Demand forecasting estimates how much of each product customers will buy in a future period, using past sales, seasonality, promotions and known events. Start with simple methods like moving averages and seasonal comparisons, measure the error, adjust for planned activity, and use the forecast to set reorder levels and purchase quantities.
- Forecast at the level you buy at, such as item or category and month or week.
- Simple averages with seasonal adjustment often beat complicated models on small data.
- Track forecast error each period so you know how much to trust the numbers.
- Combine the numbers with human knowledge of launches, offers and supply issues.
Why does forecasting matter for stock-holding businesses?
Every purchase order is a bet on the future. Order too little and you lose sales and customers; order too much and cash gets locked in slow stock, with storage costs and the risk of markdowns or expiry. In India, where festival seasons, weather and wedding calendars swing demand sharply, the swings make guessing expensive.
A forecast does not need to be perfect. It needs to be better than gut feel alone, and clear about its uncertainty. Even a modest improvement in knowing what will move helps you place orders earlier, negotiate with suppliers and plan warehouse space and staffing.
What data do you need?
The core input is sales history by item and date, ideally for at least a couple of seasons so that yearly patterns show. Use actual demand, not just sales: if an item was out of stock, true demand was higher than what you recorded, and ignoring this teaches the forecast to under-order.
Add the factors that shift demand: price changes, discounts, marketing campaigns, festivals, new store openings, competitor moves and supply gaps. Record them in a simple calendar. A clean item master and consistent codes are also essential, because merged or renamed items break history.
- Sales history by item, date and location
- Stock-out days, so lost demand is visible
- Promotions, price changes and campaigns
- Festival and seasonal calendar
- Lead times from suppliers
Which simple methods should you try first?
A moving average takes the average of recent periods as the forecast, and it suits steady items. Seasonal comparison uses the same period last year, adjusted for overall growth, and suits items with strong seasons. Exponential smoothing gives more weight to recent periods and reacts faster to change.
Combine methods with common sense. Fast-moving steady items can use averages. Seasonal items need a seasonal view. New items with no history can borrow from similar products. Intermittent items that sell rarely are hard to forecast, so manage them with safety stock rather than precise predictions.
How do you measure forecast accuracy?
After each period, compare forecast with actual sales. Common measures include the average absolute error, which tells you how far off you typically are in units, and bias, which shows whether you consistently over-forecast or under-forecast. Bias is especially valuable because it points to a systematic fix.
Measure at the level where decisions are made. Totals can look accurate while individual items are badly wrong, since errors cancel out. Review the biggest misses each month and ask what happened: a promotion not recorded, a supply delay, a competitor stock-out. Those answers feed back into better inputs.
- Compare forecast and actual every period
- Track both error size and direction of bias
- Check by item or category, not just in total
- Record reasons for large misses
How does a forecast turn into purchasing decisions?
Forecast demand over the supplier lead time, add safety stock for uncertainty, subtract stock on hand and on order, and the result is what to order. Safety stock should reflect how variable demand and lead time are, and how costly a stock-out is for that item.
Classify items by value and movement so effort goes where it counts. Give careful forecasting to high-value, fast-moving items and keep rules simple for the long tail. Link the forecast to your inventory system so reorder suggestions appear automatically, and let planners override with reasons.
When should you move to advanced models?
Machine learning models can use many factors at once, such as price, weather and promotions, and may help with large catalogues and rich data. They need clean history, ongoing monitoring and someone to interpret them. If a simple method performs about as well, the added complexity is not worth it.
Always compare any advanced model against a simple baseline on data it has not seen. A Plus Solution builds data insight and forecasting solutions and inventory systems, but the principle holds universally: improve the inputs and the process first, and the model second.
Frequently asked questions
How far ahead should we forecast?
At least as far as your supplier lead time plus review period. Longer-range forecasts help with capacity and budgeting but are less accurate.
What if we have only a year of data?
Use simple averages and your own knowledge of the seasons, and be cautious about trends. More history improves seasonal estimates.
Can Excel handle forecasting?
Yes, for a modest number of items. As the catalogue grows, a database-driven tool is more practical.
How do we forecast new products?
Use similar existing products, pre-launch interest and a small initial order, then update quickly as real sales appear.
Need help with this? Ask us a question about it — we reply within one working day.
