# Sell-out forecast accuracy

Source: https://docs.tightly.io/help/connect/sell-out-engine-accuracy
Reviewed: 2026-10-29

## What it is

These scores apply to the sell-out engine and reported consumer sales. An **Expected orders** account plan is a separate order-intake target: it does not inherit sell-out scores, store coverage or a forecast interval from this engine.

Each run of the sell-out engine is a vintage, the forecast as it stood in one week. A vintage is scored against what the account later reported. A later run never rewrites an earlier score.

## The three measures

- **WMAPE** is the total miss, whichever way, divided by the total sold. 20% means the forecast was off by a fifth of the units. Lower is better.
- **Bias** is the net miss divided by the total sold. The **Accuracy** figure's tip says **forecast runs high** when the forecast ran high, **forecast runs low** when low, and **forecast runs level** when it was even. Bias can be near zero while WMAPE is large, because high and low misses cancel.
- **Hit rate** is the share of forecast cells whose actual sales fell inside the P10 to P90 range. A range that is honest hits about 8 times in 10.

Both WMAPE and bias are blank where nothing was sold, because there is no total to divide by. Hover or focus **Accuracy** on the account's forecast for **N of M doors scored**.

## Not yet scored

A forecast week is scored only once it ended more than 4 weeks ago. An account's report trails the week it describes, and scoring a week whose report has not arrived would read the delay as a miss. So a new forecast shows a dash for accuracy, with one of these reasons in its tip.

| Reason | Meaning |
|---|---|
| Too recent to score | No forecast week is old enough yet |
| Nothing reported | The account sent nothing for the scorable weeks |
| Nothing to compare | Reports exist but no cell could be compared |
| No vintage scored | No earlier run has been scored |
| Scoring not set up | Scoring has not been enabled for this account |

An unreported week is left out and counted. It is never scored as zero. The one exception is an account that lists only the movement it had: a store it was already listing, with no row in a week it did report, counts as zero.

## Reading it well

Look at WMAPE and bias together. High WMAPE with bias near zero means the forecast is noisy but not leaning. High WMAPE with large bias means it leans one way, and a planner can lean against it. Judge a door by its own vintages on [the door page](../connect/the-door-page) before trusting the account figure, which pools them.

## Where you see it

On the account's [sell-out forecast](../connect/forecast-engine-overview) as the **Accuracy** fact, and per vintage on [the door page](../connect/the-door-page).
