I was recently wandering down a bridlepath when I came across a sign warning of a Sat Nav problem (see photo). It reminded me of watching the TV programme, Billion Dollar Chicken Shop about KFC in the UK and recent store closure decisions at Morrisons. They all illustrate the problems of believing that mathematical models do not require us to think. There is a danger that the knowledge of the archetypal old property man (inevitably it was a man and many analysts found his views statistically unfounded and somewhat irritating) is missing from the site selection process.
The sign suggests that some Sat Navs are directing delivery drivers, using a postcode destination, to houses that of course can’t be accessed through a farmyard. Knowledge of the area or a detailed map would show that a different route should have been chosen to the relevant house. The Sat Nav model was inaccurate in predicting the optimum route.
Some analysts might say, ah well the model should be improved in terms of its input data. I’d suggest that it means that model users should always be prepared to use on-the-ground evidence to test the model results and sometimes be ready to dispute the model prediction. Have the proponents of using models for decision making become so convincing that businesses think that all location decisions can be left to the ‘machine’?
The KFC TV programme raised the issue of why the Denton Rock outlet was performing badly compared to the successful Cheshire Oaks outlet. As the construction company describes it, “the landmark Denton Rock Tower cannot be missed as you approach the site from any direction, giving KFC a very prominent location (my italics) for yet another of their successful drive through projects of 2013″. Prominence does not always deliver customers.
The staff at Denton Rock came up with the revelation that, although there are many people driving by on the M60 and the other roads, not enough people want to get off on a busy roundabout. As you might say, not rocket science. However I suspect the ‘rocket science’ models will have led the decision based on traffic flows and population – a model would probably suggest an excellent drive through site performance. Somewhat differently, the Cheshire Oaks outlet has a captive shopper population ready to take a break in a KFC despite lower traffic flows.
Similarly, I think the closures of some M Locals show the dangers of a rapid roll-out with inadequate knowledge of local areas. Whilst models may have suggested a sizeable local population with the right demographics, I’d hazard a guess that the siting of some stores was such that the local market would never be adequately captured from that specific address. As The Telegraph reported, the decisions by the new management, “as well as closing stores, Morrisons has halted the opening of new M Locals. The company said it will review the ‘proposition and site selection criteria’ of its convenience stores”.
Statistical models have played a major role in improving site selection in the retail sector but will not accurately predict performance in all cases. Beware of the ‘snake oil’ salesman who suggests otherwise. It’s time for analysts (in-house or external) and their retailer clients to ensure that implementation procedures are in place to include ‘sense-checking’ processes. Model results should be critically assessed on individual sites in the knowledge that predictions will not be 100% accurate. Essentially, does the model make sense on each individual location. If such processes are not in place there is a real danger that like the users of the inaccurate Sat Nav, retailers will be taking the road to nowhere.