The purpose of any standard cluster analysis is to break out different segments of your chosen attributes.
To do this kind of analysis effectively for your business, a team gathers all of the data specific to daily store sales at the product level. By compiling this data, we want to understand:
- How store X is different than store Y.
- How store Y is different than store Z.
With some insight, we can start to develop a hypothesis about the kinds of stores that are similar, and discover their shared attributes.
For example, we may have stores that have really good sales numbers during summer, and some that are better in winter. Of course, this could be the result of the types of products each store carries.
But, it could also be the makeup of the customers who are located around their stores. Specifically, if stores are close to points of distribution, we could look at some that draw from a larger trading area than others. That might have implications for the type of marketing tactics we use. And that’s where we start when we complete a cluster analysis.
Let’s look at some possible scenarios.
With locations sprinkled across the nation, each restaurant has its own little unique challenge to overcome. Maybe one wants to increase the size of the lunch crowd, and another wants the average check value on dinner business to go up, while a third location is losing children diners at a faster rate than adults.
Clearly, a single marketing approach will not work for all three distinct situations—we wouldn’t run a “kids eat free” promotion at a location that doesn’t have a problem with child-dining business volume. We would only run that promo in the areas (or individual locations) that need to deploy that specific messaging strategy. So, we define each problem for each restaurant—solvable problems for which we can build a marketing solution—to drive revenue against the problem or opportunity.
Using our Intelligence Platform, we can input the attributes of these three locations, model them across the restaurant chain’s hundreds of locations across the country and evaluate alongside some additional data:
- Day part
- Year-over-year growth
- Product sales from each restaurant
Then, when we have all the data mapped out, we can start to identify the clusters—the similar attributes of different locations.
Continuing with this restaurant example, we might build out a program for restaurant locations that struggle with adult guests during lunchtime, throughout a standard work week. This segment’s members are probably at work, looking for a place to eat lunch, and for some reason they’re not choosing our restaurant. We could develop a plan that specifically targets this segment.
On the flip side, there may be segments that need a boost in weekend dinner guests, which would necessitate a very different approach in messaging, targeting and timing than a workday lunch crowd. Using the data we compiled, we could examine the clusters to determine which locations need what kinds of marketing plans, so we can execute tactics that are specific to locations, and not generalized to fit the entire nationwide chain.
Consumer packaged goods
Let’s say we have a product that’s distributed in Walmart. Each Walmart store performs differently, with their own nuances. Sometimes, however, with a cluster analysis, we can find really interesting similarities that correlate across large segments of sales and store locations. So, this CPG product. We have a segment of stores with great sales numbers, our highest volumes… Why?
It’s clear that our “A” stores are also “A” stores for our product. Oddly enough, our highest performing stores, our “A” stores, are located in our home state—the same state as our headquarters.
Once we can determine and establish this kind of context, we can understand that maybe it’s less of a need to develop a marketing plan to increase sales at other locations, and possibly more just a case of those high-performing stores being exceptional outliers.
Then, we can ask other questions. Do certain stores need more budget? Less? Can we reallocate to other locations?
Let’s say our executives happened to visit stores near HQ, and they see an outstanding sales picture that’s not representative of the rest of the country’s stores.
Now imagine if our executives visited stores near HQ that had the worst sales in the country. The point being that whatever we see locally, is not necessarily reflective of the big picture. We need clustered data points to show us how our stores are not created equal.
Challenges and opportunities for cluster analysis
One of many marketer’s most consistent difficulties is understanding the system of retail points.
With a single store, it’s easy to know what’s going on. It’s easy for the marketing leadership team to extrapolate a problem and have the result be representative in some way for all stores. In reality, if we have 1,000 locations, there could be 900 different problems to solve in each one. All that information is impossible to consume without data, and something to reduce 900 problems into way fewer, solvable segments.
When we look at data with this perspective, we’re able to quantify opportunities all the way down to the store level. This is where clustered datasets really shine—segmentation is all about uncovering consumer insight. From a historical marketing context, this isn’t something marketers commonly consider, but should definitely get into. Once we see our problems and their nuances form clusters, we can start to assign growth potential.
When we start to think about budget, planning and media scheduling, this can be a really strong prioritization mechanism. It also makes the case for why marketing budget needs to be what it is, and why it’s allocated to certain areas. We’re no longer just guessing what might happen. We’re able to say, with informed, increased confidence and certainty, that we expect a certain amount of money from specific segments of stores in a set period of time. That’s incredibly powerful and valuable information.