Not just a forecast — the why behind it

What gets forecast

Revenue, transactions, customers, units and average order value, at horizons from a week to a year — with honest uncertainty bands rather than one falsely precise number.

The explainability layer

Same-weekday baselines, weekend and holiday effects, paydays, month-edges, momentum, recorded promotions and events — ranked in plain English. A forecast you can interrogate is a forecast you can act on.

Guarded inputs

A nightly data-quality scan flags stale feeds, missing days and absurd values before they poison a forecast. When the data is suspect, Maya says so instead of forecasting from garbage.

A number without a range is a guess in a suit

A forecast stated as a single figure invites false confidence. Tomorrow will not be exactly $4,180; it will land in a range, and the width of that range is itself information — a narrow band on a stable weekday line means something quite different from a wide band on a seasonal one.

So forecasts carry confidence ranges, and the drivers are shown in plain English: weekday pattern, payday proximity, holidays, recent momentum, promotions. You can disagree with a driver, which you cannot do with a black box.

When forecasts are wrong

Statistical forecasting fails predictably in three cases: a business with too little history, a genuine one-off like a road closure or a competitor opening, and anything driven by a decision nobody told the system about. A promotion you ran without recording it will look like inexplicable demand.

The honest framing is that a forecast is an input to your judgment rather than a replacement for it — most useful for the routine reordering that consumes attention, least useful exactly when the week is unusual.