AI predictions
What the AI Predictions page forecasts, how often it updates and how to use its estimates without over-trusting them.
Open your dashboardWhat the page predicts
AI Predictions estimates two things from your order history: daily revenue for the next 30 days, and how likely each customer is to stop buying. It has no date controls; it always looks back from today.
Insights refreshes the predictions once a day. Both use straightforward statistics on your own history, so they know nothing about your plans: a sale next week, a new product or the start of Ramadan.
- To refresh now, click Run Forecast for the revenue forecast or Run Churn for churn risk.
- Wait for the blue note to clear. Results appear on their own, usually within 30 seconds.
Revenue Forecast (30 days)
The forecast learns from the last 90 days of daily order value and needs at least 14 days with orders in that window. It tests a few simple methods, such as a straight-line trend and a moving average, on the most recent fifth of that history and keeps the one that came closest. The chart header names the method and its MAE.
MAE is the average daily error when the method was tested. Example: with an MAE of EGP 3,200 and a forecast of EGP 20,000 for a day, a real day between about EGP 17,000 and EGP 23,000 is unsurprising.
An average forecasts a flat line and a trend a straight one, so don’t read the shape as a forecast of peaks. The forecast counts all orders placed except failed ones, including orders later cancelled, so on a cash-on-delivery store it runs above Net Sales.
Churn risk
Churn risk asks whether a customer is overdue. Insights takes each customer’s usual gap between orders, or your store’s average gap for customers with one order, and compares it with the days since their last order. The further past that gap, the higher the risk. Cancelled and failed orders aren’t counted.
One-order customers start higher. Spending that falls across a customer’s orders raises the risk, rising spending lowers it, and very regular buyers get a small reduction. Example: a customer who orders every 30 days and last ordered 60 days ago is twice overdue, which lands in High.
The cards count customers at each risk level and the pie shows the split. High Risk Customers lists up to 10 customers at Critical or High risk, highest first, with Days Since Last and Trend: whether their spending is increasing, stable or decreasing.
- Critical
- A risk score of 75% or more.
- High
- 50% up to 75%.
- Medium
- 25% up to 50%.
- Low
- Below 25%.
Use predictions carefully
Predictions are estimates, not promises. Use them to decide where to look and whom to contact first, then check them against what actually happens.
For the forecast, compare a week of predictions with Gross Order Value on Overview for the same days, which counts orders the same way. If real days often land further away than the MAE, rely on your own plans more than the chart.
For churn, contact Critical and High customers with something related to what they bought, and see whether they order again. A one-order customer at High risk may simply be a one-time buyer; a regular customer at High risk is the one worth a personal message.