AI demand forecasting at SKU, store and week level, built on your sales, promotions, prices and lead times. Stratgik turns the forecast into purchase orders your buyers approve in NetSuite, Shopify or SAP.
Sample data. The promo lift in week 15 is forecast from past promotions at the same discount depth, so the PO lands before the spike, not after it.
AI demand forecasting is the use of machine learning models to predict future unit sales for each product and location from sales history, promotions, pricing, seasonality, stock availability and external signals such as weather or events. It outputs a weekly forecast with an uncertainty range, and turns that range into recommended order quantities, safety stock levels and reorder dates that planners can approve or adjust.
Most mid-market planners still extend last year's sales in a spreadsheet, add a gut-feel uplift and hope the supplier ships on time. The result is cash tied up in the wrong SKUs and lost sales on the right ones.
Inventory distortion costs retailers $1.73 trillion a year. IHL Group puts out-of-stocks plus overstocks at 6.5% of global retail sales in 2025.[1]
North America carries $415 billion of that loss. It is not an Asia-only or big-box-only problem.[1]
Spreadsheet forecasts are refreshed too slowly. A promo, a heatwave or a late container changes demand in days, not at month-end.
Fewer lost sales from product unavailability. McKinsey reports AI-driven supply chain forecasting can cut them by up to 65%.[2]
We pull clean weekly history per SKU and location, and fix the thing most models ignore: sales lost while you were out of stock.
Gradient-boosted and hierarchical models forecast every SKU and roll up cleanly to category and region, with a range, not a single number.
The forecast becomes an order: quantity, supplier, order-by date, respecting MOQs, case packs, lead-time variance and your cash budget.
Read-only access to two to three years of sales, stock and PO history. We agree the pilot metric, usually weighted MAPE or stockout weeks, and the category.
Models forecast past quarters they never saw. You get SKU-level error versus your existing forecast, including where the model is worse.
The system drafts real POs next to your buyers' orders. Differences are reviewed together, line by line, with the reason for each.
Measured error, projected inventory and service impact, and a costed plan to extend to all categories under the Run fee, or stop.
| Criterion | Spreadsheet planning | Off-the-shelf planning tool | Stratgik build + run |
|---|---|---|---|
| Setup time | None, already in place | A multi-month implementation project | 21-day pilot on one category |
| Handles promos and stockouts | Manual uplift guesses | Yes, if data is mapped to its format | Modelled from your own promo and stockout history |
| Fits your ERP and workflow | Copy-paste | You adapt to the tool | POs drafted in your existing ERP |
| Transparency | Formula visible, logic in someone's head | Varies; often a black box | SKU-level drivers and backtests shared |
| Best fit | Under a few hundred SKUs, stable demand | Large teams wanting a full planning suite | Mid-market with messy data and specific constraints |
| Ongoing cost | Planner time | Annual licence per user or SKU | Monthly Run fee, cancel any month |
Rates are higher, so every pound, dirham or dollar sitting in slow stock costs more to finance. Meanwhile customers switch to a competitor the moment an item shows out of stock.
Two effects: less inventory carried for the same service level, and sales recovered from stockouts. The recovery rates are assumptions we measure in the pilot.
Illustrative estimate using your inputs and stated assumptions, not a quote or guarantee. The pilot measures the real figure against your baseline.
One category, backtested and run live in shadow for 21 days
Weekly forecasts and draft POs for one business unit
All categories, channels and locations
Cloud compute and data warehouse usage billed at cost with no markup; typically under $400 a month for mid-market catalogues. Taxes excluded. GBP and AED prices are indicative conversions from USD.
AI demand forecasting is typically more accurate than spreadsheet methods, and McKinsey reports AI-driven supply chain forecasting can reduce errors by 20 to 50 percent. Accuracy depends on your data: fast-moving SKUs forecast well, long-tail and brand-new items less so. In the pilot we backtest on quarters the model never saw and report error SKU by SKU against your current forecast, including where it does worse.
Two years of weekly sales by SKU and location is a good starting point, because it covers at least two seasonal cycles. One year can work with strong look-alike products. We also need stock-on-hand history, a promotions calendar and supplier lead times. Gaps and messy data are normal; cleaning them is part of the first week of the pilot.
A Stratgik AI demand forecasting pilot costs $15,000 for one category over 21 days. Ongoing Run is $3,500 a month for weekly forecasts and draft purchase orders in one business unit, and Scale starts at $9,000 a month across all categories and locations. Cloud usage is billed at cost. There are no per-SKU or per-user licence fees.
Yes. We read sales, inventory and purchase history from NetSuite, Shopify, SAP, Dynamics 365, Cin7 and most POS or WMS systems through their APIs or scheduled exports. Recommended orders are written back as draft purchase orders, so buyers approve them in the ERP they already use. Nothing is placed with a supplier without a human approval.
Often, but not always. Built-in ERP forecasting usually applies moving averages or simple seasonality, which works for stable, high-volume items. It struggles with promotions, stockout-distorted history, new products and multiple channels. The pilot compares both on the same SKUs, so you only switch where the machine learning model is measurably better.
Stockout weeks are corrected so lost demand is estimated rather than recorded as zero. Promotions are modelled by discount depth, channel and placement from your own history. New products borrow patterns from similar SKUs by category, price point and launch season, then shift to their own data after a few weeks of sales.
Send us the category that hurts most, whether that is stockouts, dead stock or both. We will scope the pilot, the data we need and the one number we will be judged on.