By Howard Roddie, FuturMaster Senior Supply Chain Consultant (@HowardRoddie)
A few years ago I sat next to a very loud recruitment executive on a train. She received a call regarding a “forecaster” for a food company. The whole carriage soon knew how easy this person would be to find as it wasn’t “rocket science”.
This isn’t the place to compare the maths and science of jet propulsion with that of demand planning, you will be pleased to know, but it is worth reflecting on some of the complexities we now face in creating realistic demand plans for FMCG, or what recruitment executives probably call “basic products”.
Not long ago, we had a single stream of historical data to create statistical models. Typically this represented the sale from producer to a range of business customers. Now we have a huge array of data we can use to fine tune this “sell in” data.
The most common form of extra data is “sell out” from our business customers to the next level of customer. It would seem logical that this data should give a much better demand signal that our sell in data as it is closer to the demand of the consumer. But this would oversimplify the picture.
If we sell via a network of distributors, we will be very lucky if we have a monopoly of supply to our distributors, and they, in turn have a monopoly of supply to their customers. I have actually seen this scenario myself and in the rare case it exists, sell out data is very useful both for statistical modelling and short term replenishment. Where we have actual and target stock levels for distributors, these can be built into our calculations to improve the accuracy of actual demand the producer needs to fulfill. Scenarios like this are rare. Most of the time our distributors have alternative suppliers. They buy and sell from each other. They sell to each others customers. Where this happens, their sell out data is likely to be unreliable and unexplained and their stock levels and target stock levels are likely to fluctuate with little logic.
For a greater level of detail we can always rely on EPOS data. This is the consumer making the final demand decision. What could be better? Certainly during promotions and launches, this data is crucial as early returns can tell us if we should increase or decrease production before we hit stock issues. However, for day to day basics, the strength of the signal may be too weak to act upon. We also have to consider stock levels in store and within the various other layers of the retailers own networks. Not to forget that they may sometimes be restocked via one of our distributors.
Sometimes, the demand signal from the consumer is too late too. Warm weather related sales may depend on a single pleasant day. By the time the sale is made the shelf is empty. Of course, we can sense the demand with weather forecast data, but which weather do we use?
Beyond EPOS, retailers now have access to a good proportion of their customers shopping lists several days in advance in the form of online orders for delivery or collection. This data ought to be used to improve overall short term forecast accuracy. Maybe, we should start calling it replenishment accuracy?
We’ve only scratched the surface of how we can best use the data available for calculating demand, so let’s bring it all back to rocket science. Nobel laureate Peter Higgs once went to bed early as a discussion about the “theory of everything” took place. This led to a failure of the physics community to make an early connection between his work and the standard theory. So a failure in the collaboration process meant that the available data didn’t get considered together, causing a delay. In FMCG, unlike rocket science, such a delay would be unacceptable!