Supply chain forecasting estimates future demand to inform production, inventory and supply commitments. Every method produces error. The goal is error that is quantifiable and manageable. The most consistent driver of improvement is accountability for forecast inputs at their source, not better forecasting technology.
In most organisations the demand planning team generates a statistical baseline, sales adds commercial overrides, and the resulting consensus forecast is owned and defended by demand planning.
When it is wrong, demand planning absorbs the accountability. The sales team whose overrides drove the error is measured on revenue, not on forecast accuracy, and has no incentive to improve their input quality.
Four methods and when each fails
| Method | Use when | Fails when |
|---|---|---|
| Statistical extrapolation | Stable products, 2+ years history | Inflection points, new competitors, channel shifts |
| Causal / regression | Strong external drivers identifiable | The driver relationship changes |
| Judgment-based | New products with no history | Systematic optimism bias |
| Machine learning | Long history plus multiple signals | Sparse or inconsistent training data |
New product forecasts supplied by commercial teams run persistently above actual launch demand across industries. The bias is structural, not dishonest: the incentive system rewards ambition and the inventory consequence lands in a different budget.
The metric problem
Mean absolute percentage error is the standard measure, but it weights every item equally regardless of financial impact. A 20% error on a $10M product costs roughly $2M. The same 20% error on a $50,000 product costs $10,000. Standard MAPE treats them as identical.
Weighted MAPE scales each item’s error by its revenue contribution, which measures what forecast error actually costs. Organisations that switch usually discover their respectable headline accuracy was driven by low-value items while high-value items were being missed.
Error attribution changes behaviour
| Stage | Typical accuracy movement | Owner |
|---|---|---|
| Naive baseline | Starting point | None |
| Statistical model with seasonality | Substantial improvement | Demand planning |
| External data integration | Further improvement | Demand planning |
| Commercial overrides | Often unmeasured, can worsen accuracy | Sales, usually unaccountable |
| Promotional adjustments | Often unmeasured | Marketing, usually unaccountable |
Measuring the accuracy contribution of each input source separately is the change that shifts behaviour, and it costs nothing beyond reporting effort. When data shows overrides consistently worsening the statistical baseline, and that data is visible to the function making them, the conversation changes.
Signal quality upstream matters as much as method, a link explored in AI forecasting and inventory planning and in the planning process covered by supply chain planning.
For continuing demand planning coverage, follow our current supply chain analytics reporting.
Frequently asked questions
What is supply chain forecasting?
The process of estimating future customer demand across a planning horizon to inform production schedules, inventory positions and supply commitments. All methods produce error; the objective is error that is quantifiable and manageable so the supply chain can size buffers appropriately rather than discovering mismatches at stockout.
What is MAPE?
Mean absolute percentage error, the average absolute difference between actual and forecast demand expressed as a percentage of actual. Its main limitation is weighting all items equally regardless of value, which allows a good headline number to coexist with systematic errors on the highest-value products.
How do you actually improve forecast accuracy?
Three changes deliver most of the gain: measure the accuracy contribution of each input source so overrides are accountable, increase update frequency for fast-moving items, and use near-real-time sell-through data to adjust short-horizon forecasts more often than the planning cycle allows.
Why do sales forecasts overestimate new product demand?
Because incentives reward ambitious launch numbers that support capacity, promotional budget and distribution arguments, while the cost of overestimation appears as excess inventory and write-downs in supply chain and finance budgets rather than in sales performance reviews.
