Friday, May 7, 2010

Obsession with Demand Forecasting

Few days back I was interviewing candidates for Demand Planning position. I asked one of the candidates to share his greatest frustration as a demand planner. What he shared was quite shocking. He had been given a target on forecast accuracy that he missed completely due to uncertainty of tender business, which contributed about 15-20% of the total business. Though company could get the sales and sales team earned handsome incentives but the poor guy lost his annual bonus.

Many companies have a utopian belief that by having a dedicated demand planner and / or a sophisticated tool, any demand could be forecasted with an accuracy should touch 90% or more. Such obsession leads to frustration and demoralizes the entire supply chain staff. The fact of the matter is that one should forecast what is forecast-able and not forecast what is not forecast-able. The demand forecasting is a mean and not an end by itself. In many cases demand forecasting may not be an appropriate mean to an end objective. Trying to improve forecast accuracy beyond a point is self defeating.

It brings out a new paradigm on “non forecast-able” component of demand – how it should be identified and managed. I call this as Demand Management, which is a more strategic topic than the demand forecasting. Let us take a hypothetical example of a product whose sales trends are shown below:

Assuming that the spikes in demand that are not due to any planned event such as pricing or marketing activity, one would attempt to normalize the spikes for generating a forecast for future. The projection will show a nice and smooth line.


However, the demand spikes are a reality and may happen in future. Normally, the supply chain planner would keep as much safety stock to cater to unknown demand spikes. Such an approach is very short sighted and results in high stocks most of the times and stock-out sometimes. It is one of the most common causes of frustration with the demand planning.

Demand management expects a deeper insight into the demand behavior and differential treatment for each kind of behavior. It requires the capability of slicing and dicing of demand data by customer type, geography and seasonality. Continuing with previous example, assume that there are two types of customers for the product i.e. B2C and B2B. Each of the customer type may reflect a different demand behavior, as shown below: This insight into demand behavior opens up many possibilities of demand management. The B2C channel is a clear candidate for demand forecasting using the statistical tools. However, B2B demand behavior shows an unpredictable pattern that may be due to:

1. B2B customers place orders and take deliveries in big lots, to take advantage of bulk price discounts or logistics cost. However, the consumption at their end may not be as fluctuating.
2. The consumption at B2B customers end is unpredictable, however they expect a short response time as and when such demand materialize.

The possible responses to such demand behavior are:
- Collaborative planning with B2B customers, by which the consumption pattern at their end is made visible to the supplying company. This would obviate the need for keeping very high stock all the times and responding to the actual situation.
- Servicing B2B demand through made to order route instead of made to stock.
- Reducing lead time for servicing demand spike by reserving production capacity, stocking of common parts or sub-assemblies, configure to order etc.

By getting an insight and segregating different demand behaviors, it is possible to design supply chain responses that go beyond the benefits of conventional demand planning or forecasting. We have taken one such example but there are numerous possibilities that demand management may throw up.

No comments: