When and how much to order: practical inventory replenishment rules

1. Decide which decision the system will make

Separate detecting a need, proposing a quantity and authorizing a purchase. They do not have to be automated together. Initially, the system can generate daily recommendations while a buyer checks supplier agreements and operational context. This exposes incorrect rules before they create financial commitments. Record how a recommendation becomes an approved order, including who can change it and why.

Set boundaries: one warehouse, a product group, a stock unit and an owner. Do not mix store replenishment from a central warehouse with external purchasing if their lead times and constraints differ. Define success: fewer emergency orders, availability of an agreed assortment or less stagnant stock. These objectives can conflict. “Minimum inventory and no shortages” is not a complete specification without assumptions. Agree which trade-offs are acceptable and who resolves conflicts between availability, budget and storage capacity. The first implementation should make those choices visible rather than burying them in unexplained system parameters.

2. Align on-hand stock, availability and inventory position

Physical stock describes units present, but some may be reserved, quarantined, damaged or awaiting inspection. Availability answers whether a new order can be fulfilled now. Replenishment also needs confirmed inbound supply and outstanding demand. Systems use different labels, so inspect field meaning rather than trusting a familiar report heading. Document whether each field includes stock at other locations or goods that cannot currently be sold.

In a simplified model, inventory position can be usable on-hand stock plus confirmed on-order supply minus unfulfilled commitments not already accounted for elsewhere. Do not subtract the same demand twice. If reservations already reduce the availability field used as your starting point, subtracting them again overstates the need. Write a formula for your actual fields and trace one order from reservation to dispatch. Likewise, an unconfirmed purchase draft is not guaranteed inbound supply. Show the components beside each recommendation so a buyer can explain the result and identify a stale or incorrectly classified input.

3. Measure time until stock is usable

Replenishment lead time is not merely the carrier’s promise. The decision may require the interval from purchase authorization until goods have been received, checked and released for use. Two days of approval and one day of receiving cannot be ignored because transport is fast. Specify the starting event clearly so records with different definitions do not enter the same calculation unnoticed.

Collect actual durations for comparable deliveries, inspect variation and investigate delay causes. Different routes or supply arrangements may not belong in one average. Mark partial receipts: the first box arriving does not mean the full order is available. For irregularly supplied items, a short history cannot establish a reliable future estimate. Use explicit review and delay scenarios rather than a confident figure with two decimal places. Check whether recorded receipt time represents physical arrival or a later administrative posting. If the timestamps do not correspond to the decision, improve the recording process before optimizing the replenishment formula around them.

4. Test a simple reorder point with clear assumptions

This is a fictional teaching example, not a business-specific recommendation. Demand is stable at an average of 6 units per day, replenishment takes 5 days and an agreed fixed buffer is 12 units. A simplified reorder point is 6 × 5 + 12 = 42 units. Under continuous review, reaching or falling below that defined inventory-position level generates a replenishment signal. It does not determine the purchase quantity by itself.

The buffer of 12 is an assumption, not a statistical service-level calculation. The example does not account for seasonality, volatile demand, supplier constraints or delays. If stock is reviewed only weekly, the review interval also matters; the continuous-review model no longer describes the actual process. Do not copy 42 into an ERP without testing the assumptions. Replay several historical decisions and inspect when the signal would have appeared and what stock would genuinely have been available. Include delayed receipts and outstanding commitments so the test does not benefit from information that was unknown at the time.

5. Separate the trigger from the order quantity

Crossing a threshold does not mean “buy up to the threshold” or “buy the same amount again”. Quantity may depend on a target level, review interval, minimum order, pack size and budget. If calculated need is 17 units and the supplier ships packs of 6, rounding to 18 should be visible. Correct arithmetic is still not purchase authorization. The buyer needs to see why the additional unit is proposed.

