The forecast page shows an accuracy figure for each site, and it refuses to forecast a site it does not have enough history for. Both are deliberate.
What "typically within N%" means
Paddl measures the forecast against weeks it already knows the answer to. It takes up to six fortnights from the last twelve weeks, forecasts each one as if it had not happened yet using only the history before it, and compares. The page says how many weeks that covered. The typical size of the miss, as a share of takings, is the figure on the page: "Typically within ±12%" means that on a normal day the forecast is within about twelve per cent of what was taken.
It is per site, and it changes as the site does. Until there are enough weeks to measure it, the page says Not yet measured rather than showing a number that would not mean anything.
What makes it better
More history. Thirteen months from the till on connection is a good start; a second year of the same season is better.
A stable pattern. A site that does roughly the same thing each week forecasts well.
Closure periods entered for the days you were shut, so a closed day is not read as a dead one.
What makes it worse
A private booking, a festival in town, a new menu or a competitor opening across the road. None of these are in the data until they have happened a few times.
A wrong figure entered by hand, which sits in the history like a real day.
A till that was offline for long stretches, so days settled late or not at all.
The "In short" summary lists the cautions it can see. It cannot see the ones it cannot see.
When there is no forecast
The page says | What to do |
|---|---|
This site needs eight weeks of daily takings | Wait, or enter or import history. The page shows how many of the 56 days are recorded |
One weekday has only a few days of takings so far | A day of the week has too little to go on. A few more weeks fixes it |
No takings have been recorded for three weeks | Check the till connection, or enter the missing days |
This site is inactive | Reactivate it under Settings > Locations if it is trading |
Every day in the next fortnight is marked closed | Nothing to forecast. It resumes when the closure ends |
Takings are recorded in more than one currency | Check the Square location's currency and any imported rows |
Why it refuses rather than guesses
A confident number with nothing behind it would be used, because numbers on a page get used. Then the rota would be built against it. Saying no, and saying why, is the more useful answer until the site has the history.
Ask Paddy follows the same rule. Ask how busy Saturday will be at a site with no forecast and it tells you how many days of takings the site has and how many it needs.