order Size: Most units on an order-line
• Average Order Size
• Max Case Order: Most number of cases on an order-line
• Average Case Order: Averaged over all order-lines requesting at least one case
• Case Lines Ordered: Number of lines for at least one case
• Max Order No Case: The maximum quantity on a line for lines that were less than a case
• Average Order No Case: Average line quantity for those lines that were less than a case
• The last columns display amounts on-hand. There are two columns for each month. The AVG column is the average of the inventory levels over all days of the month. For example, “janavg” gives the average inventory levels over the days of January
  2004. The OH is the inventory level on the last day of the month. For example, “janoh” gives the inventory level on 31 January 2004.

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “Additional sku data”).

Order history over, say, 3 months, with the following fields:

• Unique order ID (text)
• Unique customer ID (text)
• Date (date format)
• Time received (time format)
• Date released to picking (date format)
• Time released to picking (time format)
• Unique sku ID requested (text)
• Quantity (integer)
• Unique ID of assigned batch
• Size of box assigned to hold order

This will be used to address some of the other questions you raised, such as about how to batch pick-lines, whether to pick into a tote or a box, etc. Note that it was important to avoid December and January in this order history because during those months, and only then, calendars are the fastest moving items. You must always be careful of seasonalities!

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Data,” item 2).

A sales history (text file; 59MB) containing the following fields:

• item: Unique sku ID
• vnid: Vendor ID
• itdesc: Text description of sku
• ordnbr: Unique ID of this order
• linenbr: Number of the line in this order
• cusnbr: Unique ID of customer
• ordqty: How many selling units ordered
• shipqty: How many selling units shipped (may be less than ordered if, for example, out of stock. Should never be more.)
• uom: Selling unit-of-measure
• txndate: Order data and time

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “sales history”).

Product family may have implications for storage and/or handling. These tend to be particular to an industry and so require knowledge of the context. For example, product families for a drug store chain might include hair care products, dental products, shaving products and so on, which are displayed together at the retail store. For a grocery distributor product families might include dry goods, dairy, produce, refrigerated, frozen and so on. For a candy distributor product families might include chocolate (sensitive to heat), mint-flavored candies (odoriferous), and marshmallow (light and tends to absorb the smells of its neighbors), and so on. For an apparel distributor product families might include garment type, mill, style, color, or size. Note that a SKU might be in more than one product family.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, p. 237).

Order history. The order history is simply a concatenation of all the shopping lists submitted by all the customers during the preceding year. It contains the following information.

• Unique ID of this order, to distinguish it from the shopping lists of other customers and from the shopping list of the same customer on another day or later on the same day.
• Unique ID of SKU, which allows us to look up the SKU to see where it is stored.
• Customer.
• Special handling.
• Date/time order picked.
• Quantity shipped.

For analyzing warehouse operations you have to be careful where you obtain this data. Often this data is from a sales transaction database, which tracks financial rather than warehouse events. Consequently, the date recorded may represent the date the order was placed or when it was printed, not when it was processed in the facility. Similarly, the order that appears in the sales transaction database may actually have been processed at another facility. Generally, though, the information is available somewhere because each day the order pickers had to know what to pick.

Keep in mind that an order history is primarily financial information. This is good in that it is likely to be very accurate; but it can also be misleading because the transactions it represents are financial and not necessarily operational. For example, it can happen that a SKU is shown to have been requested in a negative amount; but this generally means something like the item was returned and restocked to the shelf.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, pp. 238–240).

The essential functionality you will need are the abilities to:

• Sort the rows of a table;
• Select a subset of rows of a table, such as all the order-lines from a single day;
• Count distinct entries in a table, such as the number of times each SKU appears in the order history;
• Connect the row of one table with corresponding rows in another table (for which the database jargon is join). An example is connect order-lines (rows in the orders table), through the unique SKU identifier, with additional information about each SKU (rows in the SKUs table), such as where it is stored;
• Graph results.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, p. 251).

Choose the correct unit of analysis

Aggregation rule

Choose the unit of analysis before aggregating. One order, one order line, and one SKU are not interchangeable observations.

Order header. The header describes the transaction as a whole. Typical fields include order identifier, customer, billing and shipping information, release time, service requirement, payment, and special instructions.

Order line. A line identifies one requested SKU and includes quantity, unit of measure, status, lot or revision, serial number, price, and special handling. One order may contain many lines. Each unique SKU line often approximates one storage-location visit.

SKU. A stock keeping unit is the warehouse’s unique identifier for a stocked item variant. It describes inventory diversity and differs from a manufacturer’s serial number. A readable internal code may encode category and variant, but analysis should join on the stable unique identifier rather than infer product attributes from the text of the code.

Example. An order for three monitors, three keyboards, and three printers is one order header with three order lines and three SKUs. The order-level profile counts one order; the line-level profile counts three lines; item activity is allocated to three SKUs.

Frazelle. These distinctions support the separate order and item profile families (Frazelle 2016, 28–29).

Packaging definitions change the handling mix

Table 3: Packaging conversions required for handling-mix and order-increment profiles.

Handling level	Operational meaning	Required conversion
Each	Selling unit picked individually	Eaches per inner pack
Pack	Group of eaches handled together	Packs per case
Case	Standard shipping or storage carton	Eaches per case
Pallet	Unit load containing cases	Cases per pallet and layers per pallet

Warehouse convention

“Box,” “carton,” “case,” and “inner pack” are not universal levels. Confirm the company’s definitions before calculating full-case or pallet fractions.

It is particularly important to understand the conventions the warehouse uses to distinguish among different types of storage units and selling units. For example, the word “case” is often called by other names (“carton”, “box”) and, depending on its use, can have substantially different meanings for a warehouse. For example, a vendor may ship a case that contains several inner packs each of which contains several boxes each of which contains pieces (Figure 2.2). A standard example is an office products distributor supplying a standard type of ball point pen. The manufacturer may supply the product 12 pieces to a box (as you find it in the store), 12 boxes to an inner pack (stored in a thin carton container), and 4 inner packs to a case for a grand total of 576 pens in the vendor’s case or shipping unit. While each of the terms each, box, inner pack, case, shipping unit are commonly used, there is no convention as to which level of packing they apply.

It is important to understand how this packing data is stored in the database. Often, the retail customer (of the distributor) is required to purchase an integer multiple of a selling unit. For the pen the selling unit may be an each, which means that a customer can actually order less than a full box of 12 pens. In the database this information may be stored as either the number “1” or by the symbol “EA” for each. If the customer is required to purchase boxes, then the selling unit may be listed as “12” or “box”. Now suppose the database records a customer purchase of 12, which appears on the order picker’s pick ticket? What exactly does this mean: 12 pens (1 box) or 12 boxes? If you think it means 12 pens when in fact it means 12 boxes, you would be underestimating demand by a factor of 12. If the manufacturer sells the pens as pieces, accounting will record transactions in units of a single pen, notwithstanding the restriction on the outbound side on how it is to be resold. To ensure consistency, one could always record demand in the smallest physical unit, pens, but this would require an order picker to know that 156 = 13 boxes, for example. To facilitate picking accuracy, the demand may be recorded as 13, and the order picker would know that this means 13 boxes. (Of course, if another pen is resold as pieces, this would lead to confusion.) To avoid this confusion, many facilities separate how demand is recorded for accounting purposes from how it is presented on a pick ticket, e.g., the pick ticket should say something like demand = 156 = 13 boxes of 12 each.

There is another reason why such packing data is important. It is always more efficient to store and handle product in some kind of easily-held container as opposed to loose pieces. In the case of the pen this handling unit may be the box of 12 or the inner pack of 12 boxes. For purposes of restocking a shelf of product it would be much easier to restock in units of the inner pack. For purposes of space efficiency on the shelf and order picking efficiency, it would be better to store the product as boxes neatly stacked.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 238–39).

Six linked datasets connect demand to space

Table 4: Data sources and warehouse activity profiles they support.

Data source	Essential warehouse fields	Profiles enabled
Item master	SKU, family, dimensions, case pack, status	Item, cube, handling mix
Order history	Order, customer, SKU, quantity, UOM, timestamp	Order and calendar-clock
Inventory history	SKU, on-hand, period, location	Inventory and capacity
Layout	Aisles, storage blocks, docks, travel paths	Distance and congestion
Location master	Zone, aisle, bay, level, position, SKU	Spatial activity and slotting
Storage-unit data	Address, dimensions, capacity, handling unit	Storage-mode assignment

The six-dataset grouping is an instructor-organized view. It keeps transactional demand, item attributes, stock, physical addresses, handling-unit capacity, and building geometry separate until they are joined by stable keys.

Product family may have implications for storage and/or handling. These tend to be particular to an industry and so require knowledge of the context. For example, product families for a drug store chain might include hair care products, dental products, shaving products and so on, which are displayed together at the retail store. For a grocery distributor product families might include dry goods, dairy, produce, refrigerated, frozen and so on. For a candy distributor product families might include chocolate (sensitive to heat), mint-flavored candies (odoriferous), and marshmallow (light and tends to absorb the smells of its neighbors), and so on. For an apparel distributor product families might include garment type, mill, style, color, or size. Note that a SKU might be in more than one product family.

Addresses of storage locations within the warehouse. This might include zone, aisle, section, shelf and position on the shelf.

For each location at which this SKU is stored:

• Scale of the storage unit, such as pallets or cases. This is useful in validation and error checking.
• Physical dimensions of the storage unit (length, width, height, weight), which are useful in understanding space requirements.
• Scale of the selling unit, such as cases or pieces, which is useful for validation and error-checking.
• Number of selling units per storage unit. This could be 1.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, pp. 237–238).

The essential functionality you will need are the abilities to:

• Sort the rows of a table;
• Select a subset of rows of a table, such as all the order-lines from a single day;
• Count distinct entries in a table, such as the number of times each SKU appears in the order history;
• Connect the row of one table with corresponding rows in another table (for which the database jargon is join). An example is connect order-lines (rows in the orders table), through the unique SKU identifier, with additional information about each SKU (rows in the SKUs table), such as where it is stored;
• Graph results.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, p. 251).

The six rows and the “profiles enabled” column are a course synthesis; the source supplies the operational records from which those analytical datasets can be constructed.

Item master, with the following fields:

• Unique sku ID (text)
• Short text description (text)
• Vendor (text)
• Length in inches of storage container (2 decimal places)
• Width of storage container (2 decimal places)
• Height of storage container (2 decimal places)
• Selling units per storage container (2 decimal places)
• Total number of requests (customer orders) during study period (1 year?)
• Total number of selling units requested during study period
• Maximum on-hand inventory (either in selling units or in storage units, whichever is more convenient)
• Current location(s): Zone, aisle, section, shelf, position

This will be used to support computation of re-slotting Zones B, C, and F

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Data,” item 1).

Order history over, say, 3 months, with the following fields:

• Unique order ID (text)
• Unique customer ID (text)
• Date (date format)
• Time received (time format)
• Date released to picking (date format)
• Time released to picking (time format)
• Unique sku ID requested (text)
• Quantity (integer)
• Unique ID of assigned batch
• Size of box assigned to hold order

This will be used to address some of the other questions you raised, such as about how to batch pick-lines, whether to pick into a tote or a box, etc. Note that it was important to avoid December and January in this order history because during those months, and only then, calendars are the fastest moving items. You must always be careful of seasonalities!

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Data,” item 2).

An Excel spreadsheet (1.5MB) with all of the cases that are in reserve locations for zones A,B,C,R,F,W. (This includes all zones, in case there is overstock for items in B,C, or F.) Here is what the fields mean:

• Case: The unique case id for each box
• SKU: The unique SKU in the box. Mixed cases in reserve are not allowed. So there should only be one record per case ID. (Please report if you find otherwise.)
• Quantity: This is the quantity in the box (in selling units). If the quantity is equal to the standard case quantity (which will be provided in another file), then the reserve location has a case ID on every box. This is what SPR refers to as a “case reserve” location. If the quantity is a multiple of the standard case quantity, then the reserve location has just one case ID for an entire pallet. This is what SPR refers to as a “pallet reserve” location. If it is not either, then someone has put a partial case in reserve, which they are not supposed to do, or the standard case quantity is not correct. Note, the standard case quantity is what SPR calls the “number of selling units per master case”. In reserve, SPR stores only master case quantities. There are no open boxes.
• Zone, Aisle, Bay, Level (shelf), Position: The definition of the reserve location
• Location Type: This also indicates whether a location is a Case Reserve or Pallet Reserve location. BLU and KCN are codes that refer to a Case Reserve location, with a case ID on every box. PLC and PLP are Pallet Reserve locations, where pallet quantities are stored with a case id for the entire pallet quantity. Note that for PLC locations the users can pull boxes off a pallet, just reducing the amount assigned to the case number. For PLP locations, we pull only full pallets. There are no broken pallet quantities. These are only for the few items that we sell consistently by the full pallet.

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” reserve-locations subsection).

An Excel spreadsheet (0.6MB) with all of the active locations in all zones (again, more data than we may need). There should only be one location per SKU and one SKU per location. The fields should be self-explanatory. The quantity is the current quantity, in selling units, that exists in the active location as of 01 February 2005.

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” active-locations subsection).

An Excel spreadsheet (1.3MB) with status and dimensions of the skus. Dimensions are given for both eaches and for cartons but also, in many cases, for an “inner pack” as well. Recall that this is a packing level between the carton (case) and the each (selling unit). For example, Pens are sold by the each but come packaged in inner packs of 12; and the inner packs come in cases of 12 inner packs (144 eaches). Mostly, SPR does not use the inner pack information but it might be useful to us in those instances when case information is not available. The fields are mostly self-explanatory:

• Stocked: Possible values are Y (yes), N (no and may be ignored, or X (new item that will be stocked but has no sales history)
• OMS status: Possible values are A (active), or D or X (discontinued) or P (purged). All skus marked D, X, or P may be ignored.
• SellingUOM: Selling unit of measure.
• SelLen, SelWid, SelHgt, SelWgt: Length, width, height, weight of the selling unit in inches and pounds
• ShipSelling: Possible values are Y (Selling UOM is a corrugated box that can be labeled and shipped without repacking
• Inner1Qty: The number of SellingUOM’s in an inner pack
• Inner1Len, Inner1Wid, Inner1Hgt, Inner1Wgt: Length, width, height, and weight of the inner pack, in inches and pounds
• Case Qty: The number of SellingUOM’s in a case
• CaseLen, CaseWid, CaseHgt, CaseWgt: Length, width, height, and weight of a case, in inches and pounds

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “status and dimensions of the skus”).

