Why manual warehouse slotting slows down shipping
Manual product placement relies on warehouse worker experience and "put it nearby" logic rather than shipping statistics. A fast-moving item ends up deep in the rack or on the top tier, while the picking zone fills with items taken once a month.
While the SKU count is small, the scheme holds; as the assortment grows, it stops being sufficient, and the picker on shift is the first to feel it.
Overflow of the picking zone is the most expensive point. Cells are occupied by slow-moving goods, fast-moving goods do not fit, and space in pallet storage is allocated for them.
Every order after that requires descending into the depths of the warehouse, and replenishment turns into a constant flow: some pickers wait for replenishment, others unpack a pallet right in the aisle.
The second effect is extra picker movements. The route is built spontaneously, so the picker returns to aisles already traversed, goes up to the top tier for a single unit, and looks for an item that was moved the day before.
The bottleneck of order picking is not people or equipment, but the route geometry inherited by the warehouse.
The third layer is downtime during replenishment: the rule for what and when to bring into the picking zone is not defined, replenishment starts based on the fact that "there are no cells" and lands exactly on the shipping peak, when the shift is already at its limit.
Warehouse slotting is needed to untangle these knots separately. We break down how this is addressed: what data on shipment frequency is needed, where a rule that places goods into cells is sufficient, and what to do with items that physically do not fit into the picking zone.
What automated warehouse slotting gives
Slotting produces an effect not "across the warehouse as a whole," but in measurable values taken before and after on the same SKU set and in the same load window.
The model calculates the frequency of joint demand by "item — order" pairs, distributes top positions into fast-turnover cells, and recalculates the picking route.
Without a baseline measurement, any such rearrangement turns into an argument about preferences, so we record five metrics.
| Metric | Before | After | Due to what |
|---|---|---|---|
| Order picking time | route across the entire picking zone | dense cluster of picks, travel is reduced | top positions in neighboring cells |
| Number of movements across the warehouse | replenishment as needed | planned, based on demand forecast | stock in the picking zone calculated in advance |
| Turnover of the picking zone | "dead" cells near aisles | every cell works its turnover | frequency rank instead of historical location |
| Equipment downtime | waiting for a cell to free up | fewer counter routes | picking and replenishment flows separated |
| Repairs and write-offs | damage in cramped areas | defect rate decreases | extra lifts and rearrangements removed |
Some failures disappear entirely rather than becoming less frequent. These are zero-picks because the item is physically in another cell, repeated trips for a forgotten position, and losses from rearrangements that were initiated manually for a new assortment.
We test the effect on scenarios that break a static layout:
- peak shift and seasonal demand spike;
- introduction of a new product group and phase-out of an old one;
- change in the storage scheme for oversized goods;
- picker returning from the replenishment zone with a cancelled task.
Layout rules are recalculated on a schedule and after a noticeable demand shift — otherwise the effect lasts one shift and goes away.
We compare before and after in a report using the same metrics, so the customer sees which specific value produced the result, rather than a general statement that "it got better."
Comparison of placement approaches: rules, learning, model
Warehouse rarely operates in one mode: rack, bulk, mezzanine, and cold storage place different requirements on product adjacency. On top of that, seasonality rewrites the demand map two or three times a year, and a layout built for one peak breaks at the next.
Therefore the choice is not a "correct" approach, but the one that will withstand your order flow and available data volume.
Below are alternatives between which a real choice exists, and the limitations of each that appear not in a test environment but in daily operations.
| Approach | When it fits | Limitations |
|---|---|---|
| Static placement rules | Stable assortment, one or two storage types, even demand | Does not react to seasonality; zones are reviewed manually and rarely |
| Rules with seasonal coefficients | Planned peaks by category, assortment changes slowly | Each season is configured manually; coefficients become outdated by the middle of the period |
| Trainable placement model | Hundreds of thousands of SKUs, several storage types, high turnover | Requires movement history; when the assortment changes, some decisions fall into a blind spot |
| Hybrid: model and rules together | Strict constraints on product adjacency and temperature zones | Requires forecast quality control and a regulation for falling back to rules |
The boundary between approaches runs along the volume of data on actual movements: while there is little of it, a trainable model repeats static rules, only less predictably — it has nothing to learn from.
