How PartsOS Forecast Decides Which Parts a Planner Actually Needs to Review

Alongside that: how far the forecast diverges from past consumption, how many cycles current stock still covers, how long the ERP will keep buying against this number before the next review, and how much of it rests on assumptions rather than real customer data.

Each of these answers a different question:

How big a change is the forecast? A pump that's sold 10 units a month for months, now forecast at 11, that's not news. The same pump forecast at 25 is a big jump upward and meaningful in euros. A jump to 4 is just as big a change, just in the other direction, and just as worth checking

How much time is left? Two parts with identical forecasts and identical value: one has eight months of stock and is reviewed every three months; it can wait until the next cycle. The other is nearly empty; it can't.

How long will the ERP keep buying? A seal with a two-week lead time can be corrected again after about three and a half months on a quarterly review cycle. A casting with an eight-month lead time ties up the same forecast for eleven months. If both forecasts turn out 40% too high, the seal gets fixed at the next review; the casting has already been bought against for nearly a year.

How many forecasts need a judgment call, how many are ready to send, how many can wait and when they'll return, when the last upload was and whether the next is due, and a portfolio figure showing how much additional profit the approved forecasts have generated versus the old planning method. In the sidebar, a live count per lane runs alongside, keeping the size of the outstanding work visible from anywhere in the product.

The forecast list can be filtered by lane and status, searched by part number, and sorted by status, divergence, or part code, even across a full cycle. Decided rows don't disappear, they stay visible at the bottom of the list, so a planner working across several days doesn't lose their place. Decisions can be made directly from the row, approve, reject with a reason, snooze, or undo, for a single part or for an entire filtered selection at once, and every decision is logged with who made it and when. For parts that genuinely need review, there's a dedicated page: the history, the release window, sold versus forecast quantity, and how the chosen method scored historically, all on one screen, built for the decision rather than just for looking a part up.

A planner sometimes knows something the history doesn't show yet: a customer is scrapping an entire fleet of machines in September, and the parts for it simply won't be needed anymore. The forecast still says 30 units a month, because nothing has happened in the sales history yet. The planner taps September and enters 0, then does the same for the following months. Those are now the numbers going to the ERP, and when next month's run produces a fresh forecast of 30 again, the zeros stay in place, they don't quietly come back. The second option is for the opposite case: a planner who thinks the whole forecast is roughly 20% too optimistic pulls the entire curve down by that percentage in one action, without changing its shape. This always reads against the original number, so it never compounds, and resetting it to 0% restores exactly what the system originally calculated.

1. Nur so weit veröffentlichen, wie der nächste Review noch korrigieren kann, also das eigene Zeitfenster jedes Teils, oder gleich den vollen Prognosehorizont freigeben. Ein vorsichtiger Planer veröffentlicht nur die nahe Zukunft, liegt eine Prognose falsch, sieht er es beim nächsten Review und kann noch gegensteuern, bevor das ERP zu viel vom Falschen eingekauft hat.

2. Ein Planer, der den Zahlen vertraut, gibt gleich das ganze Jahr frei, dann hat das ERP das vollständige Bild und niemand muss jedes Quartal erneut ran. Die Prognose selbst ist in beiden Fällen identisch, der Unterschied ist nur, wie viel davon das ERP schon jetzt umsetzen darf.