How do you predict seasonal peaks in demand planning?

Every year, sometime around late March, the same thing happens. One of your biggest customers calls with a rush order bigger than anything they buy from you for the rest of the year. Your planner frowns, checks the forecast, and sees a neat flat line: an average of the past twelve months. Nothing in that line hinted that April was coming. So you call the supplier again, arrange another rush shipment, and explain internally why the stock didn’t add up. Meanwhile everyone in sales already knew, months in advance, that this customer always orders in April.

What is seasonality in forecasting?

Seasonality is the recurring pattern in demand that repeats at roughly the same point every year, season, or month. Not every fluctuation in your sales figures is seasonality: a one-off spike from a large order or a temporary promotion is noise. Seasonality only counts as seasonality when the pattern repeats, year after year, with predictable timing.

The problem is that many forecasts (especially manual Excel versions and simple moving-average calculations) don’t make that distinction. They look at the past few months and extend that line forward. A customer who only orders in April disappears completely from view for the eleven months they don’t order, only to return in April as a surprise. Every single year.

Common examples of seasonal demand patterns

Seasonality shows up in almost every sector, just not always as visibly:

• Weather-dependent products: sunscreen, garden furniture, and air conditioners peak in spring; ice cream parlour supplies rise as soon as the temperature does.
• Fashion seasons and collections: textile and interior companies see demand spike around the launch of a new collection, and drop once it disappears from the shelves.
• Holidays and trade fairs: packaging, gift items, and promotional material peak around festive periods or a major trade show.
• Contractual or budget moments: some customers order structurally at the end of their fiscal year, or right before an annual maintenance round.
• Recurring large customers: as in the example above, one customer following their own yearly cycle, independent of what the rest of the market is doing.

The pattern is different every time, but the effect on your stock is the same: if your forecast doesn’t know the pattern, you’ll be too late.

How do you recognise seasonality in your data?

The first step is no more complicated than looking at your own sales data, just over a longer period. A few rules of thumb:

1. Look back at least two, preferably three years. One peak could be a coincidence. The same peak in the same month, three years running, is a pattern.
2. Zoom out to product groups, not just individual items. A single item sometimes has too little history to reveal a pattern, while the wider group it belongs to shows a clear seasonal pattern.
3. Separate customer from product. Sometimes the seasonality isn’t in the product at all, but in the ordering behaviour of one specific customer.
4. Compare with external signals. Weather data, holiday calendars, or the launch date of a new collection help explain what you’re seeing in the numbers, and distinguish it from one-off outliers.

This is exactly the point where inventory optimisation succeeds or fails: you can have your ABC classification and safety stock perfectly in order, but if the underlying forecast doesn’t recognise seasonality, you’ll still be chasing the facts.

How does forecasting software handle seasonality automatically?

Recognising seasonality by hand is manageable for a few hundred items. Across thousands of SKUs, spread over multiple customers and markets, it becomes unworkable. Forecasting software solves this by structurally searching for recurring patterns instead of relying on a moving average:

• Pattern detection across multiple years. The model compares the same period across consecutive years and calculates how strong and how consistent a peak or dip returns.
• Separating trend from season. Is demand growing structurally, or is the peak you’re seeing purely seasonal? By pulling these apart, the model avoids mistaking a temporary spike for lasting growth.
• Automatic adjustment per item or item group. Once a pattern is established, the order advice is adjusted automatically, without a planner having to enter a seasonal factor by hand for every item.
• Continuous recalculation. New sales data is factored in continuously, so a pattern that shifts (for example because a customer moves their yearly cycle) gets adjusted too.

The result is a forecast that knows April is coming, well before the phone rings.

Seasonality attributes in Innico: the colour example

One problem stays difficult, even for smart software: what do you do with an item that simply doesn’t have enough sales history yet to reveal a pattern? Think of a new item that has only been through one season.

Innico solves this with seasonality attributes. You can attach a characteristic to every item that indicates which seasonality group it belongs to. Colour is a good practical example. If you sell, say, textiles or interior products in multiple colour variants, those variants often follow the same seasonal pattern: a blue and a grey version of the same item typically peak in the same months, even if one variant sells better than the other.

Instead of judging each item separately on its own (too short) history, Innico adds together the sales data of all items within the same seasonality group. That creates a combined seasonal pattern with enough data to be reliable. That group pattern is then used, alongside the item’s own history, as extra input for the forecast. A brand-new red item benefits immediately from the seasonal patterns that all the other red items have already shown, without you having to set anything up for it yourself.

Conclusion

Seasonality isn’t an outlier in your data. It’s a structural pattern that returns every year, and one your forecast needs to recognise just as structurally. An average over the past few months misses these peaks by definition. Software that specifically searches for recurring patterns, separates trend from season, and, where needed, combines sales data across items that share a characteristic (such as colour), sees that peak coming. Even for a brand-new item, or for that one customer who only calls in April.

Banner image by Yiquan Zhang on Unsplash