Incorporating science into your Site Selection
A site selection mistake costs the business in terms of being a failure in many ways – bad publicity, disgruntled franchisees and loss of confidence in the system. As the franchisor, this can become the ultimate problem, namely a legal case aimed at showing negligence on the Director’s behalf in the site selection process.
There have been a litany of legal cases over the last 15 years, and whether you win or lose, it will have created a huge amount of stress on the business and the senior executives along the way. Cases like Billy Baxter (sent the group into liquidation), Lenards and Muffin Break all proved extremely costly and detrimental to the business, especially in these days of social media.
In an effort to minimise risk, I believe you should be asking yourself:
1. Do you have a process for selecting sites?
2. Do you have any quantitative analysis to support this?
3. Are there any sales prediction tools you use to assist in this decision?
Levels of research available to a Corporation.
The more sites you have operating, the better the information and ability to have predictive tools to use in site selection should be. I have been a great advocate of process in site selection, probably built into me from my 20 years of oil industry experience.
I believe all businesses should be able to demonstrate a process they use in site selection, which probably covers as a minimum:
• Initial site identification
• Suitability from a demographic view
• Physical site suitability
• Economic feasibility
• Final approval.
In the oil industry, we had records going back 80 years that show how old depots had been agreed upon, and over 60 years (solo marketing started in the 1950’s) to show the justification on how a petrol station was selected and approved.
Low site numbers (5 – 20 stores)
If you have 5 – 20 stores, you may be able to logically assess why the best sites are doing well, and the lower performers are not so successful. At this level of stores, a company should be able to model the demographics around the stores, and see if the better stores match your perception of the customers in terms of demographic of the area.
A Check Chart can be formulated using a combination of customer perceptions, plus show the statistical results of the demographics of the best performing stores.
Medium store numbers (20 – 50 stores)
If you have 20 – 50 stores, maybe a percentage of those in the country, then you should have a much better Check Chart based on a larger sample size. At these numbers you can undertake a regression modelling process to identify the drivers of the business, and then use some flexible logic to insert other variables you are confident assist in making better stores.
Large networks (50 - 150 stores)
For established large networks (50 - 150 stores); you can build a sales prediction model based on the sales being achieved by the network using regression modelling. This is done by a ‘Market Analysis’ where:
• All existing stores are surveyed. The survey can incorporate issues such as size of building, number of counters and tills, seating (if a food business), access, store visibility, signage visibility, parking spaces and convenience, nearest neighbours and other business generators, and many other items. A survey like this also produces digital photographs showing all aspects of the site, and gives a benchmark for comparison of stores and standards.
• Around 400 demographic variables are extracted for each store in the network either at different radii, by sales territories and/or by catchment areas.
• Competition and generators are then measured to determine which categories of business have positive or negative effects on sales. Possible distance effects, wherein the competitive or generative effect is only active within certain radii, are also examined.
• ‘Exposure’ is approximated based on traffic counts, signage and visibility, and a measure of pedestrian volume and flow.
• Sales information for all applicable outlets completes the dataset, plus any internal operations measures where available.
Statisticians then go to work to look for the best variables that explain the sales that are being achieved. Sales Prediction models typically incorporate variables from each of the above categories (survey data, demographics, competition/generators, exposure, and internal data). Though no guarantee can be given of individual results, most analytical companies would expect to obtain models that can be said to be 70 – 80 per cent accurate. The more consistent a brand is, the more accurate you would expect the results to be.
The typical graph displays all stores in a network, each point showing where that store sits in comparing the actual vs. predicted sales.
Networks of 150+ stores – Neural Networks
The large networks we work with usually extend past regression modelling, into the use of Neural Networks. This is often called artificial intelligence and works on looking for patterns in the data that allow it to combine and create new variables that give a better result than can be achieved from pure regression modelling.
One of our clients had over 200 stores all in strip shopping centres, and all selling the same product. Regression modelling gave us 78 per cent accuracy, whereas the neural network gave us an 86 per cent level of accuracy.
Applying the science once developed
The sales prediction model aims at predicting the sales on mature or established sites that have been open at least one year. In the oil industry, the sales of most service stations went through a ‘ramp up’ of 85 per cent year 1, 92 per cent year 2 and reached their full sales potential (100 per cent) in year 3. Different businesses will have different ramp ups. In the fast food industry, we have seen cases where with big opening promotions, some stores never again reach the sales level achieved in the first four weeks! Once a Sales Prediction model is built, any new sites being considered can be run through the model to give a sales prediction at maturity. This may be done by a consultant, or internally if the company has all the necessary resources. Our experience is most companies tend to leave that with the consultants as:
• They do not have the internal statistical expertise to run the models.
• They do not have all the data necessary. Often the model includes some variables from Census 2011 and ABS Business Counts.
• Staff changes, or staff are busy or on holidays, and they cannot keep up a 12 month service, but external consultants can provide these services all year round.
The Sales Prediction Modelling then becomes an integral part of the approval process that a company undertakes. It should not be seen as the only part of the decision, as exceptions do occur, however it should be seen as a good ‘flag’ as to what is expected.
Though this is never 100 per cent accurate, it should allow the company to have a set of ranges that guide further decisions in the process. For example, a sales prediction in a range below the Network Average would provide a strong warning against proceeding, and special circumstances would need to be demonstrated to achieve approval. On the other hand, if a new store’s prediction is in the top 25 per cent of network average, then a higher level of comfort in approving the proposal can be felt.
Strategic Network Planning
With a sales prediction tool in place, and the knowledge of what makes for your good stores compared to the poor stores, you are now in a position to undertake a Strategic Network Planning (SNP) process to formulate how you can expand in the future.
SNP is all about prioritising areas and opportunities, or having a plan for the Network Development Team to follow. This leads to lists of shopping centres, strips, CBD areas and other opportunities, prioritised so that there is a clear path to follow.
Use logic and science to make better site decisions. Implementing a Site Selection and a Strategic Network Planning process can be very good insurance, and should help in making better decisions for the long term. As a Company Director, you need to know there is methodology in store and site selection if you have any involvement with retailing, be it operating a company owned chain of stores, or a franchise network. Investing in having a good process for retail site selection is a good insurance policy, and should minimise risk.
Peter Buckingham is the Managing Director of Spectrum Analysis Australia Pty Ltd, a Melbourne based geodemographic consultancy. Spectrum specialises in assisting clients with analysis and decisions relating to store and site location using various scientific and statistical techniques.
Peter’s background was in the oil industry, with a strong focus in property issues, both nationally and internationally.
T: 03 9830 0077