Additional sku data in the form of a text file (1.6MB). This data complements (and, in some ways, duplicates) the previous. New fields include:

• Vendor ID: From whom the item is purchased
• Hazmat Code: Whether item is hazardous and so requires special handling. May be ignored.
• Selling UOM:
• Stocking Status:
• Item Status:
• Selling UOM Length, Width and Height: in inches
• Case UOM Length, Width and Height: in inches
• Selling UOM/Case:
• Quantity Ordered: Measured in selling units
• Lines Ordered: That is, picks, requests
• Max Order Size: Most units on an order-line
• Average Order Size
• Max Case Order: Most number of cases on an order-line
• Average Case Order: Averaged over all order-lines requesting at least one case
• Case Lines Ordered: Number of lines for at least one case
• Max Order No Case: The maximum quantity on a line for lines that were less than a case
• Average Order No Case: Average line quantity for those lines that were less than a case
• The last columns display amounts on-hand. There are two columns for each month. The AVG column is the average of the inventory levels over all days of the month. For example, “janavg” gives the average inventory levels over the days of January
  2004. The OH is the inventory level on the last day of the month. For example, “janoh” gives the inventory level on 31 January 2004.

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “Additional sku data”).

A sales history (text file; 59MB) containing the following fields:

• item: Unique sku ID
• vnid: Vendor ID
• itdesc: Text description of sku
• ordnbr: Unique ID of this order
• linenbr: Number of the line in this order
• cusnbr: Unique ID of customer
• ordqty: How many selling units ordered
• shipqty: How many selling units shipped (may be less than ordered if, for example, out of stock. Should never be more.)
• uom: Selling unit-of-measure
• txndate: Order data and time

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “Here is what we have received so far,” subsection “sales history”).

Most of the activity in Philadelphia is in zones B and C. Zone B is the blue shelving in the figure above and this is the primary picking area. The cluster on the left is 30-inches deep and 42-inches wide. The cluster on the right is 18-inches deep and 42-inches wide. Zone B was intended to hold the fastest moving of the small products. Zone C is the red shelving, some 18-inches deep, others 30-inches deep, and all 42-inches wide. Zone C was intended to hold slower moving small products. Currently zones B and C are picked independently of each other and the two corresponding parts of any order are shipped separately. However, as this has not produced the hoped-for efficiencies, the order-picking for these two zones will be combined, so that each picker will travel through both zones if necessary. SPR has been puzzled that the lines picked per person hour have been approximately the same in zones B and C. Is this because of wrong choice of product to store in B? Or because of congestion? Or is there some other reason?

Zone F is pallet rack that has been configured to hold both full pallets and, in some locations, cartons. Zone F is the gray racking on the bottom left of the figure and also the purple racking that encircles zone B in a sort of horseshoe shape. The gray area of zone F is for active pick locations of products that are stored and picked as eaches, such as brooms or garbage cans. The purple region is intended to hold overstock for zones F and B but it also holds some other items that are both stored and picked as eaches.

Here is a legend for the map above:

• The blue area is Zone B. The lighter blue is 30-inch shelving and the darker (navy) blue is 18-inch shelving. The top shelf is reserve. (The reserve locations on the top shelf always use Shelf/Level = T and Position = T so that they are distinguishable in the database by the name.) The other locations are active picking locations.
• The purple area is Zone F reserve locations and the gray is Zone F active picking locations. All of Zone F is pallet rack.
• The yellow area is pallet racking used to hold packing and shipping supplies.
• The red area is 18-inch shelving in Zone C and the orange is 30-inch shelving in Zone C.
• The green are is Zone R. The lighter green represents 30-inch shelving and the darker green represents 18-inch shelving.

Other features of interest in the layout:

• The pickup station, where orders are released is the first yellow bay from the right.
• The dropoff point where completed orders are checked and packed is at the black conveyors on the left.

Each order-picker pushes a cart and travels a complete circuit through the main aisles of the zone, picking a batch of orders. Workers pick directly into final packaging, which is cardboard boxes. Any full-case quantities are pulled from reserve. Thus if a customer requests 13 pens and the pens are packed 12 to a carton, there will be one pick from reserve (of 1 carton) plus one pick from the active pick area (of 1 pen).

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “The company and its DC operations,” layout of zones B, C, and F).

As in any real operation, the warehouse may be changing in some ways even as we try to improve it. For example, here are two changes under current study. In the circled region on the left of the figure below, you can see that one line of shelving has been removed (formerly blue, now clear) and the pick-path has been changed to be unidirectional, in hopes of reducing congestion. Similarly, the circled region on the right might be changed so that product to the far right of Zone C is moved to the left-most green area. Then Zone C would be more compact and could be visited by a unidirectional path, again to reduce congestion. Teams should feel free to evaluate these suggestions, or come up with your own.

— A Project for S. P. Richards Company (“Warehouse & Distribution Science,” n.d., heading “The project,” proposed layout changes).

Data integrity connects profile sources

Data integrity

Stable keys, time-valid conversions, and location addresses connect six warehouse data sources.

Item master. Use a stable SKU identifier, description, product family, introduction date, dimensions, weight, packaging, status, and location. The introduction date prevents a new SKU from being misclassified as inactive.

Order history. Use order, customer, SKU, quantity, unit, special handling, and operational timestamps. Financial order history may not record wave, pick-start, pack-complete, or ship-confirm events. Name any proxy honestly.

Inventory history. Daily or periodic on-hand stock supports inventory, variability, and capacity profiles. Monthly averages cannot establish a daily peak or stockout cause.

Layout and locations. Layout files estimate travel and reveal spatial concentration. Location addresses must use the same spelling and hierarchy in the WMS extract and map. Report duplicates, unmapped addresses, and pooled levels.

Storage units. Record address, handling-unit scale, dimensions, capacity, weight limit, and contained SKUs and quantities. These fields connect activity to feasible storage modes.

Source basis. The item-master, order-history, layout, and location-address requirements follow warehouse data guidance in (Bartholdi and Hackman 2019, 237–42). Inventory history and storage-unit data are explicit course extensions needed for the inventory and storage-system profiles.

Location addresses convert picks into spatial demand

Storage address

Zone B / Aisle 231 / Bay C / Level 3 / Position A

2D activity map

Use zone, aisle, and bay to analyze travel and congestion.

3D storage analysis

Add level and position to analyze ergonomics and vertical handling.

Addresses of storage locations within the warehouse. This might include zone, aisle, section, shelf and position on the shelf.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, p. 237).

The address hierarchy turns a pick record into spatial demand. Zone, aisle, and bay support a two-dimensional view of travel and congestion. Level and position add the vertical placement needed to study reach, ergonomics, and vertical handling. This 2D/3D interpretation is course synthesis; it is not source wording. The S. P. Richards reserve-location data use the analogous fields zone, aisle, bay, level, and position (“Warehouse & Distribution Science,” n.d., heading “Data,” reserve-location subsection).

Normalize location addresses before mapping

Location integrity

“B-231-C-3-A” and “b-231-c-3-a” may become different keys. Normalize format, then report duplicate and unmapped addresses.

Address hierarchy. A warehouse location commonly combines zone, aisle, bay or section, storage level, and position. The exact hierarchy must match the facility and the WMS.

Two-dimensional use. Zone, aisle, and bay can be mapped to the floor plan to show travel-intensive areas, unbalanced zones, and congestion risk. The map must preserve aisles, cross aisles, storage blocks, docks, and non-storage space. Aggregating levels should be disclosed.

Three-dimensional use. Adding level and position supports ergonomic zoning, vertical travel, load restrictions, and equipment access. A popular SKU at an awkward level may create more work than its 2D coordinate suggests.

Data control. Normalize capitalization and separators without changing the operational meaning. Check for duplicate addresses, storage locations missing from the map, mapped locations with no activity, and transactions with no valid address.

Source basis. Location addresses and their two- and three-dimensional uses follow (Bartholdi and Hackman 2019, 239–42).

Every profile ends in an operating test

1. Connect order lines, item master, inventory, and locations.
2. Validate units, dates, and keys.
3. Aggregate by the stated unit of analysis.

4. Plot the full distribution.
5. Ask whether the pattern changes a warehouse decision.
6. Define an improvement strategy, or revise the metric and stop.

Course synthesis based on Frazelle, Bartholdi and Hackman, and Baker and Canessa. Begin with the warehousing decision. Validate the fields required to support that decision. Then aggregate by the correct unit of analysis: customer, order, line, item, day, or location. A chart without an improvement strategy is descriptive output, not a completed profile. The six-step sequence on the slide is the course’s integration of the sources below; it is not a framework stated verbatim by any one source.

Warehouse activity profiling and data mining is the systematic analysis of item and order activity, seeking to identify and take advantage of patterns of opportunity. Our pattern recognition algorithms highlight the root cause of material and information flow problems, pinpoint major opportunities for process improvements, and provide an objective basis for project-team decision making.

I will start with some of the major motivations and potential roadblocks to successful profiling. Then I will review a full set of example profiles and their interpretations. The examples will serve to teach the principles of profiling and as an outline for the full set of profiles required for warehouse reengineering.

— World-Class Warehousing and Material Handling (Frazelle 2016, 21–22).

Order history. The order history is simply a concatenation of all the shopping lists submitted by all the customers during the preceding year. It contains the following information.

• Unique ID of this order, to distinguish it from the shopping lists of other customers and from the shopping list of the same customer on another day or later on the same day.
• Unique ID of SKU, which allows us to look up the SKU to see where it is stored.
• Customer.
• Special handling.
• Date/time order picked.
• Quantity shipped.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 238–40).

The analytical grain changes the question. Customer aggregation measures who generates demand; orders describe transaction size and mix; lines approximate distinct SKU requests and storage visits; items describe SKU activity; days describe calendar workload; and locations convert transactions into spatial demand. Choose one grain before aggregation and retain identifiers needed to move between grains (Bartholdi and Hackman 2019, 237–42).

Oxley (1994) provides a fairly comprehensive list of steps that incorporates the key features of the previous authors. He starts with defining the overall system requirements of the supply chain, including such factors as service levels and implementation time constraints. Again, data collection and analysis are key steps. He also introduces a new step of establishing the unit loads to be used. The following steps are again concerned with developing alternative operating methods, equipment types and layouts. He stresses that the warehouse design should be centred on the storage and handling requirements and that the building should then be designed around these.

— Warehouse design: A structured approach (Baker and Canessa 2009, 427–33).

Ten checks make a profile credible

1. Warehousing decision
2. Scope and study period
3. Unit of analysis
4. Calculation and units
5. Data preparation

6. Assumptions and limitations
7. Consistent comparison
8. Chart completeness
9. Key findings
10. Improvement strategy

Credibility rule

A credible profile addresses all ten checks. A missing denominator, period, or limitation can change the warehouse conclusion.

Course-defined formative checklist. The exact ten checks are an instructor-developed expansion of the public activity-profiling guidance. They are not a published research instrument. The public course page groups quality under context and scope, assumptions and method, quantitative precision, visualization quality, insight and next analysis, and writing and brevity (Beirigo 2026, secs. “Activity-profile options” and “Formative quality criteria”).

The checks mean:

1. Warehousing decision: name the operational choice the profile informs.
2. Scope and study period: identify population, facility boundary, and dates.
3. Unit of analysis: define what one observation represents.
4. Calculation and units: state numerator, denominator, formula, and unit.
5. Data preparation: document joins, filters, conversions, and exclusions.
6. Assumptions and limitations: separate observed fields, proxies, and gaps.
7. Consistent comparison: use compatible populations, periods, units, and scales.
8. Chart completeness: provide a decision-focused title, axes, units, legend, readable scale, and disclosed thresholds.
9. Key findings: quantify the dominant concentration, mixture, peak, or exception.
10. Improvement strategy: identify the next warehouse action and evidence test.

The course instruction asks students to “Present a summary of main assumptions and the key findings” (Beirigo 2026, sec. “Activity-profile options”).

These criteria reflect recurring weaknesses in prior activity-profile assignments: incomplete preparation, missing chart elements, incorrect calculations, unclear scope, unsupported interpretation, and vague improvement strategies. State the population, study period, unit of analysis, numerator, denominator, unit, exclusions, outlier rule, assumptions, limitations, and comparison basis. Finish with a quantitative finding and one supported warehouse strategy. Avoid generic openings such as “The graph shows.”

A warehouse is a complicated and busy place and it can be hard to get an accurate sense of what is happening. Warehouse activity profiling is the careful measurement and statistical analysis of warehouse activity. This is a necessary first step to almost any significant warehouse project: Understand the customer orders, which drive the system.

There are several simple statistics that are the first things to learn about a warehouse. Each gives some hint as to the economics of that warehouse; but these are to be treated carefully because many are simple averages and so can be misleading. Their primary advantage is to summarize the warehouse environment succinctly (but at the cost of hiding much complexity).

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 233–34).

Time permitting, make a sample request for a small piece of the data (say, one day’s worth) and check that it makes sense. Review the meaning of every data field with people from both operations and from IT. Show them summary statistics you extract from that day of data to see whether they seem reasonable. Check the text description of the biggest, busiest SKUs to see that there are no suspicious results.

Only after successfully sampling the data should you submit your request for the full data dump.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 246–47).

Four Frazelle families plus spatial activity

Table 5: Course implementation of Frazelle’s four profile families plus a spatial extension.

Warehouse activity	Profiles	Profile types	Primary decisions
Spatial activity	1	Pick frequency	Slotting and travel
Customer orders	16	Pareto, composition, volumetric, calendar-clock	Picking process and capacity
Purchase orders	4	Pareto, composition, volumetric, calendar-clock	Receiving process and capacity
Item activity	6	Pareto and demand correlation	Storage mode and location
Item inventory	3	Inventory Pareto and handling units	Inventory review and storage system
Analytical total	30

Profile taxonomy. Frazelle separates warehouse activity profiles into order profiles and item profiles. Order profiles cover customer orders and purchase orders through Pareto, composition, volumetric, and calendar-clock patterns. Item profiles cover item activity and item inventory through activity Pareto, demand correlation, demand variability, and inventory distributions.

Use in warehouse design. The taxonomy prevents unrelated decisions from being forced into one chart. Order profiles support process design, capacity, and workload planning. Item-activity profiles support slotting and storage-mode assignment. Inventory profiles support inventory review and storage-system selection.

Course implementation count. The 30 analytical profiles combine later Chapter 2 examples, named but unfigured measures, four modeled purchase-order profiles, and one spatial extension. The number 30 is not stated in Figure 2.6.

Implemented customer-order profiles. Customer Pareto; order-family mix; pallet and case handling mix; pallet, layer, and case increments; lines, units, cube, lines–cube, and weight per order; month, ordinal week, weekday, and hour.

Implemented inbound, item, and spatial profiles. Supplier Pareto; purchase- order composition, volumetric, and calendar-clock; item popularity; cube movement; popularity–cube; order completion; demand correlation; demand variability; inventory Pareto; item-family inventory; handling-unit inventory; and the aisle–bay pick-frequency matrix.

Spatial activity. The pick-frequency map is a complementary warehouse-activity view. It maps transaction frequency to the physical layout, but it is not one of Frazelle’s named profile families.

Transportation boundary. Frazelle names inbound and outbound transportation orders in the Chapter 2 prose, but Figure 2.6 does not detail those categories into profile types; they are outside this 30-profile taxonomy.

Frazelle. The family structure follows the warehouse activity profile taxonomy (Frazelle 2016, fig. 2.6, p. 29).

Decision confidence follows evidence quality

Table 6: Evidence basis and the strength of a warehouse conclusion.

