Why I Like Analog Models

Recently, I’ve spent a lot of time discussing why analog models aren’t given the respect they deserve and why I really like them.

This has made for some engaging conversation at places like the ICSC Research Connections Conference, less so at Christmas dinner.

I’d like to walk you through my thinking on why I like analog models and hopefully elicit some discussion on the pros and cons of analog models.

For those of you who aren’t familiar, analog models use the trade area and market characteristics of a given site to compare against your existing stores and find the stores that most closely match the site in question. The output of analog model is a list of your existing stores that most closely resemble the site in question along with a score of how closely they match. This can provide useful insight into the viability of the site.

In analog models it is crucial that the demographic and market characteristics being matched are indicative of something important to the viability of the store. This is usually sales. An analyst or statistician determines which factors are most relevant to driving sales and uses these to match the potential store location against the existing stores. Match scores are assigned to indicate how closely the store matches the site.

Before I dig into what I like about analog models, I’d like to first look at what people don’t like about them.  I typically hear three arguments against them.

  1. They don’t directly provide a sales forecast
  2. They don’t capture outliers well; that is, stores that will perform far, far above or below average
  3. You need to have a lot of stores in order to create an effective analog model

Argument 1 is a statement of fact. I personally believe that argument 2 is correct as well though I can’t prove it to you. Argument 3 is also correct though the definition of “a lot” depends on how geographically diverse your stores are.  I agree with all three issues but I like analog models anyway. Why?

There are a number of reasons and I’ll lay them out in no particular order.

  1. Analog models are easier and cheaper to build than sales forecasting models
    Analog models require a statistical analysis of your existing stores to determine which trade area characteristics like income, population density, competition, etc. (“independent variables”) drive sales and/or any other trait you care about.  But they don’t need to actually forecast sales so this process is simpler than a full-blown sales model.
  2. Analog models don’t require updating as frequently as sales forecasting models
    When you build a sales forecasting model, you not only rely on the assessment of which independent variables are important, you need to decide exactly how important they are. This means that if anything changes about the way a given independent variable impacts sales, your model requires recalibration or it will lose accuracy.
    But an analog model is effectively self-updating (up to a point) and can more easily withstand changes to the way your customers shop your brand.  As long as the independent variables that drive sales don’t actually change (i.e. household income becomes irrelevant and education level becomes crucial) then you probably won’t require a model update.
    For example, if customer behavior changes such that the importance of average household income changes (in tech speak – the coefficient for average household income in your regression model changes), this change is – by definition – reflected in all your existing stores because their performance will change as a result of the changes in your customers’ behavior.  When you run the analog model, it will still show you the most similar stores and the impact of the changes in customer behavior will be implicitly reflected in the performance of those stores.
  3. Analog models give more insights into a new store
    Though analog models don’t produce a sales forecast directly, they can often be used to estimate the potential sales of a site by looking at the sales of the analog stores.  These can be seen as a range of potential sales estimates.  In addition, if the analog model is properly defined, analog stores can provide insight into things like merchandising, pricing, and loss prevention.  By looking at the analog stores and how they eventually were merchandised, or which pricing model worked best, or how severe shrink is can allow new stores can be designed out of the box to match the needs of the market.  This is very powerful and can provide astonishing ROI.
  4. Analog models are easier to explain to senior management
    While most management at retail companies accepts the concept of a sales forecasting model, they are often uncomfortable with not understanding how it works.  Nothing is easier to explain than an analog model. “These stores have similar population density, income, and competition to the site.”  Management loves it! And since you’ve made it understandable, you look smart.
  5. Analog models can be put into the hands of real estate professionals
    The output of an analog model is a list of existing stores.  This is something that real estate professionals can make good use of in order to focus their attention on more viable sites. In fact, analog models can really help offset one of the biggest challenges that real estate professionals face when assessing opportunities, namely the overall market.  They are typically excellent at assessing sites and site characteristics but it can be more challenging to understand the larger market in the context of store potential.  Analogs can give them a powerful tool to address this limitation, which helps you engage their massive pattern recognition skills at yet another level.
  6. Analog models can help validate the sales forecasting models
    The stores that the analog model selects have sales associated with them.  If the sales of the selected stores are drastically different than the sales forecasting model, the results of the forecasting model should be called into question.

If you’ve read this blog before, you’ve probably heard me rail about the limitations of sales forecasting models.  Analog models are no exception.  They are not perfect by any means.  But they are a lot more useful and powerful than people realize.  Plus, it’s not like you need to choose between a sales forecasting model and an analog model.  You can have both without spending a lot more than you would on a sales forecasting model.

The bottom line is that, if you’re in a position to implement an analog model, there is a lot to be gained by doing so.  The ROI will be high, senior management will think you’re a genius, and it’ll give you something to talk about with your extended family at the holidays.


Joe Rando