Demand forecasting
Read predicted consumption, confidence, and forecast accuracy without confusing estimates with orders.
Demand forecast in Inventory Tracking estimates future item usage from the previous 84 days of recorded distributions. Use it alongside current stock, commitments, and supplier information when reviewing replenishment.
Open a forecast
- Open Reports and choose Demand forecast.
- Choose the Forecast horizon: 7 days, 14 days, or 30 days.
- Check the As of date and the number of Items forecasted.
- Review Top predicted demand, then the individual item rows.
- Click an item row to compare its recent actual usage with the daily forecast.
- Export the returned rows as CSV when you need to discuss or work with the estimates outside Matte.
What history the forecast uses
The forecast uses the 84 complete days before the workspace's current date. It looks at distribution movements, which represent stock handed out or shipped. The workspace timezone determines the day boundaries.
Receipts, transfers, and count corrections are not demand by themselves. If staff have been reducing quantities through corrections instead of recording distributions, the forecast may not have the consumption history you expect.
An item needs at least some positive distribution activity in the window to appear. The forecast does not create a prediction for every catalog item merely because that item has stock or a reorder point.
Days without distributions are part of the series. This makes consistent recording important: a quiet day and a day when staff forgot to enter usage both look like no recorded demand.
How the estimate is made
Matte fits a trend to the recent daily usage and adjusts it for the pattern of weekdays. This lets an item reflect both a general increase or decrease and a recurring weekly rhythm.
The resulting daily predictions are kept nonnegative and added together for the selected horizon. Predicted quantity is shown in the item's unit; predicted value uses current item prices when available.
This is a statistical calculation from your recorded movements. It is not a supplier forecast, an external market feed, or a language model making a prediction about your business.
The same forecast horizon can be useful for different decisions. A seven-day view can help a weekly review, while a longer horizon can help you compare an upcoming requirement with stock already on hand. The app does not automatically turn that comparison into a purchase order.
Read confidence and trend
Confidence describes the amount of usage history supporting the estimate. It is not a promise that the prediction will be close to the eventual result, nor is it a statistical probability interval.
Seasonal events, closures, unusually large orders, changes in how staff record stock, and new products may need judgment beyond the history shown. Use the expanded item chart to see whether the recent pattern is representative before acting on the total.
- Low confidence. The item has fewer than 10 days with positive usage in the 84-day window.
- Medium confidence. There are at least 10 positive-usage days, but fewer than half of the days in the window.
- High confidence. At least half of the window's days have positive usage.
- Trend. The trend indicator shows whether the fitted direction is up, down, or flat.
Understand Accuracy (MAPE)
The table can show Accuracy (MAPE), a comparison between saved forecasts and the actual usage that becomes available afterward. The app stores forecast snapshots and can score them when the relevant future days have passed.
A blank accuracy value usually means there is no eligible scored history yet. Opening the report today does not instantly create evidence of how accurate tomorrow's prediction will be.
MAPE is an error percentage, so a smaller value means a closer match on the observations it measures. Treat it as historical evidence about previous predictions, alongside the visible usage pattern, rather than as a guarantee about the current forecast.
Use the forecast in replenishment decisions
Compare the predicted quantity with physical stock, stock already committed to sales orders, purchase quantities already on order, and your supplier's practical lead time. Avoid adding expected inbound stock to on-hand totals before it is received.
Reorder prompts and suggested supplier order quantities currently use reorder points, par levels, and ordering rules. Demand forecast does not automatically adjust those settings, raise par levels, or generate forecast-driven orders.
The report defaults to a limited set of returned items, so the rows and CSV should not be read as a complete item-by-item forecast export of an arbitrarily large catalog. Reports access and reports.view permission govern this surface.
If the report says Not enough usage history yet, first confirm that distributions are being recorded. Adding a price can improve value presentation, but it cannot substitute for real consumption history.