Seasonal Demand Forecasting for SMEs Still Driven by Gut Feeling

Why Seasonal Demand Forecasting in SMEs Always Arrives Too Late

Seasonal demand forecasting in SMEs is the attempt to understand, months in advance, how much to sell, produce or purchase when the market accelerates or stalls. It is not a chart for a presentation: it is the difference between empty shelves and stationary pallets. Those who work without clean historical data, without a shared calendar and without clear reorder rules make decisions by intuition. The result is a peak that arrives earlier than expected, or a slowdown that leaves the warehouse full of goods that are not moving.

In 2026, many Italian businesses still keep sales, purchasing and inventory in separate spreadsheets. The sales team updates one file, the warehouse uses another, and production looks at the wall calendar. The actual delivery date is missing, the distinction between a confirmed order and a hypothesis is missing, and the historical data from previous years aligned to the same product references is missing. Anyone who needs to launch a batch or sign a raw material order does not read a unified picture: they read delays, chat messages and memories.

Why Seasonal Demand Catches Your SME Off Guard

Seasonal demand catches businesses off guard when signals arrive late, scattered and contradictory, not when the commercial calendar is missing.

The Christmas peak, the September restart, the industry trade fair or the April bank holiday weekend are not mysteries. The problem is that every function looks at them through a different window. Those in sales hear the phone ringing. Those in production see the order backlog growing. Those in purchasing receive urgent requests when the supplier's schedule is already full. Nobody keeps volume, lead time and actual line capacity together in one place.

In practice, seasonality is a set of constraints that move together. If a product reference turns over for only eight weeks a year, the mistake is not failing to anticipate it: it is failing to close batches, packaging, shifts and reorders in time. The management system records what was sold. It rarely records why an order slipped, whether it was a promotion or whether the customer moved up their collection. Those gaps become the next wrong forecast.

Those Who Have Little Time to Close Orders Before the Peak

Those who manage a few dozen people and handle sales, warehousing and administration at the same table do not have a dedicated planning room. Forecasting becomes an end-of-month meeting, often after suppliers have already closed their schedules. In that narrow window, decisions are made from memory: the previous year is replicated, stock is raised slightly "just to be safe," and everyone hopes the weather or the customer's campaign does not change. When the peak arrives, there is no time left to make corrections.

Those Who Produce to Order and Fear Line Stoppages

An illustrative example: a workshop with short batches and raw materials with long lead times. If September fills up and June is empty, the temptation is to build up warehouse stock in advance. Without historical data by order and by week, however, parts are purchased that will not be needed, and the ones that are needed are discovered too late. The stoppage does not arise from a lack of effort: it arises from demand being read too late relative to the actual purchasing and setup cycle.

Those Who Sell Across Multiple Channels With Non-Overlapping Calendars

Another illustrative example: a catalogue of a few dozen product references, sold partly in-store, partly through e-commerce and partly at trade fairs. The website's Christmas is not the same as the shop counter's Christmas, and the trade fair shifts volumes that are then missing elsewhere. If the three channels do not share stock and commitments, the same reference can appear both available and out of stock on the same day. In that case, seasonal demand forecasting is not a sophisticated model: it is first and foremost an agreement on what has already been sold and what is still a hypothesis.

The Hidden Costs of a Wrong Forecast

A wrong forecast costs tied-up capital, emergency purchases, end-of-season discounts and hours taken away from work that generates margin.

The visible damage is the empty shelf or the pallet that does not move. What erodes the accounts lies beneath the surface. You pay for extra transport, you pay the minimum order quantity imposed by the supplier, you pay overtime in production, you pay the discount to clear what is left over. You also pay in credibility: a customer who has heard "it arrives next week" three times stops asking. On this ground, the invisible costs that erode margins accumulate in silence, because they have no single line in the accounts.

• Capital tied up in stock covering a peak that has already passed.
• Emergency purchases at worse conditions, with lead times the line cannot meet.
• Meeting hours spent chasing gaps instead of confirming batches.
• End-of-season discounts on goods that could have stayed out of the warehouse.
• Production stoppages due to a component that "was there" in the wrong spreadsheet.
• Promises to customers that the warehouse cannot keep.
• Dirty data that, the following year, makes the new forecast even more blind.

A wrong forecast is not an isolated calculation error: it is stock sitting idle, capital tied up and people spending their days putting out fires.

Why Spreadsheets and Intuition Are No Longer Enough

Spreadsheets and intuition hold the memory of one season, not the link between what was sold, stock levels, reorder lead times and today's actual capacity.

Spreadsheets are honest: they do exactly what you ask of them. The limitation lies upstream. Every column only stays current if someone updates it. A product reference changes its code, a customer moves an order, a return comes back in and the spreadsheet stays frozen. The intuition of someone who has been with the company for years remains valuable, but it does not scale when product references grow, channels multiply and the people who "knew everything" are not all in the office on the same day. The warehouse managed on spreadsheets follows the same pattern: apparent stock, real sales and a gap in between.