Check minimum order value, shelf life, warehouse capacity, substitutes and planned transfers. Show the initial need and the reason for each adjustment. Otherwise nobody can explain why a recommendation suddenly doubled. A new product may have no demand history; its initial stock is a separate assortment decision. A discontinued product may prohibit ordinary replenishment even when stock is low. Store these states explicitly rather than relying on a buyer’s memory. Also define how a cancelled purchase or supplier rejection reopens the need so the next review neither ignores the shortage nor creates duplicate replacement orders.

6. Segment by decisions, not only ABC labels

ABC classification by annual consumption value can focus attention, but it does not automatically describe margin, criticality or shortage risk. An inexpensive component may stop production, while customers may accept waiting for an expensive slow-moving product. State the classification measure and the actual rule that changes for each group. A letter that changes nothing in review frequency or treatment adds little operational value.

Consider demand stability, substitutability, lead time and lifecycle. In apparel, model, colour and size may behave differently: overall model stock does not ensure availability of the requested size. Avoid drawing firm conclusions from a handful of sales after excessive segmentation. Rare items may need manual decisions or purchase against confirmed demand. The worksheet below illustrates three control approaches, not a universal store matrix. Its purpose is to decide who reviews a recommendation and when. Revisit classifications when the assortment changes; last year’s high-value item is not necessarily this season’s replenishment priority.

7. Do not treat sales as the complete demand signal

When stock is unavailable, zero sales may reflect inability to buy rather than lack of interest. Mechanically including those periods in an average can reduce the forecast and prolong the shortage. Mark unavailable periods and inspect reliable records of refused requests, preorders or substitutions separately. Do not invent lost demand using an arbitrary multiplier. If the evidence is weak, make the uncertainty part of the purchasing decision.

Promotions, seasonality and large one-off orders need context too. Do not automatically delete them as outliers: they may recur according to a known plan. For an upcoming promotion, agree on a scenario, horizon and post-event review owner. Separate returns from new demand; returned stock may not immediately be saleable. Preserve reasons for manual overrides so their quality can later be evaluated. If a buyer repeatedly changes the same group of recommendations, revisit the data or model instead of adding another hidden adjustment. The override history is useful evidence about where the current rule fails.

8. Apply pull and kanban where the flow supports them

Lean Enterprise Institute describes pull as replenishment driven by downstream need and kanban as a signal within that flow. The practical implication is a clear trigger, defined quantity and response owner. A board displaying task statuses does not by itself create an inventory replenishment system. For recurring consumption, investigate whether a simple visible signal can work without elaborate forecasting.

Before launch, agree what happens when a signal is lost, packaging changes, supply is delayed or consumption rises sharply. Do not apply one quantity to every item or reduce buffers solely to improve an inventory figure. Flow stability and response capability matter. Start with a bounded area and observe several cycles from consumption to restored stock. A timely signal is ineffective if nobody has authority to order. The ERP can record the signal but cannot replace agreement about the response. Include an escalation route for abnormal conditions so the team does not continue following an unsuitable rule simply because the system generated it.

9. Launch recommendations with an exception log

Initially compare recommendations with the responsible person’s decisions without automatically sending orders to suppliers. Record disagreements: inaccurate stock, late inbound supply, promotion, budget restriction or incorrect rule. Monitor not only average inventory but also shortages, urgent purchases, stagnant items and manual review time. Improving one measure can worsen another. Use comparable periods and note changes in demand or assortment when interpreting results.

Once the rules stabilize, define automatic-action boundaries: eligible products, amounts, suppliers and stop conditions. Parameter changes need an author and explanation. If inputs are stale or supply is unconfirmed, expose uncertainty rather than pretending to be precise. Review settings after changes in demand or lead time. A working replenishment process is not a one-time configuration of minimum stock levels; it is a controlled cycle of observation, decision and verification. Retain a safe manual route when automation cannot establish the required facts.

• All stock components have agreed meanings.
• Reservations do not subtract demand twice.
• Lead time ends when goods are genuinely usable.
• The reorder trigger and purchase quantity are separate.
• Exceptions and purchasing authority are explicit.