CASE STUDY

Predictive analytics: next generation retail site selection

Faced with the challenge of how to expand its national network significantly in less than three years, one successful South African retailer turned to predictive analytics and discovered a powerful method that has enabled it to evaluate potential sites much faster and more cost-effectively.

Site selection is critical to success in the retail industry. Placing an outlet or store in the wrong location can prove a costly mistake, especially during a tough economic cycle when consumer spending slows and shopping malls and high streets are strewn with outlets that have failed and been forced to close. When every customer is prized, even successful retailers need to use more than demographic profiling to select locations for stores.

Keenly aware of this, a highly successful fast food retailer requested assistance analysing locations and making site selections. Despite enjoying success using demographic profiling and mapping, the retailer wanted to find a quicker, scientific, yet cost-effective way of rapidly expanding its national network. As a result, the team of expert analysts at Knowledge Factory was tasked with developing an efficient and more accurate method for evaluating potential sites.

Model behaviour

The solution was to employ predictive analytics. The process of analysing store performance by applying statistical algorithms to an extensive range of variables, including demographic, financial, property, transport and lifestyle-related factors, predictive analytics takes site selection to the next level. The process doesn’t overload retailers with reams of data, which may be interpreted incorrectly. Rather, it enables them to build highly accurate models of store behaviour, which can then be used to forecast performance.

Once store performance has been accurately modelled, retailers can make quick decisions about prospective sites and identify those that offer the greatest market opportunity and lowest risk without having to generate time-consuming maps and reports on a site-by-site basis.

Predictive analytics models offer numerous other advantages as well. The models can be used to identify gaps and potential pockets of opportunity within an existing network. They can also be used to benchmark the performance of existing stores, helping retailers analyse and replicate the success of top performers across the network and to decide whether to close a non-performing outlet or not.

Developing the models

The team started by conducting a ‘proof of concept’ pilot using a small sample of the retailer’s existing stores in Johannesburg’s East. The purpose of the pilot was to examine the available data to identify those variables that positively or negatively correlate to store revenues. The initial results were encouraging and so the team set about creating models for each category of outlet.

Developing the models that underpin predictive analytics is a complex process. The first step is to collate and link the client’s data with proprietary datasets developed by Knowledge Factory. There are always gaps in the data that need to be built up before the necessary assumptions can be created. For example, the number of businesses within the immediate trade area might suggest a potential level of revenue, but this will be affected by the number and type of employees each business has. As a result, an accurate way of assessing employees needs to be developed as well.

Once the gaps in the data have been closed, the next step is to exhaustively test the variables for correlations and to use these relationships to determine what is influencing revenue. In turn, these can be used to create the formulas that underpin the models. The models are then tested by comparing their predictions against actual store revenues.

Applying the models

The results were excellent. Firstly, using a gap analysis the Knowledge Factory models managed to predict the revenue of 80 percent of the retailer’s existing stores in Gauteng to within 70 percent. The next step was to build a grid map of the province using predicted revenue, which was used to identify and group areas of high potential.

As a result of this process, seven additional sites were immediately identified and moved into the property acquisition process. This method will be rolled out to the other provinces in the near future.

Maximise growth, minimise risk

Given how critical site selection is to retail success, it is no surprise that retailers have continuously searched for new ways of making faster, more precise location decisions. While the use of demographic profiling and mapping has armed retailers with the comprehensive data needed to make accurate decisions about individual sites, predictive analytics represents another significant leap forward. It’s a powerful process that can model and predict store performance, enabling retailers to quickly identify sites that offer immediate or long-term growth and minimal risk.