Demand forecasting in a manufacturing SME is rarely a formal process. The sales rep “feels” the market, production makes what sold last month, and purchasing orders when stock looks low. As soon as the number of items grows and bills of materials run to several levels, stock-outs and emergency replenishment become part of daily life.
We saw this situation at close quarters in a growing SME manufacturer, with items shared across multi-level bills of materials: a poor service level, sales staff on the phone all day, and management unhappy about late deliveries. The starting point was not software, but the lack of a usable forecast.
This article explains why the forecast drives stock, purchasing and production, then sets out a simple method to get started: history, seasonality, promotions, projected stock. You do not need a complex statistical model to begin: you need a regular routine and reliable data.
Why demand forecasting affects the whole chain in a manufacturing SME
The forecast is the first link in a chain of decisions. It feeds the master production schedule (MPS), which sets what is made per period and per finished product. The MPS feeds the MRP run (net requirements calculation), which explodes the bills of materials and derives the components and materials to buy or make, after deducting stock and work in progress. Purchase orders and manufacturing orders follow from it.
Without a forecast, every link works by eye, with the same consequences every time:
- Stock-outs on fast movers, because nobody anticipated the increase;
- Overstock on slow movers, ordered “just in case”, tying up cash;
- Emergency replenishment, more expensive, disrupting suppliers and the shop floor;
- A falling service level, and a sales team that spends its days apologising.
The more levels the bills of materials have, the stronger the effect: a component shared by twenty finished products accumulates twenty forecasting errors.
The prerequisites: reliable stock and accurate bills of materials
Even an excellent forecast is useless if the opening stock is wrong: projected stock is calculated from actual stock. If stock is managed by eye, with constant manual checks, the calculation produces unnecessary purchases or stock-outs nobody sees coming.
Before forecasting, three things need to be secured:
- Goods receipts: a clear receiving procedure, recorded the same day, with a quantity check.
- Stock issues: accurate order picking, because every picking error is also a stock error.
- Items and bills of materials: unique codes, up-to-date BOMs, consistent units.
A simple five-step method
1. Build a clean sales history
Extract sales (quantities, not revenue) by item and by month over 24 to 36 months. Flag the abnormal months: stock-outs (artificially low sales), one-off orders, promotions. Correct or mark them, otherwise the forecast will repeat them.
2. Calculate the baseline and seasonality
The baseline is the regular level of sales: a moving average of recent months, excluding events, is enough to start. Seasonality is measured with a monthly coefficient: sales for the month divided by the average monthly sales for the year. A coefficient of 1.3 in May means May is 30% above average. The forecast for a month is then baseline × seasonal coefficient.
3. Handle promotions and events separately
A promotion or a product launch cannot be forecast from history: it is entered as an additional quantity, with its period and customer, kept separate from the baseline so that it does not inflate next year's base.
4. Project stock and stock cover
For each item and each month:
Projected month-end stock = opening stock + planned receipts + planned production − sales forecast
Stock cover (in months) = projected stock ÷ next month's forecast
A negative projected stock signals a stock-out; cover above target signals overstock. It is this column that triggers purchases and manufacturing orders, not the forecast alone. A fictitious example: a baseline of 1,000 parts a month, an opening stock of 1,500 parts, steady production of 1,000 parts a month, and a forecast of 1,200 parts in June.
| Month | Seasonal coefficient | Forecast | Planned production | Projected month-end stock | Stock cover |
|---|---|---|---|---|---|
| January | 0,9 | 900 | 1 000 | 1 600 | 1.6 months |
| February | 1,0 | 1 000 | 1 000 | 1 600 | 1.2 months |
| March | 1,3 | 1 300 | 1 000 | 1 300 | 0.9 months |
| April | 1,4 | 1 400 | 1 000 | 900 | 0.7 months |
| May | 1,3 | 1 300 | 1 000 | 600 | 0.5 months |
From March onwards, steady production no longer keeps up with the season: you need to produce more in January and February, or increase capacity in spring. With a component on an eight-week lead time, that decision is taken in January. This is the whole point of forecasting: turning an April problem into a January decision.
5. Measure the error and start again
Every month, compare the forecast with actual sales: absolute percentage error for accuracy, signed error for bias (systematically too high or too low?). A 30% error on a slow-moving item is acceptable; a constant 15% bias on a major item is not.
From forecast figures to the production plan
The forecast becomes useful once it feeds the MPS and then the MRP run:
- the MPS turns the forecast per item into quantities to produce per period, smoothing the load when capacity is limited;
- the MRP run explodes the bills of materials level by level and calculates, for each component, the net requirements after deducting stock and work in progress, taking lead times into account;
- the planned manufacturing orders and purchase orders are approved by the planner, who stays in control.
At product-family level and over a one-year horizon, this logic is called sales and operations planning (S&OP). A classic MRP run works at infinite capacity, as if the shop floor could do everything: we cover its limits and the workarounds in our article on infinite capacity MRP planning. The earlier the forecast is maintained, the more room the MPS has to smooth the load.
What we see in the field: foundations before forecasts
The SME manufacturer mentioned in the introduction shows the road to travel. A successful family business, it was happy to invest in property and machines, much less in IT: a management system produces nothing tangible, costs a lot, and stories of failed ERP projects were hardly encouraging. An ERP project had in fact been started, then stopped at the first dip in revenue.
Stock was managed by eye and by word of mouth, the computer data was unreliable, and finished goods were tracked by hand. Yet the foundations were good: items correctly coded, bills of materials in order, responsive teams and quick decisions. The action plan therefore started with the foundations, without waiting for a tool:
- making the system stock reliable through goods receipt procedures;
- accurate order picking to make stock issues reliable;
- introducing manufacturing orders for finished products;
- building a finished-goods tracking table that served as the master production schedule;
- developing a net requirements calculator for components and materials.
Each step made stock more reliable and streamlined production. Once these foundations were in place, sales forecasting could become a process in its own right: collect, measure the quality of the figures, improve every month. The order matters: forecasting on wrong stock would have produced nice tables and the same stock-outs. Other projects are described in our case studies.
Conclusion: forecasting starts with reliable stock
Demand forecasting in a manufacturing SME needs neither a data scientist nor a full ERP. It needs reliable stock, a clean history, a clear separation between baseline and promotions, a projected stock updated every month and a measure of the error. The rest, MPS and MRP, follows from it. What makes the difference is the regularity of the cycle, not the sophistication of the model.
Planiprev was designed to run this cycle without an Excel file held together by hand: forecasts item by item and month by month, baseline kept separate from promotions, projected stock, stock cover and anticipated stock-outs, change history and comments. It takes its data from the ERP or from your Excel workbooks. Find out more about Planiprev or request a demo.