Why segmentation is making a comeback
Much of what we thought we knew about their customers has been ‘flipped on its head.’ The pandemic has propelled many companies to revisit their existing personas and commercialization approaches through segmentation. At this point, you may be asking:
How can we understand our customers better?
How can we make our business process better or more efficient?
What caveats should we consider when operationalizing our segmentation/model?
In this interview, Julien Theys, Agilytic's Managing Partner, draws from his extensive experience with various segmentation projects to break down how data impacts commercial success with segmentation.
“Segmentation is useful to estimate and identify the different customer groups and identify their most relevant needs.” - Julien
Would you agree that segmentation is making a comeback?
Julien: Let’s use retail as a starting point: there is a move towards more personalized content. Companies can better understand and identify purchase patterns down to smaller, more granular segments. For instance, it is now possible to optimize promotions by the food product sub-category.
We’ve also seen this beyond retail, such as the cultural sector that is eager to bring customers, old and new, back to enjoying cinema, theater, and live music. We can segment by transaction data, purchase behavior, high revenue potential, and high-frequency purchases.
It is also challenging to consider COVID behavior in segmentation (i.e., switching from offline to online channels, the smaller selection available, changing purchase behavior, and buying more at less frequent trips). We have seen a massive increase in digital campaigns targeting more people and prospects, and segmentation helps optimize these campaigns.
What is the first step for mid-size retailers if they feel they don’t know their customers anymore?
Julien: The first step is to have a cohesive approach to blending the “traditional” sociodemographic data (such as age, gender, postcode) and behavioral data (purchase patterns) they already have.
Data doesn’t have to be exhaustive, granular, or 100% clean to produce trustworthy insights quickly!
There’s no point in waiting for those elusive perfect data sets.
Why is segmentation as a result of the crisis a crucial strategic element for marketing/commercial teams to revisit?
Julien: Because much has changed, a company’s idea of who their customers are may have also dramatically changed. Teams need to know their customer segments and critical indicators and dimensions, especially following this crisis.
For example, when looking at interactions, teams can discover new transaction patterns (who, how much, when, which products). Segmentation highlights potentially overlooked behaviors.
Other ‘Nice-to-have’ items include contacts or other interactions (e.g., subscribed to the newsletter, interacting with the enterprise online).
We also often look at sociodemographic such as age, region, profession, household type. For example, based on geographic information, we can use open data to provide additional insights: average revenues, unemployment rate, age, etc.
It’s not just in B2C. In B2B environments, we build onto interactional and behavioral data with company-specific (firmographics) to see what business-specific factors matter more.
In principle, the more granularity you can have on your customer, the better!
Why is it crucial to narrow the focus to a number of segments with the best revenue/demand-generation response?
Julien: Segmentation allows us to identify the customers with higher revenue potential. Of course, these customers are the most prioritized, but they are not necessarily the only focus of marketing campaigns.
Segmentation can also highlight:
A “mid-revenues” segment with the potential to move to high-spender segments (often called “up-selling”)
Potential prospects: if we identify we are popular with a given segment (whether it’s the one we intended to attract or not), it’s easier to aim for “lookalike” potential customers in digital campaigns.
How does data play a role in segmentation? What are common methods you’ve used in projects?
Julien: Simple, straightforward methodologies can generate a lot of quick wins already. But whatever approach we take, it needs to be actionable.
It’s always good to start with a quick descriptive analysis because it can help us identify concrete clusters in an uncomplicated way.
Then, for instance, here is a classical approach:
Principal Component Analysis to reduce the number of variables and improve the clustering
Silhouette and/or elbow method to find the optimal number of clusters
K-means to find and generate the segments
Description of segments : descriptive analysis of segments based on relevant data (e.g., sociodemographic + main trends on transaction behavior)
A predictive model to identify which variables distinguish each segment from the other ones.
Can you describe why our approach in achieving certain milestones can be valuable for a segmentation project?
Julien: Really, what is essential is the before and after.
That means before, we’re making sure there is adequate validation of data. Healthy data from the correct sources helps to target customers, see trends, and help us orientate as data scientists what variables should be included.
After, what you do with your newly-found segmentation knowledge is essential. After a segmentation project, we take a lot of time, much more than building a model, to describe the model and how our client can use it. That includes explaining how the model brings value in the present and the future. For instance, we may recommend for the next step to do A/B testing.
