Summary
The storage location assignment problem (SLAP) is the decision of which SKU goes in which storage location. In its standard form the objective is to minimize the sum of pick frequency times travel distance: each SKU adds how often it is picked, multiplied by the distance from its location to the depot, to the total. Warehouses call the same decision slotting.
The simple version has a simple answer: rank SKUs by pick frequency, rank locations by distance, and pair them off. That stops working as soon as a picker collects several lines per tour, SKUs are ordered together, or not every SKU fits every location. Then the problem is NP-hard, and an exact method published in 2024 solves medium-sized instances, far smaller than a real building.
So every warehouse runs a rule instead. Fast movers up front, heavy cases on the bottom beam, whatever slot is free. Each rule is a shortcut through the same model, and the research is clear about what each one assumes and where it falls over.
The storage location assignment problem objective: demand times distance
The decision in SLAP is a yes or no for every SKU and location pair. The answer is yes if the SKU is stored in that location and no if it is not. Each SKU gets one location, each location holds at most one SKU, and the objective scores the result.
The standard objective is demand-weighted distance: multiply the pick frequency of each SKU by the travel distance of its location, add that up over all SKUs, and keep the total as small as possible. Pick frequency is counted in picks or order lines per period, and travel distance runs from the location to the depot and back. A literature review of SLAP research found travel distance and travel time to be the variables most used as the objective.
| Term | What it means | Where the data comes from |
|---|---|---|
| Placement | Yes if a SKU is stored in a location, no if not | The answer the model returns |
| Pick frequency | How often a SKU is picked: picks or order lines per period | Order line history, 12 months or more |
| Travel distance | Distance from a location to the depot and back | A layout map with a position for every location |
| Space and capacity | Space a SKU needs and space a location holds (capacity variant) | Verified dimensions per SKU, a location master with sizes |
View this table on desktop for the full model.
Pick frequency comes from the order history, distance comes from the layout. The order data is usually there. The distances often are not, because a WMS location master holds a code per location and rarely a position. So the travel distance of each location has to be built from a warehouse layout map that knows the aisles, the cross-aisles and the one-way rules.
Why the objective is travel
Travel is the target because it is the largest block of picking time. An order picker spends roughly 55 percent of the time traveling, 15 percent searching, 10 percent extracting and 20 percent on paperwork and other work, according to Bartholdi and Hackman's Warehouse and Distribution Science. Picking itself is expensive: a 2025 case study in Applied Sciences puts order picking at 50 to 75% of total warehouse operating costs.
- Traveling55%
- Searching15%
- Extracting10%
- Paperwork and other20%
Variants change the weights, and the shape of the model stays. Swap distance for time and high locations get more expensive: in that same 2025 study, retrieval took 0.3 minutes per item from levels 1 to 3 and about two minutes from levels 4 and 5.
Add storage cost, a penalty for congestion, or the cost of moving a SKU from where it sits today, and you still have an assignment of SKUs to locations with a score for every pair.
Why the storage location assignment problem is NP-hard
The textbook version is easy to solve. If every trip picks one line and returns to the depot, and every SKU fits every location, the best plan pairs the most-picked SKU with the closest location, the second with the second, and so on. When SKUs need different amounts of space, the cube-per-order index sets the order: space needed divided by pick frequency, lowest ratio closest. For single-command picking, COI slotting is known to minimize travel time.
Real warehouses break that in three places, and each one makes the problem harder in a different way.
Multi-line pick tours tie the slot plan to the route. A picker collecting twenty lines walks one tour, so the cost of a location depends on which other locations are on the same pick list.
The 2024 study on storage location assignment and picker routing puts it both ways: a slot plan must be known before routes can be planned, and routes must be planned before a slot plan can be judged. That combined problem has been proven NP-hard in the strong sense, even in very basic cases.
SKUs that are ordered together turn the sum into a product. Correlated storage scores a pair of SKUs by how far apart their two locations are, so the objective multiplies two placement decisions. That is a quadratic assignment problem, and a 2023 review of correlated storage notes that most of these models are NP-hard integer programs.
Different sizes and moving demand add decisions of their own. Once a SKU can take one, two or three locations of different sizes, the model also chooses quantities. Once demand moves from month to month, it also chooses when to move stock.
Exact methods stop well short of a real building. The branch-cut-and-price method in the 2024 study solves medium-sized instances, and one benchmark set it was tested on has between 10 and 100 SKUs. A distribution center runs thousands. That is why practice runs on heuristics: rules that give a good answer fast, with no proof that it is the best one. The useful question about a rule is which conditions it assumes, because those are what fail.
Eight goals a slot plan trades off
Demand times distance is one goal. A real slot plan is scored against eight, and seven of them can be dialled up or down.
