Forecast accuracy, measured honestly

A forecast is a promise, and accuracy is the record of how those promises turned out. The application cannot produce this record, because the Working version overwrites the promise every month. Keeping it is one snapshot, and the payoff is a review that starts with the habitual missers.

Note: In Planning, Oracle's module where budgets and forecasts are built, a forecast is a promise about the future. Accuracy is how good those promises have been, measured after the fact. This page is that measurement, which the application does not keep for you.

The whole module is mapped on the Planning index.

RuleSnapshot the forecast at each month end, define the miss measure once in writing, and read accuracy by entity and month.

Nevermeasure accuracy from whatever version survives in the application. The surviving version has been corrected; the promise you made in April has not survived to defend itself.

◆ The example, and the three requirements, the 2.0% worked through, and what the measurement needs.

The worked example: in April, the forecast said June revenue would be 69.8M. June closed at 71.2M. The miss is 1.4M, which is 2.0% of the actual. One number, and it required something the application does not do: keeping April's forecast after May overwrote it.

What accuracy needs	How it exists
Past forecast versions, preserved	A month-end snapshot of the forecast lane into your warehouse, stamped with its as-of month. Inside the application, the Working version moves on and the promise is gone.
The matching actual	The same ledger rows the budget comparison uses, on the same keys.
A defined measure	Miss as a signed amount and as a percentage of actual, by entity and by month. Define it once, in writing, or every meeting redefines it.

What the trend buys: entities that habitually miss in one direction stop hiding inside the group total. If 1010-CL, the Chilean subsidiary, has run optimistic four months straight, that pattern is a row in a table, and next cycle's review starts there.

THE USE CASE, SIMPLIFIED

The problem: Nobody can say whether the June forecast was any good, because April's version of it no longer exists anywhere.

What we build: Every forecast version is snapshotted at month end, so past promises are preserved next to what actually happened.

What you get: Forecast accuracy by month, by entity, trended, and the entities that habitually miss stand out on their own.

Forecast reviews arguing about the past instead of the future?

We snapshot every forecast at month end and land it beside actuals, and accuracy by entity becomes a standing table the review opens with.

Talk to us

Terms on this page

the snapshot: The forecast lane, preserved monthly with its as-of date.
the miss: Actual minus forecast, signed, and as a percent of actual.
habitual missers: Entities biased one direction for months. The table shows them.
2.0%: The sample: 69.8 promised, 71.2 delivered, 1.4 missed.