12-Month Demand Forecasting from CSV: Preparing ERP & Sales Data for Statistical Replenishment

Why trailing averages miss seasonal swings, how statistical forecasting smooths noise, and how to prepare a clean sales history CSV from your ERP, WMS, or Excel tracker.

Why simple trailing averages fail

A trailing average — for example, the mean of the last three months — treats every recent month as equally informative about the future. That works while demand is flat, but it breaks down exactly when planning matters most:

  • Seasonal swings. An average calculated in the low season systematically under-forecasts the peak, and one calculated at the peak over-forecasts the trough. The error repeats every year because the method has no memory of the annual cycle.
  • Promotional spikes. A one-off promotion or a single bulk order inflates the average for months after it happened, so you keep ordering for demand that has already passed.
  • Trend. A steadily growing product is always under-forecast by a backward-looking average, because the average of the past is always below the present.

Statistical forecasting separates these effects instead of blending them: it estimates the underlying trend and the repeating seasonal pattern, smooths random noise, and caps one-off outliers so a single unusual month cannot distort a full year of plans. The result is a baseline you can trust, plus a confidence band that tells you how much uncertainty remains.

What a statistical forecast gives you

A good 12-month forecast from 24 months of history produces three things per item: a month-by-month baseline, a confidence interval around it, and an explicit record of how outliers were handled. The baseline drives replenishment quantities; the confidence band drives safety stock and service-level decisions; the outlier treatment keeps data errors and one-off project orders from inflating the plan.

The StockLogic Demand Forecast tool offers three statistical logics — a moving average for stable items, linear trend for growing or declining items, and seasonal logic for items with a repeating annual pattern — plus a commercial uplift percentage for the growth or decline you expect on top of the statistics, and outlier capping at 2–3 standard deviations. All calculations run locally in your browser, up to 1,000 SKUs per upload.

How to Prepare Your Historical Sales CSV (ERP, WMS, or Excel)

Every mainstream ERP and warehouse system can export historical transaction or shipment data as a CSV, and a well-kept Excel tracker works just as well. The exact menu names differ, but the pattern is the same:

  1. Open your system's sales, shipment, or goods-issue history report — in SAP this is typically a sales order or delivery extract, in Oracle and NetSuite a saved transaction or item fulfillment search, in Dynamics a sales invoice lines export, and in a WMS the outbound shipment history.
  2. Set the date range to cover as much history as you have — ideally 24 months, so seasonality can be detected.
  3. Aggregate or filter so each row carries one item per period, then export or save as CSV. From Excel, use File → Save As → CSV.

Whatever the source, only three columns matter:

  • SKU — the item identifier, so demand is tracked per item.
  • Date — so months can be ordered and seasonal patterns found; 24 months of history is ideal.
  • Net Quantity Sold / Issued — units actually shipped or issued, net of returns and cancellations.

If your ERP, POS, or Excel tracker exports the same three columns, the same forecast works — the tool is not tied to any specific system.

From export to forecast

Upload the CSV on the Upload page, open the Demand Forecast tool, choose the statistical logic that matches your demand behaviour, set any commercial uplift and outlier capping, and press Analyze. You get a 12-month forecast per item with a confidence band, ready to feed reorder points and safety stock. Nothing is uploaded to a server — the file is parsed and calculated entirely in your browser.

For the wider calculation reference, see Methodology & Formulas.