Forecast Accuracy Improvement

Forecast accuracy is a necessary tool for supply chain excellence. But it is just a means to an end and not an end in itself. Building sophisticated forecasting techniques and trying to reach perfect forecast accuracy is a no-win game in this world of short product life cycle, heavy price and promotional activity, and shifting customer choices. Some of the best practices seen in demand management which improves forecast accuracy besides using the appropriate forecasting techniques are

1. Internal and external collaboration to validate the forecast and incorporate externalities. Implement a sales and operations planning process to get the inputs.
2. Focus, unless things change dramatically, on making the plan happen rather than continually recreating forecast based on the latest market data. Continuous monitor and shape demand when necessary to meet the plan rather than accept the latest demand forecast modification request.
3. Understand the use of forecasting data – forecasting for deployment need not be for more than a couple of weeks; forecasting for long term planning can be at product family / monthly level.
4. Connect the forecast to the sales / revenue plan and fix any discrepancies.
5. Connect the order forecast from the retailer to the Sell-Out data (POS data).
6. Postpone configuration as late as possible to take advantage of higher reliability in forecast for components rather than end item.
7. Segment products based on demand characteristics. (1) For some products, you can just use customer order stream. (2) For some products, you could improve the forecast accuracy by using customer input but applying statistical techniques to it. (3) For some products, you might have to completely disregard customer input. Understand the demand characteristics before applying forecasting process. Work with customers to move them from category 3 to category 1.
8. Implement continuous learning process for the assumptions (product ramp-up curve, cannibalization curve, promotions effectiveness etc.).

In the real world, 100% forecast accuracy is unattainable and mostly unnecessary. Successful demand management is to know what can be forecasted accurately at a particular granularity and more importantly what can not be forecasted. Good enough forecast accuracy combined with intelligent inventory decisions is the best combination to achieve maximum “perfect order fulfillment” score.

Counter points? Additional insights on this subject?

Karthik Mani