Enterprise Intelligence Group
Retail demand forecasting: five checks before replacing your planning system
Before replacing retail planning software, inspect demand, overrides, replenishment rules, supply constraints and shelf execution. Use EIG's five-check review.
An empty shelf is not a diagnosis of a bad forecasting engine.
It is an outcome. The cause might be an underestimated promotion, an incorrect pack size, an unrecorded supplier restriction, an order that missed its delivery window or stock that reached the store but not the shelf.
Those problems need different responses. Replacing the forecasting platform before distinguishing them can move the same operating weaknesses into a new system.
For retail, CPG and distribution leaders, the useful starting point is a traceable decision: what was expected, what changed, what was ordered, what was available and who acted on the difference. These five checks create that trace before a team commits to more software or another layer of AI.
Understand retail demand forecasting and replenishment
1. Are you forecasting demand, or simply repeating recorded sales?
Recorded sales tell you what customers bought. They do not necessarily show what customers wanted when a product was unavailable.
Take a small sample of high-consequence item/location exceptions. Compare sales with availability, ranging dates, promotion timing and any substitution evidence. Check whether the product was genuinely available for the period used to judge its performance.
A low-selling line might have weak demand. It might also have had poor availability or incomplete execution. A neighboring product's uplift might represent incremental category demand, transferred demand or a mixture. Keep those explanations separate until the evidence supports one.
For CPG teams, distinguish consumer sell-through from retailer orders. Order patterns can reflect replenishment cycles, forward buying or network inventory, not only shopper demand.
Useful output: a short register of demand assumptions, evidence gaps and the decisions they affect. Do not label an estimated lost sale as an observed transaction.
2. Can you explain the last important forecast override?
An override is not automatically a mistake. An experienced planner may know that a local event, store reset or customer commitment is missing from the available inputs.
The problem is an intervention whose reasoning disappears. Record the original forecast, the adjustment, the reason, the evidence available at the time, the approver and when the intervention should be reviewed or expire.
Then compare the result using the same product, location, time horizon and metric. Avoid declaring an override successful because aggregate accuracy improved while service for a strategically important group deteriorated. Equally, do not claim a counterfactual outcome as fact: comparison with a retained baseline is informative, but other conditions may have changed.
Useful output: an override log that allows the team to learn which interventions deserve repetition, which need different evidence and which should stop.
3. Do ordering rules reflect the physical operation?
A demand forecast is not an order. Replenishment must account for stock, open orders, delivery schedules, lead times, order multiples and service requirements.
Inspect those inputs against the real operation. Is lead time measured consistently? Does the case-pack conversion match the supplier? Can the store receive the quantity on the planned day? Has a minimum display requirement outlived the display?
These are not capabilities that enterprise software necessarily lacks. For example, RELEX's published F&R reference describes replenishment inputs and constraint handling. The consulting question is whether the relevant capability is configured, supplied with reliable information and used appropriately in this business. RELEX F&R reference
Useful output: a parameter and constraint review with accountable owners, not an instruction to increase every safety-stock setting.
4. Who resolves a conflict between the plan and the constraint?
Consider an illustrative promotion. The commercial plan expects additional demand, but available supply cannot support the original plan in every store. One cluster also has limited backroom capacity during a reset.
Forecasting, replenishment, category and store teams can each make a reasonable local adjustment. That does not establish which combined action best serves the business.
Write down the objective, the options and the person authorized to decide. Options could include reallocating available supply, changing the promotional scope or sequencing a store change. Each has consequences for availability, margin, inventory and execution. The correct choice depends on the evidence and agreed priorities; it cannot be inferred from a departmental metric alone.
An AI agent can prepare alternatives and surface a missing input. It should not silently invent supplier capacity or rewrite the commercial objective to make its recommendation appear feasible.
Useful output: one decision record joining objective, constraint, trade-off, authority and the selected action. Connected systems help; explicit decision ownership makes the connection useful.
5. Does the feedback reach the next planning decision?
Follow the decision beyond the approved plan. Was the order received? Was the shelf change implemented? Was the product available when the promotion ran? Did the team review inventory and waste as well as sales?
Keep measure definitions and horizons consistent. Forecast-error measures are useful, but choice of aggregation and metric affects interpretation. RELEX's measurement guide discusses these distinctions; the relevant measure must fit the decision being assessed. Measuring forecast accuracy
For each material exception, name the owner of the feedback. A review that ends with "the system was wrong" does not tell the next planner which input, rule or action should change.
Useful output: a small improvement backlog with an owner, due date, test and outcome measure for every accepted change.
A practical first review
Choose one category, one location group and a manageable set of consequential exceptions. Include planning, merchandising, supply and store or customer-facing expertise as the workflow requires. Use the same five checks throughout; do not turn the exercise into an unfocused audit of every system.
Classify each finding as data, configuration, process, execution, capability or unresolved evidence. The next step may be a configuration change, a data correction, clearer ownership, a different operating practice, an AI-assisted workflow or a genuinely new platform requirement.
That is a stronger investment discussion than starting with "we need better AI."
Use the one-page Retail Planning Five-Check Review to prepare the discussion. It is a working aid, not a benchmark or certification.
Retail Planning Five-Check Review
Find the decision gap before replacing the system
Enterprise Intelligence Group | Working aid, not a maturity score or benchmark
Scope: category / location group / planning horizon / review owner / review date.
| Check | Evidence to bring | Decision to record |
|---|---|---|
| 1. Demand | Sales, availability, ranging and promotion timing | What is observed demand evidence, and what remains an estimate? |
| 2. Overrides | Original and revised forecast, reason, approver, expiry | Was the intervention justified, and what will test its value? |
| 3. Ordering rules | Lead times, pack sizes, delivery calendar, stock and open orders | Which input or constraint needs correction, and who owns it? |
| 4. Trade-offs | Supply/capacity limits and commercial priorities | Who decides when margin, availability and working capital conflict? |
| 5. Feedback | Receipts, shelf/customer execution, service, inventory and waste | What changes next, who owns it and when will it be checked? |
Illustrative finding: a promoted line runs short in one store group. The team confirms a delivery restriction was missing from the planning decision. That does not prove the forecasting model failed. Record the evidence, correct the constraint, agree the next test and review availability alongside inventory.
Record one next action: finding / evidence or unresolved assumption / decision owner / action / due date / success measure.
Want an experienced partner to work through the evidence with your team? EIG's Forecasting & Replenishment Performance Review starts with a defined workflow and produces an evidence-led improvement backlog. Discuss your planning challenge.
The promotion example is illustrative, not a client case or a claimed EIG result. Platform references describe published capabilities, not EIG partnership or endorsement.
Discuss a planning performance review
Discuss a supply chain planning and replenishment review
The EIG Library · Insights · AI Maturity Assessment · Contact