Customer forecasts
What a customer is likely worth, when they are likely to buy again, and how likely they are to have gone — worked out from their own orders.
EmailEvery customer with an order gets a small set of predicted numbers, recomputed every quarter of an hour and shown on their record: their average order value, how long they usually leave between orders, when the next one is due, how overdue it is, the risk of having lost them, and what they would be worth over the next year if they keep going at their rhythm.
There is no machine learning in this, and we would rather say so. They are the ordinary retail arithmetic: what they have spent divided by how many times; the span of their orders divided by the gaps between them; their last order plus that gap; and a risk that rises as the gaps go by — half gone at one cadence late, as good as gone at three. A number a merchant can reproduce on paper gets used; one nobody can explain gets ignored.
The point of them is segments. Under Customers → Segments you can now build a list on “where they are as a customer” (buying, slipping away, lapsed), on risk, on predicted value, on average order, or on how overdue their next order is — and then email it, text it, or put it in an automation. “Slipping away, worth over RM 500, overdue by a week” is the list worth writing to this month.
Two honesties built in. A contact with no orders has no prediction and is left out of every predicted rule rather than counted as a zero. And each prediction carries how much of it to trust: one order is a guess, a few are a rhythm, six or more is a pattern.
Visits count too. For a salon, a clinic or a gym the visit is what a customer buys, so a booking that happened counts like an order: its price (or what was paid online) goes into what they spend, and the time between visits is their rhythm. A shop that also takes bookings gets both added together. The contact page then says “Next visit” for somebody who books rather than buys.