Evidence basis	Appropriate use	Evidence required before implementation
Observed transactions	Diagnose the measured workload	Current constraints, cost, service, and stability
Substitute field	Screen a warehouse method	The intended WMS event or master-data field
Modeled scenario	Test calculation logic and sensitivity	Company observations from the proposed process

Investment rule

Capital decisions require current physical, service, capacity, labor, and cost evidence.

Course synthesis based on Baker and Canessa and Bartholdi and Hackman. The observed, substitute-field, and modeled-scenario rows are course evidence categories. The sources support the need to distinguish descriptive evidence from validated design evidence, but they do not state this three-row taxonomy.

Every profile states its warehouse decision, scope, unit of analysis, calculation, data preparation, assumptions, comparison basis, chart elements, quantitative finding, and improvement strategy. These 10 items make the analysis understandable and auditable.

Presentation completeness does not establish that a proposed operating change is feasible or economical. Observed profiles still require current facility constraints and stability tests. Proxy profiles require the missing operational event. Synthetic profiles explain the method and the shape of a useful result; they supply no company conclusion.

Rouwenhorst et al. (2000) propose a framework for making these high-level decisions. They determine two clusters of problem areas at what they describe as “strategic” level decisions: centred around system selection on the basis of technical and economic capabilities. The first cluster (i.e. based on technical capabilities) relates to this step. They identify decision areas and state that these are highly interrelated, but found no literature to assist with these decisions.

— Warehouse design: A structured approach (Baker and Canessa 2009, 427–33).

Naish and Baker (2004) describe a step-by-step approach to equipment evaluation, comprising:

• High-level technology assessment, based on such general factors as the scale of the operation and the flexibility required.
• Equipment attributes, to identify whether each equipment type is suitable for the application.
• Decision trees, which act as representations of “expert systems.”
• Full costing comparison, to calculate all the costs associated with the remaining options.
• Sensitivity analyses, to identify whether the preferred systems still perform well under alternative business scenarios.
• Computer simulation, to test the effectiveness of the preferred system under different conditions, such as crane breakdown.

— Warehouse design: A structured approach (Baker and Canessa 2009, 427–33).

For analyzing warehouse operations you have to be careful where you obtain this data. Often this data is from a sales transaction database, which tracks financial rather than warehouse events. Consequently, the date recorded may represent the date the order was placed or when it was printed, not when it was processed in the facility. Similarly, the order that appears in the sales transaction database may actually have been processed at another facility.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 239–40).

A substitute field is therefore useful only as a disclosed proxy: its meaning must be checked against the operational WMS event before implementation. A modeled scenario can test logic and sensitivity, but it does not establish a company’s physical feasibility, service performance, or economics without current observations (Baker and Canessa 2009, 427–33).

Warehouse work varies across space and time

Where

• Demand: SKU, family, and handling unit
• Space: zone, aisle, bay, level, and position

When

• Calendar: month, ordinal week, weekday, and hour
• Operation: wave, shift, cutoff, and peak period

Capacity rule

Combine spatial and calendar-clock profiles to size labor, equipment, storage, and docks.

Where and when is the work? How can we estimate the work in a warehouse? Work is generated by the customer orders; each customer order is a shopping list comprised of “pick lines”; and each pick line generates travel to the appropriate storage location and subsequent picking, checking, packing and shipping the product. Pick lines are then a strong indicator of work; and fortunately there is almost always a historical record of them because they correspond to entries in a sales invoice, which is one of the first pieces of information to be computerized.

We use this information to infer where the work is; that is, how it is distributed among

• Skus
• Product families
• Storage locations
• Zones of the warehouse
• Time (time of day, days of the week, weeks of the year, and so on)

Sometimes this is referred to as activity analysis because we examine the activity of each sku, in particular, how many times was it requested; and how much of the sku was sold? Notice that these are two different questions: The first asks “on how many customer orders did this sku appear?”; and the second asks “How many pieces, cases or pallets moved through the warehouse?”.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 240).

Spatial activity profiles

02

Hot aisle–bay cells identify slotting candidates

Figure 2: Observed pick lines by aisle–bay address with schematic rack spacing and one shared absolute scale.

Decision: Verify the travel map, then test hot addresses for re-slotting and zone balance.

Profile definition. A pick-frequency map assigns an activity count to each storage address. The unit of analysis is the aisle–bay cell; the activity measure is order-line picks.

Construction. The observed carton-detail snapshot contains 93,750 pick lines in Zones B and C from February 25 through March 25, 2005. Levels and positions are pooled to the aisle–bay address. Both panels use one six-band absolute scale. White spacing makes the address sequence read as rack sections and a cross aisle, while clearly remaining a schematic layout.

Observed pattern. Zone B carries more activity and has several concentrated aisle–bay cells. Zone C is more diffuse. The pattern identifies candidates for location review; it does not by itself establish the best assignment.

Pattern and interpretation. Concentrated hot cells indicate candidates for re-slotting closer to dispatch or into ergonomic pick positions. Adjacent hot cells may also indicate congestion risk, especially when several pickers share the same aisle.

Warehousing decision. Compare the current layout with an alternative slotting assignment that minimizes expected travel subject to storage compatibility, replenishment access, congestion, and location-capacity constraints.

Interpretive limit. The panels preserve address order but not verified aisle width, rack orientation, docks, or travel paths. The cross-aisle spacing is a visual aid, not company geometry. The profile pools levels and positions, and a pick line does not measure units, cube, travel time, or congestion. Join the addresses to the verified facility map before optimizing slotting.

Company basis. The snapshot is part of the S. P. Richards educational distribution-center data (Bartholdi and Hackman 2019).

Customer order profiles

03

29 customers generate 80% of picking activity

Figure 3: Only 29 customers generate 80% of all order-line activity.

Warehouse response

Test dedicated processing cells for the 29 dominant customers, subject to assortment stability and mixed-order consolidation.

Profile definition. A customer Pareto ranks customers by descending warehouse activity and plots their cumulative contribution. Here, activity is measured as order lines because each line approximates a pick-location visit.

Construction. The profile uses 2,249,630 order lines from 1,097 customers during 2004. Customers are ranked by line count. The cumulative line share is then compared with the cumulative customer share.

Observed pattern. The top 29 customers, or 2.6% of the customer population, generate 80% of all order-line activity. This concentration is stronger than a conventional 80/20 split.

Warehousing implication. The dominant customers are candidates for dedicated forward-pick zones, processing cells, or customer-specific service rules. The design may reduce travel and improve accountability while sharing reserve storage, receiving, and shipping resources.

Interpretive limits. A dedicated zone is justified only when customer volume, SKU assortment, service requirements, and order patterns remain stable. Compare travel savings with duplicated inventory, replenishment effort, space use, and the consolidation cost of mixed orders.

Frazelle. Customer concentration motivates the warehouse-within-a-warehouse concept (Frazelle 2016, figs. 2.7–2.8, pp. 29–31).

A processing cell needs stable concentration

Table 7: Example division of shared and dedicated resources in a warehouse-within-a-warehouse design.

Shared across the facility	Dedicated to the dominant customer or family
Receiving and reserve storage	Forward-pick locations
Building, utilities, and safety systems	Order release and picking rules
Main replenishment and shipping resources	Labor accountability and service measures
Common inventory where duplication is costly	Packing or value-added services when required

Use only when

Volume, assortment, and service requirements are stable enough to outweigh duplicated inventory, replenishment, space, and mixed-order consolidation.

Customer Paretos. Some customers place such high demands on a warehouse, represent such a large portion of the activity of the warehouse, and have such demanding customer-service requirements that it may make sense to establish a separate area within the warehouse for those customers. For example, one of our apparel manufacturing clients does so much business with JC Penney that they have a JC Penney warehouse within their warehouse. It is not uncommon for consumer products and food manufacturing companies to have areas within their warehousing dedicated to Wal-Mart activity.

The customer pareto in Figure 2.7 is from one of our food and beverage clients. Note that just three customers make up 50 percent of our client’s distribution center activity. Accordingly, we assigned each of those customers their own zones in the warehouse to reduce travel time and improve customer service.

In another example (Figure 2.8), in a large publisher’s distribution center, a central pool of reserve stock is used to support three distinct internal customers/business units—retail, trade, and periodicals. We designed the central, shared warehouse with allocated, random storage and the forward picking areas with dedicated, distinct forward picking zones. The manager of each forward picking zone has a dotted-line reporting relationship with the business unit to which his or her picking zone reports. The solid-line reporting relationship is to the director of distribution. This warehouse-within-a-warehouse design yielded the best of both worlds—shared receiving resources, efficient handling of central inventory, dedicated forward picking lines, shared shipping resources, and business unit accountability.

The warehouse-within-a-warehouse design philosophy works because small warehouses, in general, have higher productivity, response time, and accuracy performance than large warehouses. Many of the customer-order and item-activity profiles identify opportunities to subdivide a warehouse operation into self-contained warehouse processing cells, virtual warehouses, or warehouses within the warehouse. A warehouse within a warehouse is similar to a flexible manufacturing cell in a factory, where the productivity, quality, and cycle time are much better than in the factory as a whole.

— World-Class Warehousing and Material Handling (Frazelle 2016, 28–31).

62.3% of orders stay within one product family

Figure 4: 62.3% of orders contain one description-based product family.

Warehouse response

Test family-based zones and a shared merge point for mixed orders.

Validate the description-derived families against the company’s handling groups before changing locations.

Profile definition. A family-mix profile measures whether an order can be completed within one product family or requires several families. The pattern tests whether family-based warehouse zones can operate as self-contained order- completion cells.

Construction. The unit of analysis is the customer order. The item master lacks a managerial family field, so the first normalized product term in each item description supplies a reproducible family proxy. Examples include PEN, PAPER, FILE, CHAIR, and TONER. The calculation counts distinct proxy families on each of 883,910 orders in 2004.

Observed pattern. Of all orders, 62.3% contain SKUs from one description- based family. The remaining orders require movement or consolidation across several families. PAPER is the largest pure family at 3.0% of orders, while BINDER + INDEX is the most frequent exact two-family combination at 0.09%. The combination remains small because the proxy taxonomy is highly granular.

Warehousing implication. A high pure-order share supports a controlled test of family-based zoning. Evaluate whether inbound loads, reserve storage, forward picking, and labor can follow the same families while mixed orders use an efficient merge point.

Interpretive limits. Description terms can contain abbreviations, synonyms, and inconsistent granularity. Replace the proxy with the company’s managerial family and handling-compatibility fields before changing zones. Compare reduced travel with duplicated stock, workload imbalance, and mixed-order consolidation.

Frazelle. Family mix is the principal evidence for family-based processing cells (Frazelle 2016, fig. 2.9, pp. 32–34).

Mixed pallet demand limits process separation

Figure 5: Scenario: 52% partial-only, 29% full-only, and 19% mixed.

Warehouse response

Test separate pallet and case-picking areas with one controlled merge process for the 19% mixed-order segment.

Scenario basis: 4,000 orders with fixed pallet quantities.

Profile definition. The full- and partial-pallet mix classifies each order as partial-pallet only, full-pallet only, or mixed. It tests whether pallet picking and case picking should share an area or operate as separate completion zones.

Construction. A valid profile requires the shipped quantity and the SKU’s cartons-per-pallet quantity on every line. The source item master lacks pallet quantities, so the profile uses a fixed 4,000-order scenario based on Frazelle’s pattern.

Scenario result. Partial-pallet-only orders represent 52%, full-pallet-only orders 29%, and mixed orders 19%. Thus, 81% of orders can be completed within one handling process.

Warehousing implication. Separate pallet and case-picking areas are plausible. The WMS should classify orders at release and create pallet and case portions for the mixed group. Compare alternative merge points and synchronization rules.

Interpretive limits. The economic result depends on the mixed-order share, travel, staging, consolidation delay, equipment utilization, and inventory duplication. This synthetic distribution describes a design pattern, not the source company.

Frazelle. The same mix explains separate pallet and case completion zones (Frazelle 2016, fig. 2.10, pp. 34–35).

Mixed orders are 15.9% but carry 37.9% of lines

Figure 6: Mixed orders represent 15.9% of orders but 37.9% of their order lines.

Warehouse response

Size separate broken- and full-case processes and provide a merge path for the 15.9% mixed-order segment.

Profile definition. The full- and broken-case mix classifies customer orders by their required handling process: broken-case only, full-case only, or both. It tests whether each-picking and case-picking should use separate areas.

Construction. Each line compares shipped quantity with the SKU case pack. Valid pack data cover 2,133,143 lines and 857,579 orders, or 94.8% and 97.0% of the respective populations. Line classifications are then combined at order level.

Observed pattern. Broken-case-only orders represent 73.6%, full-case-only orders 10.5%, and mixed orders 15.9%. Mixed orders carry 37.9% of the valid order lines, so their merge workload is larger than their order share suggests.

Warehousing implication. Evaluate separate broken-case and full-case zones, with WMS routing and a defined merge process for mixed orders. Size each zone from its workload rather than its order count alone.

Interpretive limits. Missing or invalid case packs can change the mix. Test the decision using labor hours, travel, replenishment, packaging, staging, and consolidation cost. A low mixed share is necessary but not sufficient.

Frazelle. The profile supports separate full- and broken-case completion zones (Frazelle 2016, fig. 2.11, pp. 34–37).

Pallet demand peaks at quarter- and half-pallet loads

Figure 7: Frazelle’s reference pattern peaks at 25% and 50% of a pallet.

Warehouse response

Collect pallet quantities, confirm the recurring increments, and test preconfigured quarter- and half-pallet loads.

Profile definition. A pallet order-increment profile measures the fraction of a pallet requested on each order line. Peaks identify repeated sub-pallet quantities that may be handled as preconfigured unit loads.

Construction. The calculation divides the requested carton quantity by the SKU’s cartons-per-pallet quantity. The source item master lacks pallet quantities, so the plotted percentages reproduce the bins and values in Frazelle Figure 2.12.

Reference result. Order-line shares peak at 25% and 50% of a pallet. The 100% bin contains only 0.5%, so the profile supports preconfigured quarter- and half-pallet loads rather than a full-pallet demand conclusion.

Warehousing implication. Test supplier-built, manufacturing-built, or receiving-built unit loads. Pallet quantities should be visible to order entry, and commercial price breaks may be aligned with efficient handling increments.

Interpretive limits. Preconfigured loads can reduce storage density and may conflict with product rotation, stability, customer acceptance, or load-height constraints. Verify that the peaks persist across customers and periods.

Frazelle. Peaks and valleys in pallet fractions indicate opportunities for preconfigured quarter- and half-pallet loads (Frazelle 2016, fig. 2.12, pp. 36–39).

Layer demand concentrates at 1, 2, 5, and 10 layers

Figure 8: Reconstructed reference pattern: demand clusters at integer layer counts.

Warehouse response

Evaluate layer-picking equipment, layer-sized replenishment, and supplier pallet patterns against case-picking labor and damage.

