Analysis of location potential and comparison of projected values with the actual values. How can they be improved?

Objective

To identify the best locations for the business, select the best shopping centres to locate the retail outlets. Determine the outlet's potential and understand the reasons behind it.

Context

One of the key challenges for retail businesses is finding the right location and making the most of its potential. Here is a look at how data helps to do that and why nearly benchmark-accurate options are not always a good idea.

Solution

When we are transforming a network or helping to find locations for new outlets, we need to be very clear and reasoned in explaining to the network manager and the client's senior management why we are proposing this solution and not a different one and where the prognosis we are suggesting comes from.

To this end, we describe our two approaches in more detail. The first one is the use of statistical models. Their prediction typically accounts for 3,000+ input variables and pre-established "features", which increases the conclusions' accuracy and predictive value. The main advantage of this approach is its high accuracy. But there are also disadvantages, the main one being that the calculation methodology is hard to interpret. Some parts, such as regression, are easy to interpret and understand, but there are also calculations that take place, so to speak, within the model, and are not easy to understand for humans and even more difficult to translate from the language of programming and higher mathematics and statistics to the business user's language and to find the correct matches in the business dictionary so that all project participants would understand. Often this is simply not possible, and for the business customer, the calculations, including the high-load ones, can become a "black box" producing solutions, the reasoning behind which is unknown.

This is precisely the reason why we also use the second approach. Together with the customer, we compile a set of expert rules, machine statistics advise us which factors influence the result the most, and we assign different weight to different parameters and characteristics depending on the greater or lesser importance to the business in question.

The sum of the points is what separates the locations into best, average and worst. This procedure is the most transparent and logical evaluation and comparison possible. For example, let's look at evaluating the accessibility of a shopping centre. How many public transport stops and routes are there? What proportion of residents can get to the shopping centre without a transfer? If 20% of the city residents can get to the shopping centre, then such shopping centre can get an additional 1 point. If there are no other major landmark shopping centres within a radius of two kilometres, then the centre in question can receive another point. If there are more than 50,000 people living within a radius of 1 km – another point. How many inhabitants and how much traffic per 1 square metre of the shopping centre, for different types of its tenants, and so on. In total, there are usually 10 to 15 key expert rules. The rules are not left at the mercy of the machine; any expert can check them, recalculate, and change parameter weight, priority, order or gradation. Socio-demographic parameters are also included in the calculations. For example, if we want to target a family audience, but the profile of the shopping centre is construction-related, this shopping centre will receive 0 points or even a negative score, so that it can be immediately removed from the list.

Once we have analysed all the factors and built a prognosis, for example, stating the possibility of 10 million in potential revenue, the customer may object: "We have an outlet in this shopping centre, and it makes half as much profit." We analyse the main factors influencing the profits and determine what is missing. The first thing we do is check that there are no errors in the calculations, the source data, or the rules agreed upon for the calculations, and then we collect internal, additional factors.

For example, we approach the location of a particular outlet in a specific shopping centre and see a big difference between the expected and actual financial results. In this case, together with the company management, we analyse the conversion rate of shopping centre visitors to shop visitors, as well as the conversion rate of visitors to customers, and formulate and identify additional characteristics that influence the difference between the prognosis and the fact. Here we are more about providing tools and helping the customer find the cause. The customer can, and often does, do all this without us, but we are needed in the data preparation, mathematics and calculation stage to determine the potential against which the actual result is compared and to structure additional separating factors.

Traffic measurements at a particular point can be used as an additional factor. If its location is “outdoors”, the customer already has these calculations, as our models have calculated and accounted for them during the first stage. If this is an indoor location, such as distance from a grocery shop or a food court, we or a company employee can calculate that. The important thing is that this will leave us with no types of data not covered by the analysis and no possible reasons for inefficiencies and differences between fact and prognosis. This does not just apply to “inefficient” points. From the perspective of a working network, it may be successful, but analysis shows that a combination of factors produces a better result, which means that the potential is not being exploited fully. And this must necessarily be addressed.

In addition to standard scenarios, a joint expert team can develop several plans for positioning an outlet in a particular shopping centre location and assess where the result is better: next door to which brands, on which floor and at what distance from the cinema, escalator, food court, etc. What is the value of such additional expert assessment? Analysing various internal factors after they have been digitised and put into models for calculations will make it possible to formulate a "key to success" and to assess and analyse, in an already automated way, for example, the attractiveness of all shopping centres in the country. This is the most promising way to "run" thousands of options with high accuracy and select the best ones. Any other expert evaluation or trial-and-error method would be more expensive and take years. Here we produce a shopping centre ranking, but not a general one for the market, but a target one, using the adjusted parameters for a particular chain.

Result

When we solve network optimisation problems, we select the most suitable locations, perform predictive traffic analysis and calculate the organisation's financial bottom line, paying attention to all the types of digital data available to us. Competition, total traffic, targeted traffic, demand. Such internal and additional characteristics as staff quality, navigation and advertising. As a result, the customer gets the fullest possible set of the best solutions and an easily adaptable business-wide model for calculating and forecasting revenue, business profitability, depending on various variables.