A sustainable effect is provided by reinforcement learning on top of a warehouse simulator, once history by locations and SKUs has been accumulated.
In practice, this is more often a hybrid: the model proposes a plan, storage-type constraints filter it, and a rollback path remains without stopping operations.
How we implement the placement optimization system
Slotting begins not with an algorithm, but with an analysis of how the warehouse currently operates. First, an audit of data and established placement rules — otherwise the model optimizes a beautiful but incorrect picture, and the layout does not take hold.
Work order
- Audit of product placement: check the completeness of address storage, dimensions, shelf life, and movement history. Find items without a location and places where the actual state diverges from records.
- Storage zone scheme: rows, levels, aisles, picking zones, temperature and compatibility constraints. The scheme becomes the framework within which optimization operates.
- Historical shipments: remove one-off spikes and returns, label "item — handling frequency" pairs. This builds the demand profile by zone.
- Rule setup: weights of criteria — handling frequency, joint shipment, dimensions, shelf life, seasonality. A rule that a warehouse worker will bypass is filtered out at this step.
- Model validation on history: how many lines are picked per pass, how the picker's path is shortened. If the result is not confirmed, we adjust the rules rather than "further train" the model.
- Layout implementation: start with one zone, run control days, compare plan with actual. We connect the rest of the warehouse after a sustainable result.
- Handover and training: recalculation regulation, shift instructions, zone scheme, personnel training, and a person responsible for updating.
After handover, the layout lives in the regulation: recalculation is triggered by triggers — season, assortment change, new zone.
What we hand over to the customer after system implementation
We hand over not a report on work performed, but a working kit according to which the warehouse continues to operate without us.
The key principle is that recalculation rules remain reproducible: if after six months the product flow or rack topology changes, the layout is rebuilt using the same logic, not anew "by eye."
- Storage zone scheme — breakdown by turnover and dimensions, capacity of each zone in pallet places, adjacency constraints: temperature, humidity, hazard class, joint storage.
- Product group cards — weight, dimensions, packaging multiplicity, seasonality, and the attributes by which the system assigns a SKU to a zone.
- Replenishment rules — minimum stock threshold in the picking zone, batch size, priorities when space is scarce, behavior during a supply failure.
- Layout recalculation regulation — frequency of full and partial recalculation, list of input data, procedure for approving movements, and the window in which they are performed.
- Rollback regulation — how to return the previous layout if control metrics decline, and who makes that decision.
- System documentation — description of the distribution logic, constraints, decision log, and exports by which the warehouse can calculate metrics itself.
- Warehouse personnel training — separate sessions for warehouse workers, accounting system operators, and supervisors: what to do in case of a discrepancy, whom to call, where to look for the cause.
Separately, we hand over a responsibility matrix and a set of metrics that show the layout is working: line picking time, share of empty trips, and fill rate of picking zones.
Case: reduced order picking time after implementation
A warehouse of an electronics distributor with tens of thousands of SKUs: the layout had been based on the alphabet and warehouse worker memory for years.
A picker covered 12–14 cells per order line, and 8% of orders returned to the additional picking zone — the item was listed in a cell but physically lay in another. Each return meant a repeated pass and a disrupted shipping wave rhythm.
What was changed
We analyzed picking history over several seasons and built a joint demand matrix: which SKUs ship in the same order, how the profile shifts by day of week and during sales.
The algorithm rebuilt the warehouse scheme: high-turnover positions were placed in the active picking zone near the transport route, linked pairs in neighboring cells, and rare items on upper levels.
Recommendations go into 1C:WMS as ready-made movement tasks, and the warehouse worker confirms them in the familiar interface.
Result
Layout recalculation became event-driven rather than once a year: new SKUs, a seasonal shift, or a drop in the hit rate trigger recalculation; edits fit into one night cycle and do not stop shipping.