Profile definition. A layer-picking profile measures requested quantity as a fraction or multiple of a complete pallet layer. It identifies demand that can be served by layer-handling methods instead of individual-case picks.

Construction. The calculation requires cases per layer for every SKU. The source item master lacks this field, so the slide reconstructs the integer layer pattern visible in Frazelle Figure 2.13. The displayed shares are an approximation of that printed distribution.

Reference result. Demand concentrates at one, two, five, and ten complete layers. The peaks indicate repeatable handling units rather than a continuous distribution of case quantities.

Warehousing implication. Evaluate layer-picking attachments, vacuum or clamp equipment, layer-sized replenishment, and supplier pallet patterns. The business case should compare throughput, labor, damage, and capital cost with case picking.

Interpretive limits. Check package strength, layer stability, SKU coverage, volume persistence, pallet patterns, and customer acceptance. This synthetic profile demonstrates the decision pattern only.

Frazelle. Integer-layer peaks support layer-picking equipment and higher throughput (Frazelle 2016, fig. 2.13, pp. 39–40).

Only 5.5% of eligible lines request full cases

Figure 9: Among 1,857,186 eligible lines, 5.5% are exact full-case multiples; 6.4% fall in the 97.5%–100% case band.

Warehouse response

Test inner-pack sizes and commercial order increments at the persistent partial-case peaks.

Profile definition. A case order-increment profile measures the requested quantity as a portion of a full case. Peaks reveal stable split-case quantities that may justify inner packs or commercial order increments.

Construction. Each line quantity is divided by the SKU case pack. Exact full-case multiples appear at 100%; partial quantities retain their remainder fraction. Lines with missing or single-unit packs are excluded.

Observed pattern. Exact full-case multiples represent 5.5% of the 1,857,186 eligible lines. The 97.5%–100% band contains 6.4%. The eligible population is 82.6% of all order lines; it excludes 166,166 positive-quantity lines with a single-unit case pack and lines with missing or invalid packs.

Warehousing implication. Candidate peaks can inform inner-pack size, supplier packaging, receiving preparation, and price breaks. The objective is to minimize piece handling after order release.

Interpretive limits. Do not select an inner pack from the histogram alone. Test packaging cost, minimum order quantities, demand stability, stock rotation, customer acceptance, and the share of lines covered by valid pack data.

Frazelle. Packaging and price increments should follow persistent partial- and full-case peaks (Frazelle 2016, fig. 2.14, pp. 39–42).

57.9% of orders visit one SKU location

Figure 10: Single-line orders represent 57.9% of 883,910 orders.

Warehouse response

Batch the single-line majority in location sequence; route the multi-line tail through cart or zone-wave picking.

Profile definition. The lines-per-order profile counts unique SKUs on each customer order. Each unique SKU normally requires a storage-location visit, so lines per order is a primary driver of picking travel and handling work.

Construction. Order lines are deduplicated by order and SKU, then counted for all 883,910 orders in 2004. Five operational bands preserve the single-line majority while keeping the multi-line tail visible for process segmentation.

Observed pattern. Single-line orders represent 57.9% of orders, and the median is one line. The distribution therefore resembles Frazelle’s single-line pattern rather than the large retail-order pattern.

Warehousing implication. Evaluate a dedicated singles process, batch picking in location sequence, dynamic forward picking, or cross-docking for back orders. Large multi-line orders may follow a separate cart, pallet, or zone-wave process.

Interpretive limits. A line is not identical to labor effort. Include quantity, cube, handling unit, travel, cut-off time, and downstream sortation before selecting a process. Confirm how the WMS creates split lines and substitutions.

Frazelle. The book contrasts single-line and large-order distributions and connects them to different picking strategies (Frazelle 2016, figs. 2.15–2.16, pp. 42–45).

Half of all orders ship 3 units or fewer

Figure 11: Positive-shipment orders: median 3 units; 90th percentile 31.

Warehouse response

Use the order-size segments for commercial analysis. Use cube, mass, and pack hierarchy for equipment and container choices.

Profile definition. The units-per-order profile measures the total selling units shipped on each customer order. It describes order scale but does not identify how many locations or handling units the picker must visit.

Construction. Shipped selling units are summed within each order. The operational population contains 842,434 orders with positive shipped quantity. The 41,476 nonpositive orders are service exceptions and are excluded from the physical workload curve. The logarithmic cumulative curve retains the full positive tail.

Observed pattern. The median order contains 3 units and the 90th percentile contains 31 units. The right-skewed distribution separates a large small-order population from a smaller high-quantity tail.

Warehousing implication. Use the profile with cube and weight to select tote, carton, cart, and accumulation capacity. It can also support order segmentation and capacity scenarios for packing and shipping.

Interpretive limits. Selling units are not normalized handling units. One unit may be an each, case, or bulky article. Do not compare counts across SKUs without pack hierarchy or physical dimensions.

Frazelle. Units per order is part of the customer-order volumetric suite and recommends joint interpretation with lines, cube, and weight (Frazelle 2016, 42).

Median positive order cube is 0.0062 m³

Figure 12: The median positive-cube order occupies 0.0062 m³ before packing allowances.

Warehouse response

Define cube classes for cartons, totes, carts, pallets, and accumulation space.

Profile definition. The cube-per-order profile measures the physical volume of outbound merchandise on each customer order. It supports container, picking- method, transport, staging, and accumulation decisions.

Construction. SKU dimensions are converted to cubic meters, multiplied by shipped units, and summed by order. The operational population contains 829,328 orders with complete dimensions and positive cube. The 40,859 complete-dimension orders with nonpositive cube are service exceptions outside the physical workload profile. The chart uses metric cube bands that reproduce the categorical structure of Frazelle’s profile.

Observed pattern. The median positive-cube order occupies 0.0062 m³ before packaging and void space. The long right tail represents orders that need different containers, equipment, or staging positions from the median order.

Warehousing implication. Define operational cube classes and test carton, tote, cart, pallet, conveyor, and transport assignments for each class. Capacity should reflect peak order release and accumulation.

Interpretive limits. Item dimensions may contain unit, orientation, or data- quality errors. Product cube is not packed cube. Add packaging, nesting, and load-building allowances before sizing equipment or transport.

Frazelle. Cube classes support selection of shipping containers, picking methods, transport modes, and staging space (Frazelle 2016, fig. 2.17, pp. 44–46).

49.1% of positive-cube orders are small singles

Figure 13: 49.1% of positive-cube orders contain one line and at most 0.03 m³.

Warehouse response

Route small singles to batch picking. Use carts or pallets for the large, multi-line tail.

Profile definition. The joint lines–cube profile positions each order by location visits and physical volume. Together, these variables provide stronger evidence for picking-method selection than either marginal distribution alone.

Construction. The plot uses 829,328 orders with complete dimensions and positive cube. Lines are grouped into 1, 2–5, 6–9, and 10+ bands. Metric cube bands correspond to the cubic-foot bands in Frazelle’s bubble matrix. Nonpositive-cube records are service exceptions outside the physical workload profile.

Observed pattern. Orders with one line and at most 0.03 m³ represent 49.1% of the positive-cube population. Other regions contain small multi-line, medium, and large orders with different travel and load requirements.

Warehousing implication. Small singles are candidates for batch pick-pack carts, totes, or compartmentalized containers. Medium orders may share multi- order carts. Large high-cube orders favor single-order cart or pallet tours.

Optimization use. Treat region boundaries as decision variables. Compare labor, travel, equipment, congestion, sortation, and consolidation cost while respecting container capacity and service-time constraints.

Interpretive limits. Lines and cube omit mass, handling unit, SKU location, congestion, service class, and replenishment interaction. Re-estimate the joint distribution for peak periods before fixing the segment boundaries.

Frazelle. The joint distribution supplies evidence for order-picking strategy and order-family segmentation (Frazelle 2016, figs. 2.18–2.19, pp. 44–49).

Verified SKU mass is required before setting lift limits

Figure 14: Positive-cube scenario: median 1.5 kg; 95th percentile 40.9 kg.

Warehouse response

Load verified SKU gross mass before setting manual-lift, cart, conveyor, or pallet thresholds.

Scenario basis: fixed product density; the source item table has no mass.

Profile definition. The weight-per-order profile measures outbound mass by customer order. It supports ergonomic limits and the selection of carts, conveyors, lifting devices, pallets, staging positions, and transport equipment.

Construction. The source item table has dimensions but no mass. For method demonstration, order cube for 829,328 positive-cube orders is multiplied by a fixed-seed synthetic density with a median of 250 kg/m³. Values above the 99th percentile remain visible in the logarithmic cumulative curve.

Scenario result. The proxy has a median of 1.5 kg and a 95th percentile of 40.9 kg. This would separate lightweight manual handling from a smaller group requiring lift assistance or pallet handling.

Warehousing implication. With measured mass, define handling classes and test equipment capacity, manual-lift policy, task assignment, and transport constraints for each class.

Interpretive limits. These weights are not company evidence. Product density varies strongly by SKU, and packaging adds mass. Replace the proxy with verified SKU gross mass before making safety or capital decisions.

Frazelle. Weight per order is part of the order-volumetric suite and recommends joint profiles with lines, units, and cube (Frazelle 2016, 42).

January is the 2004 high point

Figure 15: January averages 6,975 lines per recorded operating day.

Warehouse response

Use January as a 2004 capacity scenario. Confirm the pattern across years and after isolating calendar products before changing resources.

Profile definition. A month-of-year profile shows recurring seasonal peaks and valleys in warehouse activity. Separate series can be built for receipts, inventory, picks, shipments, and returns because their peaks may occur in different months.

Construction. The 2,249,630 order lines are first aggregated to 363 dates with recorded activity in 2004. Monthly values are the mean lines per recorded operating day. The timestamp represents order activity and does not directly measure receiving, picking, shipping, or returns.

Observed pattern. January is the maximum at 6,975 lines per recorded operating day. The office-products assortment includes calendars, which may create a temporary December–January fast-mover effect.

Warehousing implication. Use near-peak inventory to size storage and near- peak transaction activity to size labor, material-handling capacity, docks, and staging. Offset peaks can support seasonal labor transfer between processes.

Interpretive limits. One year cannot establish stable seasonality. Compare several years, working days, promotions, assortment changes, and exceptional seasonal SKUs. Use process timestamps before transferring the pattern to staffing.

Frazelle. The profile is a seasonal capacity-planning view and illustrates offset peaks across receiving, inventory, shipping, and returns (Frazelle 2016, figs. 2.20–2.21, pp. 48–52).

Days 8–14 form the monthly high point

Figure 16: Order-activity proxy: days 8–14 average 6,523 lines per recorded operating day.

Warehouse response

Test whether release capacity must change during days 8–14. Confirm the pattern with WMS task and shipment events before changing labor or carrier appointments.

Profile definition. A week-of-month profile measures repeated workload concentration within the monthly or quarterly planning cycle. It can reveal receiving and shipping peaks created by commercial and replenishment schedules.

Construction. Daily line totals are assigned to five explicit bins: days 1–7, 8–14, 15–21, 22–28, and 29–31. Each point is the average for dates with recorded activity in that bin. This makes the three-day fifth bin comparable with the seven-day bins.

Observed pattern. Days 8–14 have the highest rate at 6,523 lines per recorded operating day. The fifth bin averages 6,262 lines per recorded operating day after exposure normalization.

Warehousing implication. If the pattern repeats, set labor, replenishment, carrier appointments, staging, and maintenance windows with the within-month workload. Quarter-end effects may require separate capacity scenarios.

Interpretive limits. A recorded operating day is any date with at least one transaction. Add the formal facility calendar, holidays, closed days, and several years before treating the pattern as recurring.

Frazelle. Shipping may concentrate late in a month or quarter while receiving may concentrate early (Frazelle 2016, fig. 2.22, pp. 51–53).

Wednesday carries the weekly peak

Figure 17: Order-activity proxy: Wednesday averages 9,513 lines per recorded operating day.

Warehouse response

Test weekday release capacity first. Confirm the pattern with pick, pack, replenishment, and shipment events before changing shifts or equipment.

Profile definition. A day-of-week profile measures recurring weekday workload patterns. Customer delivery schedules, order cut-offs, production calendars, and carrier arrangements can shift warehouse work toward particular days.

Construction. The 2,249,630 order lines are first aggregated by transaction date. Each weekday reports the mean across its dates with recorded activity.

Observed pattern. Wednesday is the maximum at 9,513 lines per recorded operating day. Tuesday, Thursday, and Friday also carry substantial release work; weekend activity is minimal.

Warehousing implication. Stagger shifts, replenishment, equipment availability, and support labor to the repeated pattern. Customer or carrier negotiations may also smooth avoidable peaks when service agreements permit.

Interpretive limits. The timestamp records order activity rather than a physical shipment. Validate the delay from order receipt to wave release, picking, packing, and dispatch. Control for holidays and missing weekend work.

Frazelle. Customer purchase and delivery cycles can create predictable warehouse peaks (Frazelle 2016, figs. 2.23–2.24, pp. 51–55).

The 16:00 release peak contains 21.7% of lines

Figure 18: 16:00: 488,722 lines, or 21.7% of 2004 activity.

Warehouse response

Protect wave-release, picking, packing, and staging capacity near the 16:00 cutoff.

Order timestamps indicate release demand; WMS task scans determine labor.

Profile definition. An hour-of-day profile measures receiving, putaway, picking, or shipping activity by clock hour. It supports intraday capacity and shift decisions because warehouse processes can peak at different times.

Construction. The profile aggregates 2,249,630 order lines by the hour of the 2004 transaction timestamp. The field is treated as an order-release proxy; it is not a direct labor, equipment, or dock scan.

Observed pattern. Activity peaks at 16:00 with 488,722 lines, or 21.7% of annual activity. The concentration is consistent with the documented late- afternoon order cut-off.

Warehousing implication. Protect wave-release, picking, packing, and staging capacity near the cut-off. When receiving and outbound peaks do not overlap, stagger shifts or transfer cross-trained labor between departments.

Interpretive limits. Final staffing must use WMS task start and completion times, separate processes, and account for the lag between order receipt and physical work. A transaction peak is not automatically a labor-hour peak.

Frazelle. Hourly process distributions support system sizing for peak periods and identify labor-sharing opportunities from offset peaks (Frazelle 2016, fig. 2.25, pp. 56–57).

Purchase order profiles

04

55 suppliers generate 80.2% of modeled PO lines

Figure 19: Scenario: 55 of 325 suppliers generate 80.2% of purchase-order lines.

Warehouse response

Collect ASN, receipt, discrepancy, and putaway data for the 55 dominant suppliers first.

Scenario basis: weekly supplier demand; no observed receipts.

Profile definition. A supplier Pareto ranks suppliers by inbound warehouse activity and plots their cumulative contribution. Activity is modeled purchase-order lines because each line represents SKU-level receiving, checking, system-entry, and putaway complexity.

Construction. The source contains supplier-linked outbound demand but no purchase orders or receipts. Positive 2004 shipped demand is grouped by supplier, SKU, and Monday-start week. Demand is rounded upward to complete cases and all SKU lines for one supplier-week are consolidated into one synthetic purchase order. The scenario contains 572,763 PO lines, 14,254 POs, and 325 suppliers.

