- A demand forecast predicts units by product and location for each period of a horizon you choose, refreshed as new sales, promotions and weather data arrive.
- H2O Driverless AI builds time series models with automatic lag, rolling and calendar features, and explains which drivers moved each forecast.
- For direct-store delivery, the same models forecast by SKU, route and territory, so each route starts the day with an order suggestion per outlet.
- tabH2O, H2O's tabular foundation model, returns predictions inside a spreadsheet with no model training, for the questions that never reach a data science team.
You order today for demand you see next month
Your buyers, planners and route teams commit stock, trucks and shelf space before the demand arrives. When the plan rests on last year's sales and a spreadsheet of adjustments, the gap shows up as stockouts on the items that sell, markdowns on the ones that do not, and spoilage on anything fresh.
Most retailers and consumer brands already forecast at some level, but the forecast is often built for the total, one region at a time, and rarely reaches the store, the SKU or the route where the decision is made. Key account managers and route supervisors add their own forecasts in spreadsheets, and those numbers sit outside the planning system, so no one can compare or improve them.
How demand forecasting works
The forecast is trained on your own sales history and predicts future demand from past sales, the calendar, prices, promotions and events you already know about.
Step 1: connect the history
Pull sales by product, store and day from your point-of-sale and planning systems, with prices, promotions, stock levels and store attributes. The forecast feeds your planning system; it does not replace it.
Step 2: set the level and horizon
Choose what each forecast predicts, such as SKU by store by week or SKU by route by day, and how far ahead it looks.
Step 3: build and compare models
H2O Driverless AI creates lag, rolling-window and holiday features through automatic feature engineering on your time series data, and ranks candidate models on held-out periods.
Step 4: explain the forecast
Interpretability tools show which drivers moved each forecast, such as a promotion, a price change or a holiday, so your planners can check it against what they know.
Step 5: run it in shadow mode
Run the forecast beside your current plan first, and compare both with actual sales before anyone orders from it.
Step 6: deploy, monitor and act
H2O MLOps scores the forecast on a schedule and watches for model drift. H2O AI Super Agent can turn forecast gaps into replenishment exceptions and draft the order change for a planner to approve.
Forecasting for direct-store delivery, and inside the spreadsheet you already use
In direct-store delivery, your own trucks stock each outlet, and the forecast predicts demand per SKU for each outlet on each route, then rolls up to the route and the territory. For fresh and short-life products, that same forecast is where you cut spoilage and empty shelves.
Not every forecast needs a data science project. tabH2O returns predictions from labeled examples with no training or feature engineering, and its add-ins run inside a spreadsheet, so a key account manager can predict a customer's next orders from the history already there, without a new tool.
Proof
H2O can be used to forecast demand by product, region and season, by SKU, route and territory for direct-store delivery, and inside the spreadsheets your key account managers use. No retailer or consumer brand has published a demand forecasting result on H2O, so this page does not quote one.