Demand forecasting has moved through three broad eras: history-only statistics, driver-based models, and real-time signal-rich AI. Each added something genuine. But “real-time” is not automatically better, it is better only when the decision it feeds can actually act that fast. This is the evolution, and what it means for how you forecast today.
Era 1: history-only statistics
The classical methods, moving averages, exponential smoothing, Holt-Winters, ARIMA, forecast an item from its own past. They are transparent, cheap, and still the right tool for steady demand and for the intermittent long tail. Their limit: they only know what the series itself has done.
Era 2: driver-based models
Adding external drivers, price, promotions, holidays, weather, let forecasts explain demand rather than just extrapolate it. This is where forecasting started to capture cause, not just pattern, and where the comparison with classical methods starts favouring richer models on data-rich SKUs.
Era 3: real-time, signal-rich AI
Modern models ingest live signals, point-of-sale, web behaviour, even external events, and update continuously. On fast movers with real drivers this genuinely improves accuracy and responsiveness. The promise is a forecast that reacts to the world as it changes rather than once a planning cycle.
The catch: real-time only helps if you can act in real time
A forecast that updates every minute is wasted if your replenishment cycle is weekly and your supplier lead time is six weeks. Match the forecast cadence to the decision cadence:
- If lead times are long and reorders weekly, daily-stable forecasts are plenty; real-time is noise.
- Real-time forecasting pays where the response is fast: allocation, dynamic pricing, or short-lead replenishment.
What it means for you today
Do not chase the newest era for its own sake. Segment your catalogue, use classical and intermittent-demand methods on the tail, driver-based or ML on the data-rich head, and adopt real-time only where a fast decision consumes it. Every era still hands off to the same safety-stock and reorder policy, and rests on accurate inventory data. The evolution added tools; it did not repeal the need to match the tool to the SKU and the decision. See demand forecasting for the full toolkit.
Working through this in your warehouse?
The team that wrote this also implements inventory architecture, audits operations, and advises on transformation engagements. AvanSaber’s inventory practice runs case-by-case engagements for mid-market and enterprise inventory teams.
Book a discovery conversation or describe your situation at [email protected].
Learn more about our engagement shapes: Inventory Implementation, Audits, Advisory.