Demand accuracy#
For: Manager | Admin You’ll need: A plan that includes demand forecasting — Plus or Pro — permission to view reports, and at least one week where a forecast can be compared with recorded values.
This report answers one question: is the forecast worth trusting? It scores every prediction against what actually happened — including against the naive alternative of assuming last week repeats.
Open the report#
In the sidebar, open Reports → Demand accuracy.
The report is gated on demand forecasting being enabled: when it’s off, both the sidebar entry and its card on the Reports landing page disappear.
The subtitle says it plainly: How close last week’s forecast came to what actually happened. Above the table, a headline line sums up the scored period: “Beating the benchmark in {n} of {m} weeks, average MASE {x}.”
What the three numbers mean#
Measure |
What it tells you |
|---|---|
‘wMAPE’ |
Total error as a share of the total actual — the headline percentage. Lower is better. |
‘MASE’ |
Our error divided by the error of simply repeating last week. Below 1 means the forecast is better than that. |
‘Bias’ |
Positive means the forecast runs high and you’d overstaff; negative means it comes up short. |
Read MASE first. A forecast with a respectable wMAPE but a MASE above 1 is losing to copying last week — for that driver, the forecast is not earning its place, and the fix is usually more history, or a marker on days that distorted it.
Bias is the one to act on: a consistently positive bias means you are paying for people you don’t need, a negative one means you are consistently short.
Filter it#
Pick the window from the period dropdown in the toolbar, and use the filter icon beside it to add a ‘Location’ filter (or ‘From’ / ‘To’ for a range the presets don’t cover). Driver is a column, not a filter. ‘Manage columns’, the icon next to the filter icon, chooses what the table shows.
When it’s empty#
‘Nothing to score yet — this fills in once a forecast has a week of actuals to be measured against.’ — the forecast hasn’t had a scored week yet. Keep recording values.
‘No data matches these filters. Try widening the date range.’ — the data exists, your filter excludes it.
What to do with a bad score#
Check whether abnormal days are teaching the forecast the wrong thing — How to mark an exception day.
Check that the driver is measured finely enough for how you staff — How to create a demand driver.
Give it more history. A year lets the forecast compare like with like across seasons.
Verify it worked#
The table lists a row per driver and location with all three measures filled in for the scored period.