Modeled pattern. The top 55 suppliers, or 16.9% of the supplier population, generate 80.2% of PO-line complexity. The concentration is a result of the modeled weekly scenario, not an assumed 80/20 rule.

Warehousing implication. Use the modeled concentration to select suppliers for an observed receipt-performance study. With actual PO, ASN, appointment, receipt, discrepancy, and putaway data, compare advance shipment notices, standard labels, appointments, and exception controls.

Interpretive limits. PO lines measure complexity, not pallets, cube, spend, reliability, or dock hours. The scenario omits actual lot sizes, minimum order quantities, lead times, carrier consolidation, shortages, returns, and stockout censoring. Replace it with observed PO and receipt lines when available.

Frazelle. Supplier Paretos are part of purchase-order profiles. The book develops the customer-order branch in detail (Frazelle 2016, fig. 2.6, p. 29).

8.8% of receipts carry 46.3% of modeled cases

Figure 20: Scenario: mixed receipts are 8.8% of POs and carry 46.3% of cases.

Warehouse response

Use case receiving as the standard path and test a high-throughput mixed lane for the volume-heavy minority.

Scenario basis: weekly supplier demand and modeled pallet capacity.

Profile definition. A purchase-order composition profile distinguishes case-only, full-pallet-only, and mixed pallet–case receipts. Reporting both PO share and received-case share separates arrival frequency from physical handling volume.

Construction. The common weekly purchase-order scenario is restricted to 14,115 POs with valid case packs and case dimensions. Weekly demand is rounded to cases. Cases per pallet are estimated from a 1.219 m by 1.016 m pallet base, a 1.524 m usable load height, both horizontal case orientations, and a 200-case cap. Each PO is assigned to one mutually exclusive handling class.

Modeled pattern. Case-only receipts represent 91.1% of POs and 53.7% of cases. Mixed receipts represent only 8.8% of POs but 46.3% of cases. Full-pallet-only receipts account for 0.1% of POs in the stated scenario.

Warehousing implication. Design the default process for case-level verification and putaway. Provide a high-throughput mixed lane that directs intact pallets to reserve storage and residual cases to decanting or forward- pick replenishment. PO counts alone would understate this mixed-receipt burden.

Interpretive limits. Geometric pallet capacity ignores mass, stackability, overhang, crush limits, hazardous-material separation, and supplier pallet agreements. Weekly consolidation and case rounding are modeling choices. Use actual ASN, receipt, and pallet-master data before changing the dock layout.

Frazelle. Purchase-order composition appears in Figure 2.6. The profile transfers the handling-mix logic that the chapter develops for customer orders to the receiving process [Frazelle (2016), fig. 2.6, p. 29; pp. 34–36].

6.5% of modeled POs exceed 10 m³

Figure 21: Scenario: median PO cube is 0.37 m³; 6.5% of POs carry at least 10 m³.

Warehouse response

Route compact receipts through standard verification and reserve staging for the large-cube tail.

Scenario basis: weekly supplier demand; 2.6% of POs contain estimated cube.

Profile definition. A purchase-order volumetric profile measures the physical size distribution of inbound orders. Merchandise cube supports dock, staging, handling-equipment, and putaway-capacity decisions more directly than PO count alone.

Construction. The unit of analysis is one of 14,254 modeled supplier-week POs. For each line, case-rounded units are multiplied by selling-unit cube and then summed by PO. Missing cube is replaced first by the supplier median and, if necessary, by the global median of 0.001345 m³ per unit. Imputation affects 779 of 572,763 lines and 373 POs, or 2.6% of the PO population.

Modeled pattern. Median PO cube is 0.37 m³, while the 95th percentile is 14.42 m³. In total, 68.8% of POs are below 1 m³ and 6.5% contain at least 10 m³. The distribution has a long tail that the mean would conceal.

Warehousing implication. Route small receipts to compact verification and putaway lanes. Reserve staging positions, pallet-handling equipment, and dock windows for the large-cube tail. Validate whether cube correlates with unload time before translating this distribution into labor or dock-door requirements.

Interpretive limits. Merchandise cube excludes pallets, packaging, void, orientation, mass limits, and trailer utilization. The purchase orders and receipt quantities are modeled from outbound demand. The imputed cube values support completeness but reduce precision for affected POs.

Frazelle. Purchase-order volumetric profiles appear in Figure 2.6. The receiving profile uses the cube-per-order logic developed for customer orders [Frazelle (2016), fig. 2.6, p. 29; pp. 44–46].

Balanced appointments keep weekday shares near 20%

Figure 22: Leveled scenario: weekday and appointment shares remain near 20%; daily P95 is 7,083 cases.

Capacity evidence: Appointment, dock-event, receipt, and putaway records.

Profile definition. A purchase-order calendar-clock profile measures inbound work across month, week of month, weekday, and hour. These clocks support labor, dock-door, staging, and putaway-capacity decisions.

Construction. Each supplier-week PO enters a load-leveled appointment scenario with 25 slots: Monday–Friday at 07:00, 09:00, 11:00, 13:00, or 15:00. Within each week, the largest modeled receipts are assigned first to the least-loaded slot; the starting slot rotates by calendar week. The 2004 profile contains 14,144 appointments and 1,331,692 modeled cases. Every panel retains zero-load operating-day opportunities.

Scenario pattern. Average daily receipt demand is 5,083 cases. The 95th percentile is 7,083 cases and the maximum is 8,499. Each weekday and appointment slot carries approximately 20% of annual modeled cases. Month and ordinal-week variation remain after opportunity normalization.

Warehousing implication. The scenario demonstrates a balanced baseline and the four calendar views. With observed appointment data, compare the current schedule with a leveled alternative subject to supplier, carrier, product, and dock constraints. Convert cases into dock doors only with measured cases per dock-hour, productive hours, and target utilization.

Interpretive limits. The appointments are a balanced scenario, not observed supplier behavior. The profile cannot estimate lateness, queueing, unload duration, appointment adherence, public-holiday closures, or shift changes. Replace it with PO, ASN, appointment, receipt, and dock-event history.

Frazelle. Purchase-order calendar-clock profiles appear in Figure 2.6. The chapter develops four calendar levels for warehouse activity in the customer-order context [Frazelle (2016), fig. 2.6, p. 29; pp. 48–57].

Item activity profiles

05

25.3% of SKUs generate 80% of picks

Figure 23: The top 7,955 SKUs generate 80% of all order-line picks.

Warehouse response

Give the 7,955 A items productive storage modes and accessible positions, then size their faces with cube, inventory, and replenishment data.

Profile definition. The item popularity profile ranks active SKUs by descending order-line frequency and plots cumulative pick share against cumulative SKU share. A pick represents one order-line occurrence, so the profile estimates location visits rather than unit or cube movement.

Construction. The population contains 31,423 active SKUs and 2,249,630 order lines from 2004. Each SKU’s line count is divided by total lines before the ranked cumulative curve is calculated.

Observed pattern. The top 7,955 SKUs, or 25.3%, generate 80% of picks. The curve is less concentrated than a strict 80/20 pattern, so the fast-moving class is relatively broad.

Warehousing implication. Use breakpoints to define candidate A, B, and C popularity families. Give high-frequency SKUs productive storage modes and accessible positions, including the golden zone near travel aisles and around waist height.

Interpretive limits. Popularity omits cube, on-hand stock, replenishment, compatibility, congestion, and route interactions. Treat ABC thresholds as screening rules, then confirm assignments with a slotting model that includes space, travel, replenishment, and service constraints.

Frazelle. The ABC curve connects its breakpoints to storage modes and golden, silver, and bronze positions (Frazelle 2016, fig. 2.26, pp. 56–59).

SKU cube movement spans five orders of magnitude

Figure 24: 24,283 positive-cube SKUs cover 97.3% of order lines; median movement is 0.154 m³.

Warehouse response

Screen low-cube movers for drawers or shelving and high-cube movers for larger, replenishment-friendly faces.

Profile definition. The cube-movement profile measures annual shipped volume for each SKU. It supports storage-mode assignment and space allocation because volumetric flow determines pick-face size and replenishment intensity.

Construction. For each SKU with valid dimensions, shipped selling units are multiplied by selling-unit volume and summed across 2004. The chart counts SKUs in logarithmic movement ranges because throughput spans several orders of magnitude.

Observed pattern. Median positive annual movement is 0.154 m³ per SKU and the 95th percentile is 3.324 m³. Most SKUs move little cube, while a small tail generates much higher volumetric throughput.

Coverage. The chart retains 24,283 SKUs, or 77.3% of active SKUs. They account for 97.3% of order lines. Missing dimensions remain important because unmeasured bulky items could change the upper tail.

Warehousing implication. Low-cube movers may fit drawers or bin shelving. High-cube movers need larger faces and replenishment-friendly modes such as pallet flow, push-back rack, double-deep rack, or block stacking.

Interpretive limits. Annual movement is not peak inventory or required face capacity. Correct missing dimensions and add packaging, palletization, case packs, inventory variability, and replenishment intervals before sizing.

Frazelle. Movement ranges should correspond to feasible storage-mode alternatives (Frazelle 2016, fig. 2.27, pp. 59–60).

8.2% of valid SKUs are both fast and bulky

Figure 25: 8.2% of valid SKUs exceed both 80th-percentile activity guides.

Warehouse response

Separate small fast movers from bulky high-throughput items before assigning storage modes and face sizes.

Profile definition. This joint profile combines item popularity with cube movement. Each region represents one activity combination and a candidate storage-mode family.

Construction. The internal boundaries are the separate 80th percentiles of pick frequency and cube movement for valid 2004 SKUs. The decision matrix reproduces Frazelle’s popularity–cube storage-mode map while adding the share of SKUs in each region.

Observed pattern. Of valid SKUs, 8.2% exceed both guides. These items combine frequent location visits with high throughput and are priority candidates for high-productivity storage with carefully sized pick faces.

Warehousing implication. High-pick, low-cube items favor ergonomic forward picking. High-cube movers need replenishment-friendly modes and larger faces. Low-pick, low-cube items rarely justify expensive equipment. Within each mode, place SKUs with more picks per unit of space in the most accessible positions.

Optimization use. Evaluate candidate regions using wage rate, space and capital cost, replenishment, compatibility, congestion, travel, and the planning horizon.

Interpretive limits. Percentile guides are internal screening rules, not optimal boundaries or external benchmarks. The profile also omits inventory variability, replenishment batches, and route interactions.

Frazelle. The joint distribution supports storage-mode assignment, face sizes, and golden, silver, and bronze locations (Frazelle 2016, fig. 2.28, pp. 59–62).

Popularity and cube require different storage responses

Table 8: Storage and picking-method screening from the popularity–cube profile.

Popularity	Cube movement	Candidate warehouse response
High	Low	Ergonomic forward pick, small face, frequent replenishment
High	High	High-throughput flow storage with carefully sized reserve
Low	Low	Dense shelving, drawers, or shared low-cost storage
Low	High	Large face or reserve-oriented storage with selective access

Warehouse management is all about careful use of space and time (that is, labor or person-hours). Both space and time are expensive and so one would like to use as little of each as possible in delivering product to customers. Figure 2.3 shows a plot of the popularity (number of times requested, or picks) of each SKU of a warehouse together with the physical volume (flow) of the SKU moved through the warehouse during one month. There is little correlation between popularity and flow, and this is one of the challenges of warehouse management, because it is hard to design processes that work well with SKUs that may be any combination of popular/unpopular and low-volume/high-volume.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 2.3, pp. 13–15).

Popularity–Cube-Movement Profile. Done properly, slotting takes into account both the item popularity and cube-movement. An example popularity–cube-movement profile for broken-case picking is presented in Figure 2.28.

In this example, those items exceeding a certain cube-movement threshold are assigned to a carton flow rack. Items with high cube-movement need to be restocked frequently and need a larger storage location compared to items with medium and low cube movement. Hence they need to be assigned to a storage mode that facilitates restocking and condenses large storage locations along the pick line–carton flow rack. Items with low cube movement and high popularity generate many picks per unit of space that they occupy and do not occupy much space along the pick line. They need to be in a highly productive picking mode. In this case, light directed carousels are recommended because the picking productivity is high, and we can afford the carousels for items that do not need large storage housings on the pick line and do not need to be restocked frequently. (Carousels do not lend themselves to restocking and are expensive per cubic foot of space.)

Items with low popularity and low cube movement cannot be justifiably housed in an expensive storage mode. Hence they are candidates for bin shelving and modular storage drawers.

Items in the bottom right-hand quadrant of each preference region generate the most picking activity per unit of space they occupy in the storage mode. Hence they should be assigned to positions in the golden zone (most accessible pick locations) of their storage mode. Items in the upper right-hand and lower left-hand quadrants generate a moderate number of picks per unit of space they occupy in the storage mode. Hence they should be assigned to positions in the silver zone. Finally, items in the upper left-hand quadrant of the distribution generate the fewest picks per unit of space they occupy and they should be assigned positions in the bronze (least accessible) zone.

This example is not an end-all recommendation for slotting broken-case picking systems. This depends on many other factors, including the wage rate, the cost of space, the cost of capital, the planning horizon, and so on. Instead, this example is presented to illustrate how the popularity–cube-movement distribution is used in slotting.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Popularity–Cube-Movement Profile,” pp. 59–62).

The four candidate responses on the slide are a multi-source course synthesis. Popularity indicates likely pick frequency; cube movement indicates physical flow. Together they screen face size, storage mode, replenishment intensity, and whether scarce ergonomic positions should be protected for frequently picked SKUs.

Match storage response to activity

Storage rule

Popularity and cube movement screen storage modes and accessible positions. Replenishment interval, inventory, compatibility, and capacity determine the face size.

Popularity–Cube-Movement Profile. Done properly, slotting takes into account both the item popularity and cube-movement. An example popularity–cube-movement profile for broken-case picking is presented in Figure 2.28.

In this example, those items exceeding a certain cube-movement threshold are assigned to a carton flow rack. Items with high cube-movement need to be restocked frequently and need a larger storage location compared to items with medium and low cube movement. Hence they need to be assigned to a storage mode that facilitates restocking and condenses large storage locations along the pick line—carton flow rack.

Items with low cube movement and high popularity generate many picks per unit of space that they occupy and do not occupy much space along the pick line. They need to be in a highly productive picking mode. In this case, light directed carousels are recommended because the picking productivity is high, and we can afford the carousels for items that do not need large storage housings on the pick line and do not need to be restocked frequently. (Carousels do not lend themselves to restocking and are expensive per cubic foot of space.)

Items with low popularity and low cube movement cannot be justifiably housed in an expensive storage mode. Hence they are candidates for bin shelving and modular storage drawers.

Items in the bottom right-hand quadrant of each preference region generate the most picking activity per unit of space they occupy in the storage mode. Hence they should be assigned to positions in the golden zone (most accessible pick locations) of their storage mode. Items in the upper right-hand and lower left-hand quadrants generate a moderate number of picks per unit of space they occupy in the storage mode. Hence they should be assigned to positions in the silver zone. Finally, items in the upper left-hand quadrant of the distribution generate the fewest picks per unit of space they occupy and they should be assigned positions in the bronze (least accessible) zone.

