Question

Dear Dr. Jain,

I work for a large multinational FMCG – it is considered a group norm to use a high level, statistically modeled forecasting tool for our demand forecasting. However, I believe the tool we have for this is old fashioned (as is the company’s thinking).

Within the section of the business I work in, we use a more granular forecast process (surrounding big retailers, customer implants, dedicated business planners etc.) but we are not happy with the overall result (either at customer level or top line). Let’s say in this model we have 8 customers at 80% of our total volume.

The top down decision is to implement the high-level forecasting tool and work on an aggregated forecast for all of these customers combined.

I am struggling to see that this is the correct and modern-thinking approach. Everything I have learned says that going to customer detail (when you have said detail) is the correct approach when dealing with sophisticated customers with their own promotional plans etc. But I am struggling to convince my peers.

What do you think on this point? Is there any modern best practice I can refer to?

Thanks,

Chris,
Large UK based FMCG company

Answer

Dear Chris,

I agree with you that we should forecast by customer, particularly where a large percentage of sales come from just a few customers. Forecasts are likely to improve if we incorporate their plans, especially their market plans, into the forecasting process. I am not sure whether your company has tried enough. Maybe you would like to see whether or not customer-based forecasts are better. If customer-based forecasting does not give the accuracy you want, then we have no choice other than to forecast at an aggregate level, which your company is currently doing. Forecasts tend to improve when forecasted at an aggregate level.

Better forecasting tools always help. But if you are thinking about forecasting tools only in terms of sophistication of forecasting models, then I am not sure. Forecasting is simply matching the data pattern with the pattern that model captures. With the right marriage between the two, we can have the best forecasts. It is not unusual for a data pattern to match with the pattern that a simple model captures. Among three different types of forecasting models (Time Series, Cause and Effect and Judgement), Time Series models are the easiest and most often used. Based on the recent IBF survey, 62% of the companies use Time Series models. Within Time Series models, 56% of them use the simplest models, which are, Averages and Trend. The best approach, therefore, should be to start with simple models. If they don’t give the accuracy you need, then move to more sophisticated ones. 

I hope this helps.

Dr. Chaman Jain

St. John’s University