Resources · 3 min read
Retail demand forecasting: start with what you already have
You don't need a data lake or a six-month project. You need your sales history, twenty items, and an honest way to measure whether it's right.
A forecast isn't a number: it's a decision
“We'll sell 340 units” is useless on its own. It's useful when it changes a decision: how much to buy, when to restock, what to promote before the season expires. Before forecasting anything, define which decision the number will change. If the answer is “none, it's just to know”, the project is already dead — it just doesn't know it yet.
You already have the minimum dataset
Sales by item and by date — that's all it takes to start. Twelve months or more if you want to capture seasonality; with less, you start anyway and improve with every month. You don't need to clean the whole catalog or integrate every system: you need the history to come out of the system where it lives, not from a spreadsheet someone builds by hand every Monday.
Censored demand: the error that ruins forecasts
Last month that item sold 40 and ran out on the 18th. How much demand was there? More than 40 — how much more, you don't know: days without stock don't record the sale that couldn't happen. If the model learns from those numbers uncorrected, it learns to repeat your stockouts. Flagging stockout days in the history is the least glamorous and most profitable correction in the whole project.
Start narrow: twenty items, not the catalog
Pick the twenty that hurt the most — the ones that sell the most or ran out most often. Forecast only those, compare every week against what actually happened, and expand when the accuracy holds. A narrow pilot that wins defends itself in any meeting; a wide one that “sort of works” doesn't survive its first quarter.
What not to buy yet
Not the twenty-thousand-dollar platform, not the data warehouse, not the “digital transformation”. The first version is a read layer on top of your ERP plus a backtest that says how often it's right. If that version doesn't improve your buying decisions, no platform was going to — and you found out for a fraction of the price.
Rather go through it on your own case?
A 15-minute audit of your real operation. You leave with an honest read on what to automate first, and what to leave alone.
Book a 15-min audit