This example is not an end-all recommendation for slotting broken-case picking systems. This depends on many other factors, including the wage rate, the cost of space, the cost of capital, the planning horizon, and so on. Instead, this example is presented to illustrate how the popularity–cube-movement distribution is used in slotting.

— World-Class Warehousing and Material Handling (Frazelle 2016, 59–62).

The top 10% of SKUs complete 37.9% of orders

Figure 26: The top 10% of SKUs can complete 37.9% of all orders.

Warehouse response

Test a compact order-completion zone for the top 10% of SKUs, with explicit replenishment and mixed-order merge rules.

Profile definition. The item-order-completion profile asks whether a compact SKU subset can fill a large share of complete orders. It differs from popularity because an order counts only when every required SKU is available in the subset.

Construction. SKUs are ranked by order-line popularity. For each order, the completion rank equals its least-popular required SKU. The curve reports the cumulative order share completed as additional ranked SKUs enter the candidate zone.

Observed pattern. Among 883,910 orders and 31,423 active SKUs, the top 10% of SKUs can complete 37.9% of orders. A steep initial curve indicates meaningful coverage from a compact assortment.

Warehousing implication. Evaluate a warehouse-within-a-warehouse or compact order-completion zone with short travel, dedicated labor, and simplified control. Search for a driving family such as brand, supplier, product group, size, color, or kit.

Interpretive limits. Compare processing and travel savings with duplicated inventory, replenishment, stockout, consolidation, congestion, workload balance, and space cost. Test whether completion coverage is stable by season.

Frazelle. The cumulative construction supports completion zones organized around compact SKU and product families (Frazelle 2016, figs. 2.29–2.31, pp. 62–66).

Most frequent SKU pair has 6.3x lift

Figure 27: Most frequent pair: 2,781 observed orders versus 442 expected, a 6.3x lift.

Decision: Test stable high-lift pairs only when travel savings exceed congestion and replenishment cost.

Profile definition. The demand-correlation profile ranks SKU pairs by the number of orders containing both items. Each unordered pair is counted at most once per eligible order.

Construction. The calculation screens the 120 most popular SKUs and retains 52,324 orders containing 2–12 distinct screened SKUs. Orders above that limit are excluded to control combinatorial dominance. Expected co-occurrence equals \(N_A N_B / N\) within the same eligible population. The chart ranks the 10 most frequent pairs by observed support.

Observed pattern. The most frequent pair appears in 2,781 orders versus 442 expected under independence, a 6.3x lift. The pair still covers only 0.31% of all 883,910 orders, so a relocation needs a measured travel benefit.

Warehousing implication. Stable affinity can support family zoning and short picking tours. Co-locate or sequence related SKUs only when the tour reduction exceeds congestion, replenishment, relocation, and space costs.

Interpretive limits. Lift can be unstable for rare pairs. Require minimum support, validate each association across periods, and disclose the SKU and order screens before changing locations.

Frazelle. Pair frequency informs an apparel slotting design that zones by size, groups styles, mixes colors, and protects popular golden-zone positions (Frazelle 2016, figs. 2.32–2.33, pp. 64–68).

Product structure and affinity shape warehouse zones

1. Define handling-compatible product families.
2. Measure SKU-pair co-occurrence.
3. Compare observed pairs with an independence baseline.
4. Protect popular golden-zone positions.
5. Test tour distance, congestion, and workload balance.

The overall operating strategy of a warehouse should be influenced by the extent to which orders require items from multiple families. If the majority of the orders are pure, drawing from a single product family, then zoning the warehouse on that basis could create virtual warehouses within the warehouse.

For example, the family-mix profile in Figure 2.9 was developed for a wholesale distributor of fine papers, copy/laser paper, and envelopes. Printers make high-quality brochures from flat stocks of fine papers. A carton of flat stock is 30 inches long, 24 inches wide, and 9 inches deep. A carton weighs 80 pounds. Cut stock is basic 8½- by 11-inch copier and laser printer paper. A carton of cut stock is 24 inches long, 10 inches wide, and 10 inches deep. A carton weighs 20 pounds. The third product family is envelopes—extremely small and lightweight merchandise.

In this example (Figure 2.9), we are evaluating the merits of zoning based on those three product families—flat stock, cut stock, and envelopes. If the orders are mixed—that is, flat stock, cut stock, and envelopes tend to appear together on customer orders—then in pallet building, we would start with flat stock, put cut stock on top of that, and put envelopes on top of that. If we zone the warehouse that way, we may pay a big travel-time penalty passing pallets between those zones. However, if the orders are pure—that is, they tend to be “complete-able” out of a single product family—then zoning the warehouse by product family will establish efficient warehouse processing cells, especially since inbound loads already arrive in those same product families.

— World-Class Warehousing and Material Handling (Frazelle 2016, sec. 2.3, pp. 32–34).

Product family, which may have implications for storage and/or handling. These tend to be particular to an industry and so require knowledge of the context. For example, product families for a drug store chain might include hair care products, dental products, shaving products and so on, which are displayed together at the retail store. For a grocery distributor product families might include dry goods, dairy, produce, refrigerated, frozen and so on. For a candy distributor product families might include chocolate (sensitive to heat), mint-flavored candies (odoriferous), and marshmallow (light and tends to absorb the smells of its neighbors), and so on. For an apparel distributor product families might include garment type, mill, style, color, or size. Note that a SKU might be in more than one product family.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, sec. 14.2, p. 237).

Assignment of location by correlation. The products are related in a characteristic, it can be weight, type, or they are found in different client orders multiple times, etc. These correlated products are placed in continuous spaces or in the same corridor to reduce OP distances. The data that creates the correlation must be known, and [34] and [18] illustrate some examples.

— The Storage Location Assignment Problem: A State-of-the-Art Review (Medrano-Zarazúa et al. 2024, item 3, pp. 31–32).

Just as a minority of the items in a warehouse makes up a majority of the picking activity, certain items in the warehouse tend to be requested together. In the example (Figure 2.32), pairs of items are ranked based on their frequency of appearing together on orders. In this case, we are examining data from a mail-order apparel company. The first three digits represent the style of the item (i.e., crew-neck sweater, V-neck sweater, turtle-neck shirt, pleated pants, etc.), the middle digit represents the size of the item (i.e., 1 = small, 2 = medium, 3 = large, and 4 = extra large), and the last digit represents the color (i.e., 1 = white, 2 = black, 3 = red, 4 = blue, 5 = green, etc.).

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

What do you think people tend to order together from this mail-order company? (I thought it would be shirts and pants that looked good together in the catalog.) What does the profile in Figure 2.32 suggest? In this case, customers tend to order items of the same style and size together. The explanation is that customers tend to get comfortable with a certain style and tend to order in multiple colors to add variety to their wardrobes. Of course, they order the same size unless they will return one for fitting. This pattern was a surprise to me. It was also a surprise to the marketing people. Surfacing truth is the crucial benefits of profiling process. (The myriad SKUs, order patterns, suppliers, and interdependent decisions make it difficult to form a reliable intuition about logistics operations.)

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

How do we take advantage of this demand-correlation information in slotting a warehouse? We zone the warehouse by item size first, creating a zone for the smalls, mediums, larges, and extra larges of all styles (Figure 2.33). Within each size area, we store items of the same style together, mixing colors within a style. This zoning strategy allows us to create picking tours based on size and style. As a result, order pickers pick many items on short-distance picking tours. Simultaneously, we manage congestion by spreading out the sizes. Golden zoning is used to slot the most popular color for each style at or near waist level.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

The category of operational efficiency includes variables focused on improving the use of resources within the storage system; for this reason, the use of equipment, energy consumption, the association index, the number of orders processed, the number of replacements, and the maximum load are considered. On the other hand, the space and distance category include variables related to the use of the area and storage capacity, e.g., locations and travel, as well as space use and availability, the number of locations, and the travel distance.

Within the time category, measurements of chronological duration for certain operations are summed; for example, travel time, picking time, and relocation time. The human factors category includes variables focused on workers and related activities; for this reason, they include workload, effort, discomfort, and energy used by the workers. The infrastructure category includes variables focused on resistance issues and geological conditions in the operations center, including structural stability. Finally, the cost category is related to monetary measures, including costs of storage, replenishment, and recovery. Table 10 presents the performance measures according to their category and optimization approach, either maximized (Max) or minimized (Min), and the authors who used them.

— The Storage Location Assignment Problem: A Literature Review (Reyes et al. 2019, sec. 3.3, pp. 214–215).

To test whether a given cell is dependent, we must determine if the actual count in cell \(r\) differs sufficiently from the expectation.

In the chi-squared test, expectation is calculated under the assumption of independence. Thus, \(E[i_j] = O_n(i_j)\) for a single item, \(E[\bar{i_j}] = n - O_n(i_j)\), and \(E[r] = n \times E[r_1]/n \times \cdots \times E[r_k]/n\). Then the chi-squared statistic is defined as follows:

\[ \chi^2 = \sum_{r \in R} \frac{(O(r)-E[r])^2}{E[r]}. \]

In short, this is a normalized deviation from expectation.

— Beyond Market Baskets: Generalizing Association Rules to Correlations (Brin et al. 1997, sec. 3, p. 267).

The five numbered steps are instructor-organized. The sources support compatible families, measured co-occurrence, an independence comparison, golden-zone access, and performance checks; they do not present this five-step order as a published method.

Test affinity before changing zones

Apparel example

Zone by size, group styles, mix colors, place the most popular color near waist height, and spread popular sizes to control congestion.

Demand-Correlation Profile. Just as a minority of the items in a warehouse makes up a majority of the picking activity, certain items in the warehouse tend to be requested together. In the example (Figure 2.32), pairs of items are ranked based on their frequency of appearing together on orders. In this case, we are examining data from a mail-order apparel company. The first three digits represent the style of the item (i.e., crew-neck sweater, V-neck sweater, turtle-neck shirt, pleated pants, etc.), the middle digit represents the size of the item (i.e., 1 = small, 2 = medium, 3 = large, and 4 = extra large), and the last digit represents the color (i.e., 1 = white, 2 = black, 3 = red, 4 = blue, 5 = green, etc.).

What do you think people tend to order together from this mail-order company? (I thought it would be shirts and pants that looked good together in the catalog.) What does the profile in Figure 2.32 suggest? In this case, customers tend to order items of the same style and size together. The explanation is that customers tend to get comfortable with a certain style and tend to order in multiple colors to add variety to their wardrobes. Of course, they order the same size unless they will return one for fitting. This pattern was a surprise to me. It was also a surprise to the marketing people. Surfacing truth is the crucial benefits of profiling process. (The myriad SKUs, order patterns, suppliers, and interdependent decisions make it difficult to form a reliable intuition about logistics operations.)

How do we take advantage of this demand-correlation information in slotting a warehouse? We zone the warehouse by item size first, creating a zone for the smalls, mediums, larges, and extra larges of all styles (Figure 2.33). Within each size area, we store items of the same style together, mixing colors within a style. This zoning strategy allows us to create picking tours based on size and style. As a result, order pickers pick many items on short-distance picking tours. Simultaneously, we manage congestion by spreading out the sizes. Golden zoning is used to slot the most popular color for each style at or near waist level.

— World-Class Warehousing and Material Handling (Frazelle 2016, 64–68).

Daily variation exceeds the mean for 65.4% of SKUs

Figure 28: For 65.4% of 500 SKUs, daily standard deviation exceeds the mean.

Warehouse response

Set pick-face allowances from the replenishment interval and service target, then test the 500 active SKUs across seasonal periods.

Profile definition. The demand-variability profile compares mean daily demand with its standard deviation for each SKU. It supports pick-face sizing when intrashift replenishment should be limited.

Construction. The 500 most active SKUs are aggregated by calendar day across 2004. Zero-demand days are inserted before calculating each SKU’s mean and sample standard deviation. The daily trace shows a representative variable SKU, and the companion bars summarize the full screened population.

Observed pattern. Of the screened SKUs, 65.4% have a standard deviation above their mean and therefore a coefficient of variation above one. The daily trace shows why average movement alone can undersize a pick face on peak days.

Warehousing implication. Set the face allowance from the replenishment interval, desired service level, forward-space cost, and reserve-to-forward capacity. The objective balances face-space use against expected replenishment work and shortage risk.

Interpretive limits. Variation may reflect seasonality, promotions, trends, stockouts, partial shipments, or calendar effects rather than random demand. Test seasonal subperiods and stock availability before setting capacity.

Frazelle. Incorporating daily variability can virtually eliminate replenishment during the pick shift (Frazelle 2016, fig. 2.34, pp. 67–70).

Item inventory profiles

06

15.2% of stocked SKUs hold 80% of units

Figure 29: 4,212 stocked SKUs (15.2%) hold 80% of average selling-unit count.

Warehouse response

Review the 4,212 inventory-dominant SKUs first. Add cube, handling units, and service targets before changing storage capacity.

Profile definition. An inventory Pareto ranks stocked SKUs by on-hand inventory and plots cumulative inventory share against cumulative SKU share. It identifies the concentrated SKU subset that dominates recorded stock.

Construction. The population contains 27,791 SKUs with positive inventory. For each SKU, average on-hand selling units are calculated from 12 monthly records covering January–December 2004. SKUs are ranked from highest to lowest average inventory before both cumulative shares are calculated.

Observed pattern. The curve reaches 80% of average on-hand units after 4,212 SKUs, or 15.2% of the positive-inventory population. Inventory is therefore concentrated, although the empirical split is not assumed to equal 80/20.

Warehousing implication. Prioritize these SKUs for inventory-parameter, location-capacity, and storage-mode review. Combine average units with item cube, handling unit, replenishment frequency, lot size, and service targets in the final capacity or assignment model.

Interpretive limits. Average units are not occupied volume, pallet positions, inventory value, or excess stock. Monthly averaging conceals daily peaks, seasonality, and stockouts. The profile prioritizes investigation; it does not prove that inventory should be reduced.

Frazelle. Inventory Pareto analysis belongs to the item-inventory profiles and frames it as separating a concentrated minority from the remaining SKU population (Frazelle 2016, 24, 29, fig. 2.6).

A items hold 65.8% of average selling units

Figure 30: A items generate 80% of picks and hold 65.8% of average selling-unit count.

Warehouse response

Use this unit-count screen to select SKU-level reviews. Diagnose policy with days of supply, value, lead-time demand, fill rate, and backorders.

Profile definition. The item-family inventory profile compares each ABC class’s share of picking activity with its share of on-hand inventory. It tests whether inventory allocation is broadly aligned with warehouse demand.

Construction. SKUs are ranked by 2004 order-line picks. A items generate the first 80% of picks, B items the next 15%, and C items the final 5%. For the same SKU population and period, monthly-average on-hand units are summed by class and normalized to total inventory.

