Demand, Replenishment and Waste Forecast

Define the replenishment decision first

A forecast can support purchasing, distribution, production, labour or financial planning. These decisions occur at different levels and horizons. A monthly category forecast cannot directly determine tomorrow's order for one product in one store.

Marketways defines the product-location unit, order cadence, lead-time exposure and action before choosing a model. We also state which outcomes management wants the replenishment policy to protect. This definition ensures that the forecast arrives with the detail and timing required for a real order.

Reconstruct demand hidden by stockouts

Sales equal demand only while customers can buy the product. Once stock reaches zero, the sales record becomes a lower bound: it shows what the store sold, not how many customers still wanted to buy. Repeated stockouts can therefore train a model to forecast the shortage created by the previous policy.

Marketways aligns point-of-sale records with inventory position, stockout periods, deliveries and product availability. We estimate censored demand under explicit assumptions and preserve uncertainty where lost sales cannot be observed. The replenishment model can then learn from customer demand rather than from availability alone.

Account for substitution and displaced demand

A customer who cannot buy the preferred product may choose another pack, brand, channel or location. The original item records a lost sale while the substitute records demand that its normal pattern did not generate.

Marketways examines product relationships and transaction context where the evidence permits. We distinguish transferred demand from category loss and delayed purchase. This step prevents the forecast from underestimating one item while overestimating the product that temporarily absorbed its demand.

Represent price, promotion and events as causes of changing demand

Promotions, price changes, holidays, weather, local events and assortment changes can alter demand. However, a correlation with higher sales does not establish that every future event will create the same lift. The effect may depend on stock, display, competing offers, timing and customer response.

Marketways defines each event consistently and estimates its effect under relevant conditions. We separate recurring calendar structure from exceptional interventions and retain uncertainty for events with limited precedent. The model can then produce a baseline forecast and a documented adjustment rather than embedding unexplained manual uplift.

Match the forecast horizon with lead time and review cadence

An order placed today must cover demand until the next replenishment can arrive. The relevant horizon therefore depends on supplier lead time, order frequency, delivery schedule and any delay the policy must absorb.

Marketways models demand over the complete protection period. We use the distribution of lead time rather than a convenient average when delays are material. The forecast and inventory policy then address the same period of exposure.

Forecast at the level where the decision occurs

Retail demand forms a hierarchy. Company demand contains regions, categories, products and locations, while management may also plan across channels and time periods. Independent forecasts at each level can contradict one another.

Marketways evaluates which signals are reliable at each level and reconciles the forecasts so that the totals and detailed decisions remain coherent. We do not force weak product-location histories to stand alone when related products or locations provide useful information. The method preserves local differences while borrowing strength where the evidence supports it.

Use probability distributions rather than one expected value

Two products can have the same expected demand but different uncertainty. Ordering only the expected value gives management no basis for choosing the inventory protection required for a target service level.

Marketways estimates prediction intervals or demand distributions and checks their calibration. We test whether actual demand falls within the stated ranges at the promised frequency. The replenishment policy can then convert uncertainty into safety stock, order quantity or an escalation rule.

Give forecast errors their correct commercial cost

An under-forecast for a staple can cause lost sales and erode trust. The same numerical over-forecast for a short-life product can create waste, while a long-life item may mainly consume space and capital. Seasonal goods can also lose value sharply after a fixed date.

Marketways represents shortage, spoilage, markdown, holding and handling costs by product and location where material. We evaluate forecasts through the resulting decision cost as well as statistical error. A model is useful when it improves the order outcome, even if another model has a slightly lower average error on a metric that ignores business consequences.

Include shelf life and inventory age

Total on-hand stock does not show how much inventory will remain saleable through the forecast horizon. Several units may expire before demand occurs, and newer deliveries can conceal older stock that the operation fails to rotate.

Marketways represents age buckets, remaining life and relevant rotation rules. The order decision considers which stock can satisfy future demand and which stock is likely to require markdown, transfer or disposal. This distinction connects replenishment with waste prevention rather than treating waste as an outcome observed later.