Picker steps per order line — 6–7 instead of 12–14, the share of orders picked without returns to the zone grew from 91% to 98.5%. Wave picking time decreased by about a third, and new employees reach the norm faster: the route is suggested by the system, not by a mentor.
What to do if storage zones are already assigned to products
Assigned storage zones are not a ban on dynamics, but a boundary condition of the task.
The layout is changed not by a one-time order across the entire warehouse, but in iterations: first, what cannot be touched is fixed, then we look for what can be rearranged without stopping receiving and shipping.
We start by taking the current assignment matrix: SKU, cell, picking frequency, weight, dimensions, temperature regime, joint storage rules.
Then we calculate the target layout by handling frequency and overlay it on warehouse constraints — rack height, aisle width, zones with special conditions, items that cannot be separated.
The first step is a pilot on one zone: take an area with dense picking flow and homogeneous goods, leave some cells assigned, and give some over to dynamic placement.
The metrics are simple and verifiable — line picking time, share of empty transitions between cells, number of rearrangements per shift.
Step-by-step recalculation expands zone by zone, and each subsequent iteration takes into account the rearrangements already made — picker routes are not rewritten entirely.
The previous assignment matrix is preserved as a return point: if metrics decline, the configuration is rolled back in a specific area without affecting the rest of the warehouse.
Warehouse management system provides not an order, but a proposal for rearrangement: each cell has a rationale for why it is beneficial to move it. The shift supervisor approves the decision, so familiar routes are broken only where it pays off in picking time.
Common technical doubts before starting a project
Engineering doubts usually concern not the model itself, but its integration with warehouse reality: where to get data, how to account for seasonality, what to do with multiple storage zones. Below are the things we most often discuss at the assessment stage.
Where do you get data for calculations?
The basis is an export from the accounting system: SKUs, dimensions, weight, product groups, and movement history over several cycles. For new positions without history, we substitute reference attributes of the category so that placement works from the first wave.
How is product seasonality taken into account?
We take not average demand, but a profile by weeks: peak, decline, rotation speed of new positions. For seasonal groups, we reserve capacity in the fast picking zone in advance, and after the decline we return the space to fast-moving assortment.
We have several storage zones — does this complicate the project?
Each zone is described by its own rules: temperature regime, height constraints, equipment accessibility, packaging multiplicity. The algorithm distributes goods across zones as separate resources, not as a single rack.
How do you check the quality of the solution?
We calculate metrics on historical data and compare them with the actual placement: picker path, number of visits to one cell, share of fast-moving positions in the fast access zone. Then — launch in one zone and comparison of forecast with actual.
How to deal with personnel resistance?
A warehouse worker sees the placement result, not the calculation logic, so in the working interface we show the reason for the move: demand increased, the position went to the fast picking zone.
Manual correction remains available, and the system takes it into account in subsequent recalculations.
Let's discuss the task: what we will clarify before starting work
To come back with an estimate and work plan, rather than general considerations, we need a short analysis of your warehouse: it is the inputs that determine both the placement model and the labor costs for implementing it.
Some data is usually already maintained in accounting systems; some is collected in one walkthrough together with a technologist.
- Product flow — SKUs, turnover over several months, dimensions, weight, and storage conditions. This is used to build the classification and rules for product compatibility in a cell.
- Warehouse topology — scheme of zones, racks, and aisles, receiving and shipping points, height and tier load constraints. Without this, optimization runs into physics.
- Data sources — which systems maintain stock balances, what is exported automatically, in what form tasks and movement history arrive. We look at where data is missing and what needs to be closed before start.
- Current placement rules — how decisions are made now: based on warehouse worker experience, assignment zones, or classification. This is the basis for before-and-after comparison.
- Priorities and constraints — what is more important: picking speed, stock accuracy, volume utilization, or reducing the number of movements, and which rules cannot be changed.
- Pilot metrics — share of orders in the experiment, observation period, and target indicators by which we recognize the result.
Describe the task in free form — we will come back with an estimate and work plan. If some input data is missing, we will advise how to collect it and what can realistically be achieved at the first stage.