Perspectives on Sales & Operations Planning

The Optimal Statistical Forecast Level

Working with a client recently, I came across an old problem that challenges many manufacturers trying to produce accurate forecasts. They were forecasting at too low a level and the result was unstable and unreliable forecasts.

The challenge for this client was to predict future sales for each customer. Using the most recent 2 years of demand history, we observed that:
• 36% of customers had no demand
• Of the customers with at least one purchase during past 24 months, 62% had 6 months or fewer of purchase activity

To further complicate matters, demand showed huge spikes and negatives (returns). Facing sparse data and high variability, the statistical forecasting engine produced forecasts ranging from credible to unbelievable.

The solution was to aggregate historical demand and generate forecasts using a more stable demand picture. But just how to aggregate? Turns out there was a natural grouping by industry that served this purpose very well. By combining sales history by industry group, we could isolate demand patterns that were unique to those industries and produce a meaningful and credible forecast.

The next challenge was to design rules to distribute the industry group forecast to the individual customers. To preserve the spike demand nature inherent in this client’s business, we used the prior year’s demand pattern for each customer as basis for distributing the industry group monthly forecasts.

In the end, we achieved a statistical forecast which will serve as a solid demand signal input to their Demand Planning and Sales and Operations Planning (S&OP) process.

Sphere: Related Content

Tags: Demand Forecasting, S&OP, sales and operations planning