- Short travel: fast movers near the depot. The cost is that those aisles get crowded.
- Low congestion: fast movers spread across aisles so pickers do not block each other. The cost is longer walks.
- Ergonomics: fast and heavy items at safe working height. The cost is that the good heights fill up fast.
- Safe stacking: sturdy items picked before fragile ones. The cost is that the route stops following velocity.
- Fewer mispicks: look-alike SKUs kept apart, which splits product families up.
- Grouped families: one short loop per family, which puts look-alikes side by side.
- Fill rate: items in bins close to their own size. The best-fitting bin is rarely at the best picking height.
- Compliance: hazard classes, weight limits, temperature zones. This one stays fixed, and everything else is optimized around it.
The cost of each goal usually lands on another goal in the same list. Grouping families puts look-alikes side by side, short travel crowds the aisles, and fill rate fights ergonomics for the same shelves. No assignment wins on all eight, so any plan that does not say which goal it weighted highest is hiding that choice. Pick a goal in the chart to see which ones pay for it.
Test data, for illustration.
Trade-off radar from the Optioryx slotting page. Pick a goal to push it up and see which goals pay for it. Test data, for illustration.
Congestion is the goal with the most counter-intuitive research behind it. The obvious remedy is wider aisles, so pickers can pass each other. Pratik Parikh and Russell Meller modelled pick-face blocking in exactly that setting and found that a wide aisle has less blocking than a narrow one when each picker takes a single SKU at a pick face. When pickers take more than one, blocking rises monotonically with the number of SKUs picked there (IIE Transactions, 2009, 41(3), 232 to 246).
Open access work on picking route blockages describes the everyday version. Aisle blocking happens when a forklift is parked in the aisle, when carts or pallets are sitting in the area, or when someone else is picking there (Mathematics, 2024). Both kinds of blocking depend on how many pickers need the same aisle at the same time, and the slot plan decides much of that.
What constrains the assignment in a real warehouse
The objective is clear enough. Eight kinds of constraint act on the assignment at once, and each one needs data that most warehouses have not checked.
One of these blocks more projects than the rest: missing cube data. The basic model only needs pick frequency and travel distance. The moment it also weighs the space each SKU needs against the space each location holds, every rule that reasons about fit needs verified dimensions per SKU. An item master built up over years holds a mix of measured, estimated and inherited numbers, and the estimated ones look identical to the measured ones in the record.
Checking the file against an item master data standard is the cheapest way to find out how much of it is real. Slotting is one of several processes that run on item dimensions, next to receiving, cartonization and load building, and why item dimensions matter covers the others. Teams that measure the missing SKUs with an iPhone or iPad instead of a tape use Flux for it, described in the Flux product sheet.
The one constraint that moves on its own
Seven of the eight constraints in that table hold still. Inventory behavior does not. Demand shifts with season and promotion, so an assignment fitted in March can be the wrong one by November without anyone having touched the plan. In the model, the pick frequencies change while the assignment stays where it was.
Peak is when that surfaces, because a rule that has worked without trouble all year meets peak volume at once. Concentrating fast movers near the depot shortens the walk and fills those same aisles with pickers, so the goal worth the most in a normal week is not the one worth the most in November. The peak weightings are set out next to the normal-week ones in 6 slotting mistakes warehouses make before peak season.
Storage assignment policies: the rules warehouses use
Seven rules cover almost everything running in warehouses today, sorted here by how tightly a SKU is bound to a location. Random, closest open location, dedicated, full-turnover and class-based storage follow the standard taxonomy of storage assignment in the order picking review by de Koster, Le-Duc and Roodbergen, which also describes family grouping. Scattered storage is the newer arrival.
| Rule | How it assigns | What it needs | Where it breaks |
|---|---|---|---|
| Random | Every incoming load goes to a free location picked at random. | A live map of empty slots | Does nothing for travel. Mostly used as the benchmark others are measured against. |
| Closest open location | The first empty slot the operator passes. | Nothing | Concentrates stock near the entry point and degrades as the building fills. The default when nobody has chosen a rule. |
| Dedicated (fixed) | Each location is reserved for one SKU and holds nothing else. | A fixed SKU-to-slot map, sized for peak | Costs space. Every slot is sized for peak stock and stands part-empty the rest of the year. |
| Full-turnover | Every SKU ranked by turnover, fastest nearest. | Pick frequency and space per SKU | The ranking has to be recomputed and physically re-slotted whenever demand shifts. |
| Class-based (ABC) | SKUs cut into popularity classes, each class gets a zone, random assignment inside it. | A demand ranking coarse enough to band | The compromise between full-turnover and random. Its coarseness leaves travel savings on the table. |
| Family grouping | Products ordered together, or from the same family, stored in the same region. | Order line history, not only pick counts | Competes with turnover for the same prime slots. |
| Scattered | Small quantities of each SKU spread through the building, so a piece of anything is always nearby. | A WMS tracking piece-level locations | Needs much more replenishment labor. |
View this table on desktop for the full data.