Observed pattern. A items generate 80.0% of picks and hold 65.8% of average units. B items generate 15.0% of picks but hold 28.2% of units. C items generate 5.0% of picks and hold 6.0% of units. The largest aggregate imbalance is in the B class, not the C class.

Warehousing implication. Review A-item fill rate, backorders, lead time, and forward-pick replenishment. Review B-item lot sizes, safety stocks, inactive stock, and capacity at SKU level. Retain required C items when service-parts, assortment, or order-completion obligations apply.

Interpretive limits. Inventory share alone cannot diagnose shortage or overstock. Add days of supply, fill rate, backorders, lead-time demand, value, and SKU-level exceptions. Do not transfer Frazelle’s illustrative excess-C conclusion to this company distribution.

Frazelle. The profile identifies inventory imbalances and explicitly warns that slow-moving items may be required for service or assortment coverage (Frazelle 2016, fig. 2.35, pp. 70–71).

Slow movers need explicit retention and storage rules

1. Justify retention

Check service obligations, assortment completion, contracts, lead time, and inventory age at SKU level.

2. Assign storage

Choose the least-accessible feasible storage position. Test dense reserve or batch picking against retrieval work and the service promise.

Do not infer obsolescence from class C

Use demand, inventory age, service requirement, value, and stockout evidence at SKU level before disposal or relocation.

On-Hand Inventory Profile. I receive a lot of phone calls that begin with clients complaining about the lack of space in their warehouse. More often than not, the problem is not too little space, but too much inventory. This profile (Figure 2.35) helps us identify the source of the inventory problem. As is true in the example, most companies have too little type A inventory (back orders and customers screaming for those products) and too much type C inventory (obsolete stock that nobody wants and nobody has the courage to discard). By drawing the picture, we can at least illustrate the magnitude of the problem to management and present a list of “problem items” for their review.

In some cases, C items should be removed from inventory. However, in many cases, warehouses are required to house the C items. One example is in the service parts business, where you may be required to support a certain model number in the field for up to 5, 10, even 20 years. Another example is in retailing when some key type C items protect the sales of items A and B. On a recent project in the grocery industry, the chairman of the company was presented with a recommendation to eliminate the C items from inventory. Let’s pause and consider the consequences. How many items do you buy on your weekly trip to the grocery store to restock the kitchen cabinets? Say its 50 items. In this case, there is at least a 70 percent chance that one of those 50 items is a C item. Why did you go to that grocery store? In the case of this grocery chain, it was probably because it stocked that C item. Otherwise, you could and would shop at a warehouse store.

Even though you may not be able to eliminate the C item inventory you can at least efficiently store and pick the items. To conserve space, you can store the C item inventory in dense, high-rise racking or on the second or third level of a mezzanine. To achieve high productivity at the same time, you can batch-pick the C item pick line and locate the batch in a dedicated location along the forward pick line or introduce the batch into an automated sorting system.

— World-Class Warehousing and Material Handling (Frazelle 2016, 70–71).

32.6% of SKUs average no more than one case

Figure 31: 8,300 SKUs (32.6%) average at most one case on hand.

Warehouse response

Screen the low-case ranges for shelving or carton flow and the higher ranges for rack, reserve, or bulk storage.

Profile definition. A handling-unit inventory profile converts on-hand stock from selling units into warehouse handling units. This makes the distribution usable for storage-system planning.

Construction. Average on-hand selling units are divided by each SKU’s units per case. The population contains 25,469 SKUs with positive inventory and valid case packs. The profile pairs the SKU count with total average case equivalents for seven inventory bands from no more than one case through more than 50 cases.

Observed pattern. The no-more-than-one-case band contains 8,300 SKUs but only 2,818 average case equivalents. The more-than-50-case band contains 313 SKUs and 108,550 average case equivalents. The contrast rejects one universal storage mode.

Warehousing implication. Screen low-case ranges for shelving, carton flow, modular storage, or small faces. Screen higher ranges for case rack, reserve storage, or denser modes. Select the final assignment by minimizing space, travel, picking, and replenishment cost subject to capacity, access, FIFO, load, selectivity, and throughput constraints.

Interpretive limits. Case equivalents omit peak stock, case dimensions, cases per pallet, stackability, mass, and replenishment frequency. Frazelle’s pallet-based storage rules cannot be transferred directly to this case profile.

Frazelle. Inventory is converted into operational handling units and uses pallet ranges to screen storage systems (Frazelle 2016, fig. 2.36, pp. 71–73).

Combining profiles for warehouse decisions

07

Warehouse redesign uses profile combinations

Table 9: Profile combinations that support common warehouse redesign questions.

Redesign question	Required profile combination
Dedicated processing cell?	Customer Pareto + family mix + order completion + calendar clock
Separate case and pallet processes?	Handling mix + order increments + lines and cube per order
Which storage mode?	Popularity + cube movement + handling-unit inventory + variability
Where should each SKU be assigned?	Heatmap + popularity–cube + correlation + location capacity
How much peak capacity?	Calendar clock + volumetric + service target + measured processing rate

The five-row matrix is a course synthesis. Frazelle supplies the component profiles and the decisions they inform; the source does not print this exact matrix.

Customer Paretos. Some customers place such high demands on a warehouse, represent such a large portion of the activity of the warehouse, and have such demanding customer-service requirements that it may make sense to establish a separate area within the warehouse for those customers. For example, one of our apparel manufacturing clients does so much business with JC Penney that they have a JC Penney warehouse within their warehouse. It is not uncommon for consumer products and food manufacturing companies to have areas within their warehousing dedicated to Wal-Mart activity.

The customer pareto in Figure 2.7 is from one of our food and beverage clients. Note that just three customers make up 50 percent of our client’s distribution center activity. Accordingly, we assigned each of those customers their own zones in the warehouse to reduce travel time and improve customer service.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Customer Paretos,” pp. 28–31).

Order-Increment Profiles. Order-increment profiles illustrate the portion of a unit load requested on a customer order. For example, suppose that there are 100 cartons on a pallet and a customer orders 50 cartons. In this case, the customer ordered 50 percent of the pallet. If there are 80 cartons on a pallet and that customer orders 20, they ordered 25 percent of the pallet.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Order-Increment Profiles,” pp. 37–42).

Item-Order-Completion Profile. The item-order-completion profile identifies small groups of items that can fill large groups of orders. Those small groups of items often can be assigned to small order-completion zones in which the productivity, processing rate, and processing quality are two to five times better than those found in the general warehouse.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Item-Order-Completion Profile,” pp. 62–66).

The item-order-completion profile is constructed by ranking items from most to least popular. Beginning with the most popular item, then the two most popular items, then the three most popular items, and so on, the items are compared with the order pool to determine what portion of the orders a given subset of the items can complete. In this example, 10 percent of the items can complete 50 percent of the orders. Suppose that I walk into your warehouse and identify 10 percent of the items that can completely fill 50 percent of the orders. What would you do with those 10 percent of the items? I hope that you would create a warehouse within the warehouse or order-completion zone for those 10 percent.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Item-Order-Completion Profile,” pp. 62–66).

The design principle is similar to that used in agile manufacturing, where we look for small groups of parts that have similar machine routings. Those machines and those parts make up a small-group technology cell wherein the manufacturing efficiency, quality, and cycle time are dramatically improved over those found in the factory as a whole.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Item-Order-Completion Profile,” pp. 62–66).

We recently worked with a large media (i.e., compact discs, cassettes, videos, etc.) distributor and helped to identify 5 percent of its 4,000 SKUs that could complete 35 percent of its orders. We assigned those 5 percent to carton flow-rack pods (three flow-rack bays per pod, one operator per pod) at the front of the distribution center. Operators could pick/pack orders from the flow rack at nearly six times the overall rate of the distribution center. The distribution center has won its industry’s productivity award each of the last two years.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Item-Order-Completion Profile,” pp. 62–66).

Very frequently, there is a driving force behind order completion. Examples could include customers ordering within a brand (Figure 2.30) or, as depicted in Figure 2.31, a product group, a supplier, a size, a color, a kit, and so on. Our pattern-recognition algorithms seek and surface those patterns in an attempt to identify order-completion zoning opportunities within the warehouse.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Item-Order-Completion Profile,” pp. 62–66).

Demand-Correlation Profile. Just as a minority of the items in a warehouse makes up a majority of the picking activity, certain items in the warehouse tend to be requested together. In the example (Figure 2.32), pairs of items are ranked based on their frequency of appearing together on orders. In this case, we are examining data from a mail-order apparel company. The first three digits represent the style of the item (i.e., crew-neck sweater, V-neck sweater, turtle-neck shirt, pleated pants, etc.), the middle digit represents the size of the item (i.e., 1 = small, 2 = medium, 3 = large, and 4 = extra large), and the last digit represents the color (i.e., 1 = white, 2 = black, 3 = red, 4 = blue, 5 = green, etc.).

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

What do you think people tend to order together from this mail-order company? (I thought it would be shirts and pants that looked good together in the catalog.) What does the profile in Figure 2.32 suggest? In this case, customers tend to order items of the same style and size together. The explanation is that customers tend to get comfortable with a certain style and tend to order in multiple colors to add variety to their wardrobes. Of course, they order the same size unless they will return one for fitting. This pattern was a surprise to me. It was also a surprise to the marketing people. Surfacing truth is the crucial benefits of profiling process. (The myriad SKUs, order patterns, suppliers, and interdependent decisions make it difficult to form a reliable intuition about logistics operations.)

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

How do we take advantage of this demand-correlation information in slotting a warehouse? We zone the warehouse by item size first, creating a zone for the smalls, mediums, larges, and extra larges of all styles (Figure 2.33). Within each size area, we store items of the same style together, mixing colors within a style. This zoning strategy allows us to create picking tours based on size and style. As a result, order pickers pick many items on short-distance picking tours. Simultaneously, we manage congestion by spreading out the sizes. Golden zoning is used to slot the most popular color for each style at or near waist level.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Correlation Profile,” pp. 64–68).

Demand-Variability Profile. The demand-variability profile indicates the standard deviation of daily demand for each item. Unfortunately, an item’s daily demand is not predictable. During a recent project, we were trying to size the pick faces along a case picking line such that each pick face held a day’s worth of stock. The motivation was to make sure that we did not need to restock a location during the day. The current design had the pick faces sized for an average day’s demand, and the client could not figure out why it had to restock so many locations during the course of a day. You can probably see why. If the pick face is sized for the average day, unless the same quantity is picked every single day, there will be many days when the pick face is oversized and many days when the pick face is undersized. Once the pick faces were resized to accommodate this variability in demand, the restocking during the pick shift was virtually eliminated.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Demand-Variability Profile,” pp. 67–70).

Handling Unit Inventory Profile. The item-family inventory profile is not very useful for storage systems design because the information is not presented in material-handling terms (i.e., pallets, cases, eaches, etc.). This is a common problem with most corporate data used in planning warehouse operations—the data are expressed in terms of dollars, pounds, pieces, days of supply, turns, and so on. Although this is useful for business planning purposes, the data are not very helpful for planning and managing warehouse operations. This is another motivation for the profiling exercise—to give managers and designers of the warehouse operations a presentation of the activity of the warehouse in their terminology.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Handling Unit Inventory Profile,” pp. 71–73).

In this example (Figure 2.36), we convert the item-family inventory distribution into a distribution describing on-hand inventory in terms of pallets of merchandise on hand. As a result, we can recommend the appropriate mix of pallet storage modes. For example, the 91 SKUs with less than a pallet of inventory on hand probably should be stored in shelving or decked racking. The 104 SKUs that have one or two pallets on hand probably should be stored in single-deep pallet racks. The 68 SKUs that have three to five pallets on hand probably should be stored in double-deep and/or push-back racks. The remaining SKUs, those with more than 10 pallets on hand, probably should be stored on the floor in deep block stacking lanes, in drive-in/through racks, and/or in pallet flow lanes.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Handling Unit Inventory Profile,” pp. 71–73).

The warehouse-within-a-warehouse design philosophy works because small warehouses, in general, have higher productivity, response time, and accuracy performance than large warehouses. Many of the customer-order and item-activity profiles identify opportunities to subdivide a warehouse operation into self-contained warehouse processing cells, virtual warehouses, or warehouses within the warehouse. A warehouse within a warehouse is similar to a flexible manufacturing cell in a factory, where the productivity, quality, and cycle time are much better than in the factory as a whole.

— World-Class Warehousing and Material Handling (Frazelle 2016, 31).

Family-Mix Profiles. The overall operating strategy of a warehouse should be influenced by the extent to which orders require items from multiple families. If the majority of the orders are pure, drawing from a single product family, then zoning the warehouse on that basis could create virtual warehouses within the warehouse.

— World-Class Warehousing and Material Handling (Frazelle 2016, 32).

The profile strongly suggests that separate areas for full and partial picking should be established. Should we have separate case picking and pallet picking areas? If so, would we pay a big penalty for mixed orders that require merging of the partial- and full-pallet portions of the order? No, only a small minority of the orders are mixed. When the orders are released into the warehouse management system, it should classify them immediately as pallet pick orders, carton pick orders, or mixed orders. For mixed orders, the warehouse management system should create a pallet portion and a case pick portion and either pass the full-pallet portion to the case pick area or merge the case pick and pallet portions downstream from picking.

— World-Class Warehousing and Material Handling (Frazelle 2016, 34–37).

Our customer order volumetrics include lines-, units-, cube-, and weight-per-order profiles; and joint profiles comprised of the same. The lines-per-order profile portrays the number of unique stock-keeping units (SKUs) on an order. It is one of the most important profiles because each unique SKU or line on an order represents a visit to the unique location for that SKU. Location visits are the majority of the workload in any warehouse. Consequently, managing location visits is perhaps the master key to maximizing warehouse productivity.

— World-Class Warehousing and Material Handling (Frazelle 2016, 42–44).

The item activity profile is used primarily to slot the warehouse, to decide for each item (1) what storage mode the item should be assigned to, (2) how much space the item should be allocated in the storage mode, and (3) where in the storage mode the item should be located. The item activity profile includes the following activity profiles:

• Popularity profile
• Cube-movement profile
• Popularity-cube-movement profile
• Order-completion profile
• Demand-correlation profile
• Demand-variability profile

— World-Class Warehousing and Material Handling (Frazelle 2016, sec. 2.4, p. 56).

Calendar-clock profiles reveal peaks and valleys in warehouse activity so that all the elements of the warehouse infrastructure—workforce, material-handling systems, storage systems, and dock doors—can be sized optimally. Calendar-clock profiles include the following:

• Month-of-year (MoY) activity profile
• Week-of-month (WoM) activity profile
• Day-of-week (DoW) activity profile
• Hour-of-day (HoD) activity profile

— World-Class Warehousing and Material Handling (Frazelle 2016, sec. 2.3, pp. 48–49).

The most popular and traditional warehouse performance measure is productivity (Figure 3.18). The formal definition of productivity is the ratio of the output of an entity to the resources consumed achieving that output. We recommend that our clients monitor the productivity and utilization of all key assets in the warehouse—labor, space, material-handling systems, and warehouse management systems. We typically measure overall labor productivity as the ratio of units, orders, lines, or weight shipped out of the warehouse to the number of person-hours spent in operating, supervising, and managing the warehouse.