What is usually missing is a single version of the truth. Missing is the date when goods are actually available, not the promised date. Missing is the supplier calendar aligned with your peak weeks. Missing is the end-of-season review, compared line by line with what you had forecast. Without that comparison, the following year the same mistake is copied with greater confidence.

How Demand Forecasting Works With Artificial Intelligence

Artificial intelligence compares historical data, the calendar and stock constraints, then proposes a demand range that a person can accept or adjust.

In simple terms, the system reads past sales by product reference and by week, recognises recurring patterns (holidays, back-to-school, trade fairs, weather), and places them alongside what you already know: lead time, minimum order quantity, safety stock, line capacity. It does not "predict the future." It lines up facts that a spreadsheet keeps scattered. If the facts are vague, the proposal is vague. If the facts are clean, the proposal becomes a useful starting point for a meeting, not an oracle.

The human step remains decisive. Anyone who knows that a customer has closed down this year, or that a promotion will not be repeated, must be able to adjust the range. Automation serves to avoid forgetting minor product references, to flag deviations and to close the loop: forecast versus actual, every month. The same principle applies to artificial intelligence applied to warehouse management: it does not replace the person in charge, it makes visible what previously existed only in one person's head.

Artificial intelligence does not replace the judgement of those who know the customers: it compares historical data, the calendar and stock constraints when spreadsheets can no longer keep up.

How to Get Started Without Overhauling Business Processes

You start by consolidating historical data by product reference and separating confirmed orders from hypotheses, without changing the entire management system overnight.

The temptation is to buy "the forecasting software" and hope it cleans the data on its own. That does not happen. First you decide which product references are worth the effort (those that tie up cash, space and long lead times), then you put what already exists in order. A single file for weekly sales, a shared calendar of peaks, a written rule on minimum stock and reorder lead time. Only after that does it make sense to automate the repeated steps: extraction, comparison, the alert when the actual result deviates from the forecast.

1. Consolidate sales history by product reference and by week, using the same code everywhere.
2. Mark trade fairs, closures, bank holidays and already-decided campaigns on the real calendar.
3. Clearly separate confirmed orders, expected orders and simple hypotheses.
4. Set minimum stock and reorder lead time for the product references that tie up cash.
5. Compare forecast against actual every month, and note down why the forecast was wrong.
6. Automate only the repetitive steps, leaving the final correction to a person.

Those who already have a management system do not need to discard it. They need to make it say the same things the warehouse says. Those starting from spreadsheets can still close a short cycle: clean historical data, even if brief, is worth more than years of inconsistent columns, although reading seasonality does require at least one complete seasonal cycle. The goal is not an elegant model. It is reaching the next peak with fewer emergencies and with a list of mistakes already seen, not ones to be rediscovered under pressure.

How Leomat Makes Planning Concrete for SMEs

Leomat designs artificial intelligence automations and custom management systems so that sales, warehousing and quotations speak the same language.

Leomat was created to bring frontier technology to those who run businesses. That is why the work starts from real processes, not from a platform to be forced into shape. One of the ways we do this is by building a custom ERP tailored to the company, designed to bring together the operational facts that today live in separate spreadsheets. Without a unified picture, seasonal demand forecasting remains an exercise in intuition.

One already-measured example is ERP Costruzioni, in the construction sector: drawing up a quote went from 8 hours to 5 clicks in 30 days. It is not seasonal magic, it is the same principle that applies to demand: removing hours from a slow and repetitive step. If you want to understand whether a management system tailored to your workflows can close the loop between orders, stock and planning, the starting point is Leomat and AI automations for SMEs.

How Is Seasonal Demand Forecasting Done in an SME?

You start with clean historical data by product reference and by week, mark the real peaks on the calendar, separate confirmed orders from hypotheses and set minimum stock and reorder lead time. Then you compare every month what you forecast with what you actually sold. Artificial intelligence helps keep these facts together and flag deviations. The final decision stays with those who know the customers, suppliers and line capacity.

Is a Spreadsheet Enough to Manage Warehouse Seasonality?

It is enough as long as product references are few, a single person updates the file and the sales channels align. When codes, shifts and customer commitments grow, the spreadsheet falls behind the actual stock level. The link between what was sold, incoming goods and supplier lead time is missing. The intuition of someone who has been with the company for years remains useful, but it does not replace a unified picture. The spreadsheet can stay as a check, not as the sole operational memory.

What Data Is Needed for AI-Based Demand Forecasting?

You need sales by product reference and by period, returns, promotions, lead times, minimum order quantities and safety stock. You also need to know what is already committed and what is merely expected. If codes change every season without a bridge to the historical data, the model reads noise. A few product references with consistent data are better than the entire dirty catalogue. Artificial intelligence confirms the facts when they match what you see in the warehouse and in production.

Can You Get Started Without Changing the Entire Management System?

Yes. You choose a narrow scope, usually the product references that tie up cash and involve long lead times, and you consolidate that historical data. You write a reorder rule and close a monthly comparison cycle. Automation comes later, for the repetitive steps. A custom ERP makes sense when spreadsheets can no longer keep up and departments are not sharing the same numbers. Overhauling everything on the first day lengthens timelines and leaves the data in the same condition as before.

This post was created with AI