Forecasts are always wrong — the question is how wrong, and whether you can plan around the error. For small and mid-sized manufacturers, good-enough forecasting doesn't require data scientists. It requires a simple method, clean sales history, collaboration with sales, and the discipline to measure accuracy and learn. This guide covers practical methods that work at SME scale.
Why forecasting matters for manufacturers
Lead times for materials and production are usually longer than customers are willing to wait. That means you must decide what to buy and make before orders arrive. Forecasts drive raw material purchases, capacity plans, staffing and stock levels. Poor forecasts lead to stock-outs of what sells and excess of what doesn't — tying up cash and frustrating customers.
Start with clean history
Forecasting begins with sales history by product and month. Clean it first: remove one-off bulk orders that won't repeat, adjust for periods when you were out of stock (sales were limited by supply, not demand), and account for product changes or replacements. Garbage in still means garbage out.
Method 1: Moving average
The average of the last N periods — for example, the last three or six months. Simple and robust for stable products. A shorter window reacts faster to changes; a longer window is smoother. Weighted moving averages give more weight to recent months.
Method 2: Seasonality
Many products sell more in certain months — fans in summer, paints before festivals, school supplies before term. Seasonal indices compare each month's average sales with the annual average. A baseline forecast multiplied by the month's index gives a seasonal forecast. Two or more years of history make indices reliable.
Method 3: Trend
Growing or declining products need trend adjustment. Comparing this year's sales with last year's for the same months gives a growth rate that can be applied to the seasonal baseline. Be cautious extrapolating trends far ahead.
Method 4: Customer schedules and pipeline
For B2B manufacturers, customers often share schedules or forecasts. Combine these with statistical forecasts — customer schedules for the near term, statistical forecasts further out. Sales pipeline data (quotes likely to convert) adds information about new demand.
Consensus forecasting
Statistics don't know about the promotion next month or the competitor who just exited. A monthly consensus meeting — planning, sales, marketing and management — reviews the statistical forecast, adds market knowledge and agrees the number that drives planning. Record the reasons for adjustments so you can later see whether they improved accuracy.
Forecasting at the right level
Forecasting individual SKUs with erratic demand is difficult. Forecasting product families is usually more accurate, with SKU splits based on historical mix. Similarly, forecasting by region or customer group can be more stable than by individual customer.
Measuring forecast accuracy
Track accuracy every month. A common measure is MAPE (mean absolute percentage error) — the average of |actual − forecast| ÷ actual. Also track bias: are forecasts consistently too high or too low? Bias is often more damaging than random error and easier to fix.
Using forecasts in planning
Forecasts feed MRP for material purchases, production planning for capacity, and safety stock calculations for buffer. The forecast error informs how much safety stock is needed: more uncertainty, more buffer. See safety stock calculation.
Common forecasting mistakes
- Forecasting from orders including one-off bulk deals.
- Ignoring stock-out periods in history.
- Letting sales targets replace forecasts.
- Never measuring accuracy.
- Over-adjusting statistical forecasts based on opinions.
- Forecasting at too detailed a level.
How software helps
Forecasting software pulls sales history automatically, applies methods by product, shows seasonality and trend, captures adjustments with reasons, measures accuracy and feeds approved forecasts into MRP and planning. The planner's time shifts from compiling spreadsheets to understanding demand.
Corposol builds demand forecasting software integrated with planning. Talk to us.
A simple monthly forecasting routine
- Week 1: close last month's sales; update history; calculate accuracy and bias by product family.
- Week 1–2: generate the statistical forecast for the next 6–12 months.
- Week 2: sales and marketing review — promotions, new customers, lost customers, price changes, competitor moves.
- Week 2: consensus meeting; agree the forecast and record reasons for any adjustments.
- Week 3: publish to planning; run MRP and capacity checks; flag constraints back to sales.
This rhythm, repeated every month, matters more than the sophistication of the method. Teams that follow it consistently see accuracy improve steadily over a year.
Forecasting new products
New products have no history. Use analogies — sales of a similar product at launch — adjusted for distribution and marketing plans, and review weekly in the first months. As real sales arrive, switch to statistical methods. Planning conservatively with fast replenishment is often safer than building large stock for an uncertain launch.
Intermittent demand
Spare parts and slow-moving products may sell zero in most months and a few units occasionally. Standard averages handle this poorly. Forecast these items at a family level, use longer windows, or manage them with reorder points and safety stock rather than period forecasts.
Forecast value added
A useful check is "forecast value added": compare the accuracy of the statistical forecast with the final consensus forecast. If manual adjustments make accuracy worse, it's a sign to adjust less — or to require stronger evidence before overriding the numbers.
Frequently asked questions
Do we need advanced AI forecasting?
Most SMEs get most of the benefit from simple, well-run methods — moving averages, seasonality and good collaboration with sales. Advanced models help when data is rich and patterns complex.
How much history do we need?
Twelve months gives a start; two to three years is better for seasonal products.
Should we forecast every SKU?
Forecast at the level where demand is stable — often product families — and split to SKUs using historical mix.
How often should forecasts be updated?
Monthly for most businesses, with weekly adjustments for fast-moving products or volatile demand.
Who should own the forecast?
Typically planning or supply chain, with input from sales and marketing in a monthly consensus meeting.
