AI Predictions
See what is coming before it happens: sales for the next 4 weeks, 3, 6 and 12 months with a safe range, the products that will run out before an order can arrive and how much to order, which categories are rising or falling, the regular customers likely to stop buying and the days that looked unusual. Every forecast is tested on your own past sales first, and the method that was most accurate is used, with its error shown. Part of the AI Predictions add-on for industrial supply stores in Saudi Arabia.
Steps
- Open Dashboard โ AI Predictions. Sales outlook shows the next 4 weeks' expected sales, the safe range, the change against the last 4 weeks and a bar a day. Pick 3 months, 6 months or 1 year above it for the longer outlooks, with a bar a week; 6 months needs about 8 months of sales and 1 year about 14.
- Stock-out risk lists the products out of stock or that run out before an order can arrive first, then the ones to order now, with the quantity to order up to the safe stock (rounded up to the reorder rule's pack size).
- Category demand shows each category's expected units for the next 4 weeks and how it moved against the 12 weeks before. Customers at risk lists regular customers by their chance of not buying in the next 90 days, with when they last bought and their usual gap.
- Unusual days lists the days of the last 90 that were much higher or lower than the usual for that weekday. Accuracy shows each method's error on your last four periods and which one is used.
Tips
- Each business type sees the boards that fit it: a business that sells only services (a clinic, a salon, a law firm) has no stock boards, and where people buy once in a long while (a showroom, a hotel, real estate) there is no customers-at-risk board. A clinic sees patients, a salon or agency clients, a co-working space members.
- Four methods compete: weekday average, same weekday, trend and level, and same weeks last year. Each one forecasts your last four periods as if they had not happened yet; the one with the smallest error wins, and the safe range is its largest miss.
- Once a week a machine-learning run (Chronos-2, LightGBM and statistical models) forecasts the same four periods. Its forecast is used only when its error is smaller than the best of the four methods; the Accuracy tab says which one is used and when it last ran.
- Single products sell in lumps, so the stock check uses a safe level (enough for 9 of 10 past periods of the lead time) rather than a guess of the exact quantity. Large one-off project orders are shown apart and do not inflate the everyday forecast.
- The customer model is learnt from your own customers' history and its AUC is shown (1 is perfect, 0.5 is a guess). Predictions start after about two months of sales and refresh every few hours.
In the app, press F1 or ? on this screen to open this guide next to your work.
This guide belongs to the AI Predictions module. Learn more
Related guides
More guides in Dashboards & reports
- Maintenance & owner statements
- Profit planner
- Hotel dashboard
- Signage dashboard
- Real estate dashboard
- Butcher dashboard
- Spare parts dashboard
- Accessories dashboard
- Agency dashboard
- Fleet dashboard
- Hospital dashboard
- Fuel dashboard
- Ayurveda dashboard
- Your business type's dashboard
- Restaurant dashboard
- Inbox report
- SLA report
- Asset register report
- Bank reconciliation report
- Waiting-time report
- Hours & overtime report
- StartAI daily brief
- Business Advisor
- StartAI Radio
- Business health
- Business dashboard
- Operations dashboard
- All branches
- Sales analytics
- AI assistant chat
Still need help?
Our team answers in Arabic and English by email and phone.