Full-turnover and class-based storage are the sorting answer from the textbook model, applied to a whole building. Full-turnover ranks every SKU and places them in order of distance. Class-based storage, often called ABC storage, cuts the ranking into a few classes, gives each class a zone and places SKUs at random inside it.
A SKU then only moves when it crosses a class boundary, which is why ABC is so much easier to keep up than a full ranking. The animation from our slotting page shows the effect on one floor.
The same stock placed at random, then sorted by pick frequency with the fast movers up front. Slow movers in blue, fast movers in orange. Animation from the Optioryx slotting page, no data behind it.
What the table leaves out
- The cube-per-order index is a ranking rule inside full-turnover, and it only has its guarantee for single-command picking. For multi-line tours the same paper shows its worst case has no limit, and most warehouses do not run single-command picking.
- Shared storage answers a different question: whether one SKU may occupy more than one location at a time. Spreading a SKU means less of it sits in each location, so locations empty sooner and the space returns to the pool. Scattered storage is that idea taken to its limit.
- Golden zone placement is vertical. It reserves waist-to-shoulder height for the fastest movers, so it layers on top of any of the seven rules.
Which one is best
None of them, universally. A 2025 comparison found dedicated storage better on waiting time and tardiness in most instances, while randomized storage gave the lower transportation cost in 11 of 12 large-scale cases in the same study. Same instances, opposite winners, depending on which goal you are judged on.
One finding does carry over. The 2025 case study in Applied Sciences ran three storage allocation rules against three routing methods in a working warehouse for five weeks. The storage rule had a stronger influence on picking efficiency than the routing method, and the family-based rule cut total picking time by 8 to 12 percent.
How to solve the storage location assignment problem in practice
On a working floor the model needs one more term: the cost of moving. A plan that scores slightly better on paper but moves most of the building will not get run. A useful solver counts every move and keeps the current location as an option, so leaving a SKU where it is stays a valid answer.
That is how Pulse handles slotting. Pulse scores every SKU and location pair and solves the assignment, with a cost that combines frequency-weighted distance to the start and end of a tour, how often SKUs ship together, and the location's tags, weight and size limits.
The result is a ranked move list, small enough for one operator to walk at the end of a shift, rebuilt the next week against fresh order data. A full reslot uses the same model when a season or a layout changes.
Before anything moves, the scenario shows what the moves buy. Each run reports travel distance, pick time, the distance the moves themselves take and the compliance rate, all against the as-is plan. It starts in the web app with an order history and a layout, without a WMS integration, and your WMS stays the system of record.
Pulse solves slotting and picking as one problem. A scenario slots first and then routes the pick tours across the new layout, so the two gains add up. Combined, order clustering, pick path routing and slotting cut picker walking distance by up to 50%. To put a first number on the picking side for your own team, the order picking ROI calculator turns four inputs into a yearly saving and a payback time.
The practical steps, from finding the moves worth making to running them without stopping the floor, are in the guide to smarter warehouse slotting and in how to reslot a warehouse without shutting down operations.
Questions?
The standard objective is to minimize demand-weighted travel distance: the sum, over every SKU, of its pick frequency times the distance from its location to the depot. Variants use travel time instead of distance, or add storage cost, congestion, ergonomics and the cost of moving SKUs. The decision stays the same: which SKU goes in which location.
Most real versions are. The basic case, with single-line picks and locations that fit every SKU, is solved by sorting SKUs by pick frequency and locations by distance. Multi-line pick tours, SKUs that are ordered together and SKUs that do not fit every location make the problem NP-hard, which is why warehouses use heuristics and storage rules.
They describe the same decision. SLAP is the name used in operations research, where the decision is written as a model with an objective and constraints. Slotting is the name used on the warehouse floor for deciding, and keeping up to date, which SKU sits in which location.
Class-based storage, often called ABC storage, ranks SKUs by pick frequency, cuts the ranking into a few classes and gives each class a zone, with the fastest class closest to the depot. Inside a zone, SKUs go to any free location. It sits between full-turnover storage and random storage and is easier to maintain than a full ranking.
At minimum, order line history of 12 months or more for pick frequency, and a layout map with a position for every location for distance. Dimensions and weights per SKU, location sizes and weight limits, and zone or hazard tags are needed as soon as the model has to respect space and compliance.