In this article
A distributor sells fewer cartons this week. Should it order less next week? If the shelves were empty for three days, the sales report may be telling the purchasing team how much stock was available to sell. Treating that figure as demand can lead to a smaller order and another stockout.
Make the records tell the same story
Choose the demand signal that fits the decision. Customer order dates can help estimate when buyers wanted stock, while dispatch dates show when the warehouse fulfilled those orders. Keep both. Link partial shipments to the original order so a delivery split does not create extra demand. Record returns, cancellations and stock transfers separately, with reasons where available.
Build the history at a consistent product, location and time level. Map old product codes to their replacements where they represent the same item, convert cartons and individual units to a common measure and check for duplicated orders. Mark missing records separately from genuine zero-sales periods. If there was a stockout, use available order, backorder or enquiry records to estimate unmet demand where possible. Flag any estimate and preserve the original record.
Separate repeatable seasonality from one-off events
Use the commercial calendar to interpret peaks and dips. A recurring school-opening surge can inform next season’s plan, while a one-off bulk order belongs in a separate event record. Promotions can recur, but their timing and expected effect need to be built into the forecast. Compare equivalent trading weeks when holidays shift and account for changes in opening days.
Investigate unusually high or low values before changing them. Correct a duplicate, retain a genuine seasonal peak and tag an unusual event so planners can decide when it belongs in future demand. Keep a record of what changed and why.
A distributor with three misleading weeks
Consider an illustrative stationery distributor. It normally sells about 100 cartons a week. One week it ships 40 cartons after running out of stock, while confirmed customer orders show demand for 90. Another week it sells 180 cartons during a planned back-to-school promotion. A third week appears to show 200 cartons, but a duplicate import doubled the true figure of 100.
A simple average of the three raw figures suggests 140 cartons per week. Correct the duplicate, use the 90 cartons of confirmed orders as the observed demand signal for the constrained week and tag the promotion separately. The two ordinary weeks then average 95 cartons. That is an illustrative baseline for a normal week, based on limited history. For a future promotion, use comparable events and planned activity to estimate an uplift. Any customers who gave up without ordering remain outside the recorded 90.
Measure whether the forecast is getting more useful
Stockout exposure. Divide out-of-stock trading days by tracked trading days for each product and location. Review high-exposure items before training or scoring a forecast on sales history.
Data exception rate. Divide records with unresolved issues by records checked. Group exceptions by cause, such as duplicate orders, missing dates or invalid unit conversions.
Forecast error and bias. Compare forecasts saved before the period with actual demand where it can be observed. Divide total absolute error by total actual demand for weighted absolute percentage error. Use signed forecast minus actual differences over the same total to identify persistent overforecasting or underforecasting. Report unit errors when total actual demand is zero.
Order fill rate. Divide units filled at the first fulfilment opportunity by units ordered. Review it alongside forecast error to see whether changes are improving availability.
If unreliable history is leading to stockouts or excess inventory, enquire with SERPA for a tailored quote to review your sales records and improve the data behind your replenishment decisions.