There was a post on one of the Linked-in Groups that got my attention today. How do you reduce demand volatility using better forecasting?
Forecast accuracy and Demand volatility are two different things, though there is a strong relationship between accuracy and volatility in practice. For the purposes of this discussion, I want to use just common man terminology.
Demand volatility does NOT necessarily mean the demand is unpredictable. Demand volatility also does not mean the demand is an ugly scatter of points across the map.
For example, the calculated demand volatility for a product could be the result of a highly seasonal demand profile. Such highly seasonal demand can look very pretty on a graph and could be easily predictable by what I call as an extrapolation by hand-drawing.
In summary, volatility is not necessarily bad – growth causes volatility, so do seasonal and cyclical swings. There could be other reasons – promotions and phase in, phase out, bonus packs, price actions, just plain old management actions to make the quarter.
Corporate management can try to stay away from some sub-optimal business practices and hence cause a reduction in demand volatility. They can promote so customers may buy through out the year. But can we temper seasonality and growth? Do we want to?
Life is essentially seasonal and we have to live with it. Christmas does not occur every month of the year! People like to go to the beach on sunny days, and more sunny days occur during the summer.
Forecast accuracy is an expression of how well one can predict the actual demand – volatile or not. You may be able to create a very accurate forecast from a volatile demand distribution. This can be done by good modeling and diligent forecasting.
How ever there is another distribution that the forecaster needs to consider. The distribution of the actual outcomes around the forecast — if the forecaster can say his forecast has at most 5% upside and a 5% downside, then hats off! However if the plausible outcomes are widely varying, then more needs to be done to lower the expected error:
- Model and forecast at the right level and use proportioning techniques
- introduce more magic in modeling (however you do it, software or otherwise)
- collect that market intelligence only foretellers can come up with
- discipline your corporate management to reduce demand volatility (hopefully without losing your job)
- S&OP process so one hand knows what the other hand is doing and reduce the bias in the forecast coming from functional groups.
The last point is more important. More seasonal products suffer from more game playing before the season, so do scarce promotional packs. Similarly, long lead-time products also cause more game playing. You can expect more upward forecast bias in these products as well.
In conclusion, improving forecast accuracy involves doing a variety of things and reducing or mitigating volatility is just one of those things.
