Snapshots and forecast accuracy

Assumptions change, and actuals accumulate — which makes it easy to lose track of what you actually believed at an earlier point in time. Snapshots solve that: they freeze a forecast so you can come back to it later and see how it held up against reality.

​

Freeze a snapshot

Click Freeze snapshot at the top of the project. Vector saves the project’s assumptions, the full calculated forecast and the date. Your working model isn’t affected. A snapshot takes the last saved version of your assumptions. If the dot next to the project name shows unsaved changes, click Save now first. A snapshot never changes: editing assumptions or recording new actuals later leaves it as it was. Your snapshots are listed in the Snapshots — accuracy of past forecasts section of the project’s dashboard, and the list stays quick to open however many you collect. To delete a snapshot you no longer need, click × on it and confirm. Only that snapshot goes; the project and its actuals stay.

​

Compare a snapshot against reality

In Snapshots — accuracy of past forecasts, pick a snapshot, and choose the metric to check: workspaces or revenue. The chart then shows three things together: the forecast that snapshot implied, your actuals, and your current forecast — which makes it easy to spot a systematic pattern, like a snapshot that was consistently too optimistic. Once the metric has actuals, Vector also says how far reality has been from the snapshot on average, and in which direction — above or below what you expected. With no actuals for that metric yet, no figure is shown.

​

See whether your forecasting improves

Once you have several snapshots and actuals for the months they forecast, Forecast accuracy over time compares each past snapshot with what really happened, and says at the top whether you’re ↓ improving, ↑ worsening or → flat. For each snapshot, in date order:

• Avg miss (MAPE) — the typical size of the miss, in percent. Lower later on means you’re forecasting better.
• Bias — the direction you lean: over means you were optimistic, under pessimistic, or on target.
• months scored — how many months the figures rest on.

Below the table, Miss by forecast horizon groups the misses by how far ahead each prediction was, so you can see how much to discount distant months. Predictability by metric shows which metrics are hardest to predict — the ones whose assumptions need the most scrutiny.

​

How accuracy is scored

A few rules keep these numbers honest:

• Only months where a snapshot’s forecast overlaps with a recorded actual are scored. A snapshot with no overlap yet is marked not yet evaluable — never shown as 0% error or a perfect forecast.
• A month with a zero actual is excluded from the percentage-error figure, since a percentage against zero isn’t meaningful; the remaining months are still scored.
• Every scored figure shows how many months it’s based on, so a number from one overlapping month isn’t presented with the same confidence as one from a year of them.
• Each scored point also carries how far ahead it was forecast, so near-term accuracy isn’t mixed together with long-range accuracy.

Until a snapshot has actuals to compare with, the accuracy view says No accuracy yet and explains what it needs, instead of showing an error or a misleading zero.

​

Where to go next