— World-Class Warehousing and Material Handling (Frazelle 2016, “Warehouse Productivity Performance,” p. 95).

For the final row, the service target is an input requirement and the measured processing rate converts workload into required capacity. Scenario and sensitivity tests should then challenge growth, order-profile changes, and abnormal peaks (Baker and Canessa 2009, 427 and 432–433).

The heatmap and location-capacity checks in the fourth row are course analytical inputs, not named Frazelle profiles. The heatmap locates activity in the current layout; the capacity check tests whether a proposed assignment physically fits.

Combine only decision-relevant profiles

Decision rule

Use the smallest set of profiles that changes the warehouse decision, then test the proposed design against capacity, travel, service, and cost.

Course synthesis based on Frazelle, Bartholdi and Hackman, and Baker and Canessa. The smallest-set stopping rule is a course heuristic. The sources support selecting measures that illuminate the decision and eliminating unsuitable alternatives. The complete FIFO, ergonomics, congestion, service, load, access, space, travel, and capital list is a course-defined feasibility checklist, not a named framework in the cited literature.

Combined evidence. Every profile omits part of the warehouse. Customer concentration does not show whether orders share products. Popularity does not show item cube. Handling-unit inventory does not show pick frequency. Calendar profiles do not provide a processing rate.

Decision sequence. Start with one redesign question. Select the profiles that describe its workload, add the required operational parameters, and define the current and proposed alternatives. Compare alternatives with the same study period and service requirement.

Optimization connection. A slotting or storage-mode model may minimize expected travel, handling, space, and capital cost. Feasibility constraints must include location capacity, product compatibility, replenishment access, load limits, FIFO, ergonomic requirements, congestion, and service.

Stopping rule. Stop adding profiles when another view will not change the candidate alternatives, constraints, or evaluation criteria. The analysis is complete when it supports an implementable experiment and defines how the result will be measured.

The Power of Profiling. Done properly, the patterns in profiling quickly reveal a small set of candidate process flows and material handling equipment types for each warehouse activity. Conversely, profiling quickly eliminates process and equipment options that really do not warrant consideration. Many warehouse reengineering projects go awry because we waste valuable resources analyzing concepts that were doomed from the start.

— World-Class Warehousing and Material Handling (Frazelle 2016, 22–25).

An activity profile is essential to really understand what matters in a warehouse. You can build this from data about the physical layout of the warehouse, the SKUs stored therein, and the patterns of your customer orders. The activity profile will enable you to understand, manage, and improve use of labor, space, and equipment.

Warehouse activity profiling is a special case of data-mining, which is simply the rummaging through databases to look for patterns that might be exploited to improve operations. As in mining for minerals, success depends on having good tools to support the search and on knowing where to look. Therefore you must be comfortable with SQL and databases and you must understand warehouse operations.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 251).

Oxley (1994) states that this step is largely concerned with validating the operational and technical feasibility of the proposed solutions, checking that it meets the requirements of step one (i.e. the initial requirements), and undertaking capital and operational cost evaluations.

Scenarios can be built, either by simulation or other modelling, to consider a series of different situations in which the flexibility of the design can be tested. These scenarios may include for example alternative growth forecasts, changes in order profiles, and abnormal peak requirements.

— Warehouse design: A structured approach (Baker and Canessa 2009, 427–33).

Profile quality workshop

08

One function reproduces any profile

The shared loader validates the transaction population before a named profile is rendered.

Listing 1: The named profile determines the aggregation, chart, and warehouse interpretation.

from snippets.activity_profiles import load_profile_data
from snippets.activity_profiles import render_profile

profile_data = load_profile_data()
figure = render_profile("customer_pareto", profile_data)

Internal software provenance; public gap remains. The behavior described here is implemented in warehousing-activity-profiling/snippets/activity_profiles.py: load_profile_data() prepares the shared data and render_profile() dispatches a registered profile name. The recovered commit-pinned GitHub URL is private and returns 404 for an anonymous student. It therefore cannot serve as a student-checkable citation. This note remains unresolved until the code or a versioned technical specification is published at an anonymously accessible stable URL.

The customer Pareto uses one observation per customer. It counts order lines, ranks customers by descending activity, and calculates cumulative customer and line shares. Replace the profile name to render another warehouse distribution from the same validated transaction population. Each profile retains its own unit of analysis, calculation, filters, and decision interpretation.

The first chart cannot support a staffing decision

Figure 32: The source chart shows dates and series, but omits the year, measure definition, population, and decision threshold.

Why it fails

• “Orders” does not identify receipt, release, pick, or dispatch.
• The year and exact study period are absent.
• The vertical axis has no unit or population.
• No capacity requirement or warehouse decision is stated.

Course-artifact observation. The missing year, undefined event, unlabeled unit, unspecified population, unexplained zero-shipment days, and absent decision threshold are defects observed directly in the supplied course figure. The literature below explains why operational definitions, workload measures, and readable charts matter; it is not provenance for those observed artifact defects.

Ask students to identify only genuine defects. The source names its series and shows calendar dates. It does not state the year, define what an “order” event means, label the vertical unit, identify the population, explain zero-shipment days, or connect the pattern to a capacity decision. The generated trace is a versioned course reconstruction from the supplied exercise image, so every repair uses the same fixed arrays. The private first-introduction commit c8cb79376e1505d60e9dcf72224081b8f542f9be establishes internal continuity, but the image identifies no author, publication, page, facility, year, or caption and is not anonymously accessible. Upstream provenance remains unresolved.

A warehouse is a complicated and busy place and it can be hard to get an accurate sense of what is happening. Warehouse activity profiling is the careful measurement and statistical analysis of warehouse activity. This is a necessary first step to almost any significant warehouse project: Understand the customer orders, which drive the system.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 233–34).

Use graphs to show large scale patterns, especially when you want to compare with other patterns. And when you want graphs to be compared, draw them to the same scale. Show percentages. Use tables for closer looks when the actual numbers are important. Scale the data to make it easier to understand. For example, it is better to report average lines per day rather than total lines over the period of study. Most people will find it easier to understand the implications of the numbers. The defaults on standard office software, such as MS Excel, are generally poor. If you use these tools, you will need to intervene to make a quality graph.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 251).

Defined events and units expose the pattern

Figure 33: Defined warehouse events and units make the digitized daily pattern interpretable.

What changed

• Define customer orders received and shipped.
• State orders per calendar day.
• Retain the source dates and disclose the missing year.
• Preserve the same digitized values and vertical scale.

Course reconstruction based on Bartholdi and Hackman. The course version adds event definitions and units without changing its fixed reconstructed values. Receipt and ship-confirmation are instructional definitions; the unattributed image does not establish those event meanings. The missing source year, facility identity, author, publication, and page remain unresolved. The book supports the semantic repair, not the provenance of the trace.

The second version repairs the semantics without replacing the reconstructed course trace. The upstream year and facility remain unavailable, so they are disclosed instead of invented. The reader can now distinguish the instructional receipt and ship-confirmation series, but must still identify the recurring operational pattern.

For analyzing warehouse operations you have to be careful where you obtain this data. Often this data is from a sales transaction database, which tracks financial rather than warehouse events. Consequently, the date recorded may represent the date the order was placed or when it was printed, not when it was processed in the facility. Similarly, the order that appears in the sales transaction database may actually have been processed at another facility. Generally, though, the information is available somewhere because each day the order pickers had to know what to pick.

Keep in mind that an order history is primarily financial information. This is good in that it is likely to be very accurate; but it can also be misleading because the transactions it represents are financial and not necessarily operational. For example, it can happen that a SKU is shown to have been requested in a negative amount; but this generally means something like the item was returned and restocked to the shelf.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 239–40).

Weekly shipping peaks warrant a backlog test

Figure 34: Recurring shipping peaks exceed an assumed 400-order daily capacity; the backlog mechanism remains a hypothesis.

Warehouse response

• Quantify the recurring weekly peak.
• Label 400 orders/day as an assumed capacity.
• Test weekend closure and backlog records.
• Compare temporary staffing with release smoothing.

Course reconstruction and explicit assumption. The zero-shipment days and following peaks are observations from fixed arrays in the local course code, not from an identified public dataset. The recovered GitHub artifact is private and cannot be checked anonymously. The upstream author, publication, facility, period, and event provenance therefore remain unresolved. The 400-orders/day line is an assumed course scenario parameter, not a measured warehouse capacity.

The reconstructed course trace shows recurring zero-shipment days followed by weekly shipping peaks. That pattern is consistent with closure-related backlog, but the chart alone does not prove the operating calendar or cause. The red line is an explicit scenario parameter: assumed nominal shipping capacity equals 400 orders/day. Validate closure, backlog, and ship-confirm records, then compare temporary peak staffing, weekend dispatch, and order-release smoothing on service, backlog, and labor cost.

Scenarios can be built, either by simulation or other modelling, to consider a series of different situations in which the flexibility of the design can be tested. These scenarios may include alternative growth forecasts, changes in order profiles, and abnormal peak requirements.

— Warehouse design: A structured approach (Baker and Canessa 2009, 432–33).

This passage supports testing alternative conditions. It does not establish the course trace, a weekend-closure cause, real backlog, or the 400-orders/day assumption.

Repair one profile for a warehouse manager

Choose one profile and revise it in three passes:

1. Chart completeness: Define population, period, unit of analysis, denominator, units, legend, and thresholds.
2. Evidence: Quantify the dominant concentration, peak, mixture, or exception.
3. Improvement strategy: State one warehouse action and the assumption that must be validated.

Manager test

Hide the report text. Can a warehouse manager understand the evidence and decision from the title, figure, caption, and warehouse-response box?

Course-defined exercise protocol. The three passes and manager test are instructor-created. The public formative criteria require contextual scope, explicit assumptions and method, quantitative precision, appropriate and readable charts, operational insight, and the next analysis required before a recommendation (Beirigo 2026, sec. “Formative quality criteria”). The literature below explains the analytical reasons for those course checks.

Students should preserve the underlying calculation while improving its semantic and decision quality. Ask them to report any excluded outlier and its rule, avoid pie charts with many categories, and confirm that compared metrics use compatible periods and units.

Use graphs to show large scale patterns, especially when you want to compare with other patterns. And when you want graphs to be compared, draw them to the same scale. Show percentages. Use tables for closer looks when the actual numbers are important. Scale the data to make it easier to understand. For example, it is better to report average lines per day rather than total lines over the period of study. Most people will find it easier to understand the implications of the numbers. The defaults on standard office software, such as MS Excel, are generally poor. If you use these tools, you will need to intervene to make a quality graph.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 251).

Oxley (1994) provides a fairly comprehensive list of steps that incorporates the key features of the previous authors. He starts with defining the overall system requirements of the supply chain, including such factors as service levels and implementation time constraints. Again, data collection and analysis are key steps. He also introduces a new step of establishing the unit loads to be used. The following steps are again concerned with developing alternative operating methods, equipment types and layouts. He stresses that the warehouse design should be centred on the storage and handling requirements and that the building should then be designed around these.

— Warehouse design: A structured approach (Baker and Canessa 2009, 427–33).

Present a checked profile in five parts

1. Decision: Name the warehouse choice the profile informs.
2. Unit: Define the observation, period, numerator, denominator, and unit.
3. Distribution: Show the full pattern with readable labels and scale.
4. Finding: Quantify the concentration, mix, scale, timing, or exception.
5. Test: Compare one warehouse change using service, capacity, travel, and cost.

Final check

A warehouse manager should understand the observed work, the proposed change, and the evidence still required.

Instructor-created mnemonic mapped to the public course criteria. The five-part structure compresses the course-defined ten checks; no cited literature source presents this exact mnemonic or claims that it is sufficient by itself. Decision maps to context, insight, and next analysis. Unit maps to scope, assumptions, method, and precision. Distribution maps to visualization quality. Finding maps to precision and insight. Test maps to the next analysis required before a recommendation (Beirigo 2026, sec. “Formative quality criteria”).

These five parts compress the ten-check evaluation list for a live presentation. Ask students to state them without looking at the checklist. The decision determines which profile belongs in the analysis. The unit and denominator prevent calculation errors. The distribution shows the operating pattern. The finding names its magnitude. The test connects the profile to a feasible warehouse alternative and states the missing evidence.

Warehouse activity profiling and data mining is the systematic analysis of item and order activity, seeking to identify and take advantage of patterns of opportunity. Our pattern recognition algorithms highlight the root cause of material and information flow problems, pinpoint major opportunities for process improvements, and provide an objective basis for project-team decision making.

— World-Class Warehousing and Material Handling (Frazelle 2016, 21–22).

An activity profile is essential to really understand what matters in a warehouse. You can build this from data about the physical layout of the warehouse, the SKUs stored therein, and the patterns of your customer orders. The activity profile will enable you to understand, manage, and improve use of labor, space, and equipment.

— Warehouse & Distribution Science (Bartholdi and Hackman 2019, 251).

References

Sources cited in this deck

Baker, Peter, and Marco Canessa. 2009. “Warehouse Design: A Structured Approach.” European Journal of Operational Research 193 (2): 425–36. https://doi.org/10.1016/j.ejor.2007.11.045.

Bartholdi, John J., and Steven T. Hackman. 2019. Warehouse
& Distribution Science. Release 0.98.1. The Supply Chain
& Logistics Institute, Georgia Tech. https://doi.org/10.13140/RG.2.1.4264.0481.

Beirigo, Breno Alves. 2026. “Warehouse Activity Profiling.” https://brenobeirigo.github.io/course-wh/assessment/activity-profiling/assignment_activity_profiling.html.

Brin, Sergey, Rajeev Motwani, and Craig Silverstein. 1997. “Beyond Market Baskets: Generalizing Association Rules to Correlations.” Proceedings of the 1997 ACM SIGMOD International Conference on Management of Data (New York, New York, USA), 265–76. https://doi.org/10.1145/253260.253327.

Frazelle, Edward. 2016. World-Class Warehousing and Material Handling. Second Edition. McGraw-Hill Education.

Medrano-Zarazúa, Lucy, Jania Astrid Saucedo-Martínez, and Johanna Bolaños-Zuñiga. 2024. “Storage Location Assignment Problem in a Warehouse: A Literature Review.” In Computer Science and Engineering in Health Services, edited by José Antonio Marmolejo-Saucedo, Roman Rodríguez-Aguilar, Pandian Vasant, Igor Litvinchev, and Brenda M. Retana-Blanco. Springer International Publishing. https://doi.org/10.1007/978-3-031-34750-4_2.

Reyes, Juan José Rojas, Elyn Lizeth Solano-Charris, and Jairo Rafael Montoya-Torres. 2019. “The Storage Location Assignment Problem: A Literature Review.” International Journal of Industrial Engineering Computations, 199–224. https://doi.org/10.5267/j.ijiec.2018.8.001.

“Warehouse & Distribution Science.” n.d. Accessed May 16, 2023. https://www.warehouse-science.com/wh/projects/spr3/main.html.