Enterprise Intelligence Group
Retail demand forecasting: connect the forecast to planning and replenishment
Understand retail demand forecasting, how it differs from replenishment, and how to evaluate data, forecast performance and the decisions behind inventory.
Retail demand forecasting estimates future demand for a defined product, location and period. Its purpose is to inform a decision: purchasing, replenishment, inventory positioning, promotion preparation or another planning activity.
The forecast is not the order. An order also reflects available stock, incoming supply, lead times, pack sizes, minimum quantities and operating policies. A useful review follows both the demand estimate and the rules that turn it into action.
Sales history is evidence, not a complete description of demand
Observed sales reflect what customers bought under the conditions that existed. Stockouts, distribution changes, price changes and promotions can influence that history. Low sales during an availability problem should not automatically be interpreted as low underlying demand.
For CPG businesses, orders from a retailer or distributor can also reflect their ordering cycle and inventory decisions. They are not necessarily a direct record of consumer purchases. Define whether the planning process is forecasting consumer demand, customer orders or another quantity before evaluating accuracy.
For a new product, there may be no directly comparable history. Document the chosen analogues, launch assumptions and uncertainty rather than presenting a single estimate as established demand.
Define the decision before selecting the method
A forecast for next week’s ordering may need different detail and timing from a forecast supporting a longer-term commercial plan. Specify the product and location level, time bucket, forecast horizon and refresh cycle. Then ask which inputs were genuinely available when each forecast was made.
Methods may use historical patterns, events, causal inputs or combinations of approaches. A more complex method is not automatically more useful. Compare the output against an appropriate baseline and the operational decision it supports, using the same evaluation conditions.
Evaluate more than one accuracy number
Forecast error describes the difference between an estimate and an observed outcome under the chosen definition. Bias examines whether errors systematically lean toward over- or under-forecasting. Their interpretation depends on aggregation, horizon and data quality.
Strong aggregate performance can conceal poor results for important products or locations. Percentage measures can also become difficult to interpret when actual quantities are very small or zero. Agree the treatment of these cases and inspect meaningful segments rather than relying on one headline figure.
Avoid using information from after the forecast date to create an apparently better historical test. Retain the forecast version and the evidence available at the time. Where availability makes observed sales an incomplete measure of demand, document the limitation.
For background on test data and the limitations of percentage-error measures, see Forecasting: Principles and Practice, evaluating forecast accuracy.
Follow the forecast into replenishment
Consider an illustrative situation: demand is estimated reasonably, but a pack-size rule and an outdated lead time produce an unsuitable order. Changing the forecasting model alone would not resolve that problem.
Alternatively, planners may repeatedly override the forecast for a known promotional event that is absent from the inputs. The issue may be event data or workflow ownership. The useful review separates these possibilities before recommending a system change.
| Decision checkpoint | Evidence to inspect | Question for the team |
|---|---|---|
| Forecast purpose | Defined product/location/time/horizon | What decision will use this estimate? |
| History | Sales, availability and distribution changes | Where might observed sales misrepresent demand? |
| Events | Promotions, price changes and new-product assumptions | Which events were known when the forecast was made? |
| Evaluation | Forecast versions, baseline and metric definitions | Are comparisons made on the same basis? |
| Overrides | Reason, owner and subsequent review | Is judgment recorded and evaluated? |
| Ordering | Stock, supply, lead times, packs and policies | Does the order problem originate after the forecast? |
| Execution | Receipts, availability and relevant store exceptions | Was the agreed plan actually implemented? |
| Improvement | Named owner, proposed change and validation | What will be tested before the process is expanded? |
Make improvement a scoped piece of work
Begin with a defined category, location group or planning cycle. EIG can assess the workflow, evidence and assumptions, then propose a prioritized improvement backlog. Relevant outcomes may include forecast performance, availability, inventory, waste and planner workload; measures and responsibilities are agreed for the engagement.
AI-assisted analysis can help organize exceptions and identify patterns. It does not remove the need to validate inputs, assess alternative explanations or retain accountability for material decisions.
For forecast creation, overrides and planning responsibilities, start with a planning performance review. For the handoff into ordering and supplier constraints, discuss a supply chain planning and replenishment review. Software replacement is one possible outcome—not the assumption behind either engagement.
Explore the planning performance review
Explore the supply chain planning review
Discuss your forecasting or replenishment challenge
The EIG Library · Insights · AI Maturity Assessment · Contact