Inventory forecast calculation
Forecasting stock starts by measuring how quickly it disappears.
A simple forecast projects current quantity using a consumption rate. The difficult part is not subtraction. It is selecting representative demand, separating real usage from corrections and refusing to invent a date when history is not reliable enough.
01
Separate movements
An outgoing usage movement tells you about consumption. Receiving, an inventory correction or a transfer describes something else. Mixing every quantity change together creates a false demand rate even when the arithmetic itself is flawless.
History therefore needs movement types. Invumi uses consumption movements as the forecasting signal and keeps corrections to explain stock reliability rather than treating every decrease as equivalent customer demand.
02
Choose a window
An average across all history can respond too slowly when activity changes. A recent window reacts faster but becomes more sensitive to unusual weeks. Weighting recent periods or comparing similar periods is one way to balance those weaknesses.
No window knows the future. Promotions, holidays or a process change can break the trend. Forecasts should therefore remain estimates and expose enough context for the user to understand where the number came from.
03
Handle missing data
A new product does not have a useful history yet. Displaying a precise curve anyway replaces missing information with graphics. An explicit insufficient-data state is more useful for a real purchasing decision.
Invumi allows a monthly consumption fallback to be entered. The product can then be projected from a known assumption while clearly distinguishing that source from a rate estimated from observed movements.
04
Move to replenishment
Projected quantity becomes interesting when it meets a threshold. The system can estimate when stock will cross that level and leave time to account for supplier lead time. Forecasting zero is often a date that arrives too late for action.
Reorder point, safety stock and coverage complement the forecast with more deterministic rules. Use them together: the trend describes what may happen, while thresholds describe when your process wants to react.