Include supplier and ordering constraints

Minimum order quantities, case packs, delivery days, supplier capacity and unreliable lead times restrict the quantities a store can receive. Storage, display and labour capacity create further limits after delivery.

Marketways includes these constraints in the optimisation or simulation. We identify whether a poor service outcome comes from forecast error, an infeasible ordering rule or supplier execution. Management can then correct the relevant part of the system instead of asking the forecasting model to compensate for every constraint.

Treat new products and sparse demand separately

New products have no direct history, and slow-moving items produce many zero-sales periods. A standard model may interpret both conditions poorly, especially when similar products have different launch support or lifecycle patterns.

Marketways uses product attributes, analogues, lifecycle assumptions and pooled evidence where appropriate. We update the forecast as actual sales and availability accumulate. The method states the uncertainty rather than manufacturing confidence from a short record.

Coordinate markdown, transfer and disposal decisions

Replenishment is not the only action that affects waste. A store may transfer stock, reduce the price, change the display, stop ordering or remove a product when safety or quality requires it.

Marketways defines these actions, their timing and their effect on future demand and margin. We avoid teaching the forecast that discounted clearance represents ordinary full-price demand. The decision system can then compare prevention, recovery and disposal before product life expires.

Test the policy through simulation and controlled deployment

A forecast may perform well historically while the replenishment policy performs poorly. Back-tests that ignore lost sales, lead-time variation, case packs or waste cannot establish the operating result.

Marketways simulates the complete inventory policy under historical and stressed conditions. We then compare predicted and observed availability, waste, markdown, margin and inventory during a pilot or phased release. The test identifies whether the forecast, policy parameter, supplier or store execution requires correction.

Monitor bias, calibration and operating outcomes

Average accuracy can hide persistent under-forecasting for one region or systematic over-ordering of perishables. A model can also remain statistically stable while supplier lead time or store execution deteriorates.

Marketways monitors forecast bias, interval calibration and decision outcomes by relevant product and location groups. We set review thresholds and record overrides with their reasons. This evidence shows whether the system has changed, the data have weakened or management has learned something the model does not yet represent.

Use Agentic AI to execute an approved replenishment policy

An Agentic AI workflow can retrieve the current forecast, inventory age, supplier constraints and approved event adjustments before preparing an order. It can execute routine orders within limits and refer cases involving unusual demand, low-quality data, capacity conflicts or material waste risk.

Marketways tests stale inventory, missing deliveries, conflicting promotions, duplicate orders, supplier-tool failure and unauthorised overrides. The agent records the inputs, rule, proposed action and final decision. It applies the replenishment policy; it does not invent demand or conceal an exception to meet an availability target.

A fresh-food order illustrates the complete decision

Suppose a supermarket sold out of strawberries during a weekend promotion. The following week's sales data show a low volume on the stockout day and high sales of a substitute berry. Reordering from recorded sales alone would repeat the shortage, while treating every substitute sale as incremental strawberry demand would overstate the opportunity.

Marketways would reconstruct a range for lost demand, estimate promotion and substitution effects, account for the remaining life of incoming stock and simulate feasible case-pack orders. The selected quantity would balance service, waste and margin under stated uncertainty. Later results would show whether the policy improved availability without shifting the problem into markdown or disposal.

How Marketways connects forecasting with replenishment

Marketways combines Sales Forecasting and Demand Planning, Operational Performance Diagnostic, probabilistic forecasting and Operations Research and Optimisation. We reconstruct demand, build coherent forecasts, define the cost of error, optimise feasible orders and test the complete policy.

The result tells management what each location should order, how uncertainty changes that quantity and whether the policy improves availability, waste, margin and working capital.

Related guides and evidence

• Perishable Cold-Chain and Route Optimisation
• Store and Service Execution Audit
• Assortment, Price and Pack Architecture
• AI Workflows for Retail and Consumer Distribution
• Fresh-Food Demand, Cold-Chain and Route Optimisation: case study