Problems with forecasts
My friend S works in management of a shop in Exeter. It is one of a national chain, selling consumer goods. Earlier this year, the chain opened a second branch in Exeter (500metres away, as the crow flies). I asked him how the second shop was going, and what the effect had been on "his" branch.
He said that like for like sales per week had fallen by about 15%, but the sales at the second shop were running at about 80% of the sales at the first. So, total takings in the two shops combined were significantly higher than in the first store. The expansion seemed to be successful. But, there was a problem.
Head office sets each branch a target for weekly sales. It is based on the corresponding time period a year earlier, together with an underlying trend upwards. And the first branch is failing to meet its targets, because some of their sales have been shifted to the second branch.
It looks as if head office has a forecasting model which is overlooking the launch of the second branch. And this suggests that the forecasting model that head office is using is a naive one - certainly S thinks so. It is based on the assumption that tomorrow will be just like today, with - perhaps - some growth. There are times when this model is as good as you can get, especially when you don't understand the underlying cause of the time series generating the data. But, there is a good explanation for changes in the data about sales at the first shop. And the evidence ("Fallen by about 15%") gives a measure of the change. So it would be easy to adjust the targets to reflect the consequences of the decision to open the second branch.
It is a warning that forecasting models and tools should not be based on simple "off the shelf" software. Human beings need to be involved, to make sure that what is being modelled has not been changed in a significant way.
A newspaper report today tells of another business which is suffering from forecasts which don't appear to take all the relevant information into account. A well-known company which hires our men's formal clothes has reported drops in its sales. And the drop is due, they say with hindsight, to the decline in the number of couples getting married. Maybe this could have been included in the forecasting model for their work?
He said that like for like sales per week had fallen by about 15%, but the sales at the second shop were running at about 80% of the sales at the first. So, total takings in the two shops combined were significantly higher than in the first store. The expansion seemed to be successful. But, there was a problem.
Head office sets each branch a target for weekly sales. It is based on the corresponding time period a year earlier, together with an underlying trend upwards. And the first branch is failing to meet its targets, because some of their sales have been shifted to the second branch.
It looks as if head office has a forecasting model which is overlooking the launch of the second branch. And this suggests that the forecasting model that head office is using is a naive one - certainly S thinks so. It is based on the assumption that tomorrow will be just like today, with - perhaps - some growth. There are times when this model is as good as you can get, especially when you don't understand the underlying cause of the time series generating the data. But, there is a good explanation for changes in the data about sales at the first shop. And the evidence ("Fallen by about 15%") gives a measure of the change. So it would be easy to adjust the targets to reflect the consequences of the decision to open the second branch.
It is a warning that forecasting models and tools should not be based on simple "off the shelf" software. Human beings need to be involved, to make sure that what is being modelled has not been changed in a significant way.
A newspaper report today tells of another business which is suffering from forecasts which don't appear to take all the relevant information into account. A well-known company which hires our men's formal clothes has reported drops in its sales. And the drop is due, they say with hindsight, to the decline in the number of couples getting married. Maybe this could have been included in the forecasting model for their work?
Comments
Post a Comment