Individual segments let you identify, study and use specific behavioral patterns in your customers or users. That’s as true in digital as it is in physical people-measurement. Those segments are a powerful tool for analysts. By creating a segment, you can isolate the behaviors around a single type of visit and track how that type of visit is changing over time – including your success rates for a specific type of customer and function. But defining a bunch of individual segments isn’t optimal for every purpose. In particular, if you want to integrate segments into reporting (and you should) there are real challenges around their lack of exclusivity.
Suppose you sit down and build a formal two-tiered or journey segmentation scheme for your location or website. That schema will describe every class of customer and all of the key visit types across the entire set of customers. That segmentation scheme is a powerful tool for helping the business think a LOT better about the business. By tracking the success of each segment by visit type, you dramatically reduce the noise in the KPIs you’re using. Every KPI becomes specific to the visit and customer type it’s measuring. That’s a huge improvement. But you also get a vastly improved view of how important each type of visit/visitor is to the business, how the mix of visits/visitors is changing over time, and how well the business is handling the customer journey at every step.
The first of these (how significant is each segment to business performance) should help steer stakeholders to a better understanding of which customer and visit types matter most. If reporting can do that, it’s adding real value.
By tracking how the distribution of customers across segments is changing, you give stakeholders a great way to separate out the impact of acquisition and exogenous market changes vs. changes to the actual experience (whether online or on location). This is one of the hardest problems to solve in analytics reporting. If you see that May numbers are better than June, it’s nearly always a question whether that change in performance is driven by a change in the population of customers/users or in the quality of the experience delivered. By tracking segment distributions, you give stakeholders a better understanding of how external and acquisition factors are changing (or not), helping stakeholders understand whether changes in the numbers are driven by experience or externals.
Finally, by tracking the performance of cohorts of customers over time by journey step, you give people a unique and incredibly interesting way to track your real performance. If a business is succeeding at more stages of the journey and moving more customers successfully along that journey, then the cohort of customers acquired last month will be ahead of the cohort of customers acquired 3 months before in terms of their 1 month performance. And that lead should hold up as each month elapses. A business that is improving its performance with customers on a cohort adjusted basis is demonstrably getting better at what it does. That’s huge.
But if you try to build a set of segments that capture every important group and journey stage for your customer/user base, you’ll almost certainly find that there is significant overlap. This is particularly true because you will often use highly-specific behavioral cues for certain segments and much more general ones for others. In a retail store, for example, you might define a product-returner segment based on people who go to a particular cash-wrap area as the first part of their visit. But these product returners may also wander into a shopping area (and, of course, may even buy something). If you have a segment designed to pick out product returners and another designed to pick out shoppers, some visitors will fall into both.
You might think this isn’t a big deal, but it creates a number of problems. First, it means that whenever you report out segment totals – whether for visits or purchases or time – you’ll be overcounting. It’s hard to explain why segmented totals add up to more than 100% of shoppers or why they account for more than 100% of successes – and it’s often hard for stakeholders to sort out the resulting overlap.
Worse, this overlap creates a kind of noise in the KPIs – the sort of noise that segmentation was designed to get rid of in the first place. It’s likely true that product return visits have a lower conversion rate than normal shopping visits. So when you include product returners in your shopping segment, you are likely lowering the conversion rate against the KPI. And if the percentage of product returners goes up, it will make it look like the store is performing worse for shoppers – which isn’t the case at all.
This isn’t a White Lotus kind of exclusivity. Segments are exclusive and prioritized not because they are good or bad, high or low value. They are exclusive because they are designed to reflect specific types of customer visit and journey steps.
When creating segments, you often have a implicit hierarchy in mind that’s designed to capture – to the best extent possible – the visit intent. For someone returning a product, that’s likely the primary focus of the visit and the segmentation should reflect that. In fact, Product Return visits may be your highest priority segment even though the actual value of these visits may be very low or even negative. That’s because when you’re returning a product to a store, it’s nearly always the main purpose of your visit.
Ideally, then, when you create a set of segments you should be able to make them mutually exclusive; and mutual exclusivity means that you should be able to set a priority order. If a visit is classified as X, it cannot be a Y or Z. If it is classified as Y, it cannot be a Z.
You can do this within a segment builder by using a NOT of every previous criteria. In other words, you might define a shopper as someone who exceeds X seconds of dwell in a shopping area and did NOT first go to the product return area.
That works, but as you grow out a segment set, it creates unbelievable complexity. By the time you get to the fourth or fifth or eighth segment, you have to create massively complex rules that capture all the NOTs above them.
Instead of forcing users to do that, we’ve created a separate entity called a Segment Set in our DM1 platform. It’s designed to capture a hierarchical ordering of segments and to enforce it automatically. Here’s what the Set Builder looks like:
You can add any individual segments you’ve created and you can add as many individual segments as you need. Then you can order them by the priority of assignment. You don’t have to add any logic to the segments themselves. Finally, you can name the set and control where it should be used (UI, Feed, Playback). These settings complement the ones for the individual segment. If you create an individual segment and flag it for the UI, then add it a Set and flag the Set for the UI, then the individual segment will be available both as a Custom Segment (standalone) and as part of the Set. What’s more, it will usually give different results depending on which of these is used.
Why?
If the segment is part of a Set and is not the highest priority segment in the Set, some of its qualifying visits when used as a standalone segment will NOT qualify when used as a part of a Set since they have already been assigned to a higher-priority segment in the Set. In other words, some of the “shoppers” have become “product returners”.
Building a Segment Set lets you create mutually exclusive segmentations that cover every customer and every journey and ensure that there is never overlap or double counting. That makes Segment Sets particularly valuable for the uses described above (business impact, distribution, cohort performance). Any by making segments and segment sets independent, you get all the modularity and benefits of compact segment definition along with the ability to combine segments and enforce mutual exclusivity when desired.
Typically, I think of Segment Sets as being something an organization will define at a strategic level and use repeatedly across a wide-range of business and reporting contexts. Individual custom segments may be a part of that, but they are also the kind of thing that an analyst may create for a specific project or view of the data. You might have hundreds of individual segments, but you’ll likely only have one or two Segment Sets. Because while exclusivity can be a very good thing, it’s not something you need all the time!
In my next (and final post) in this series, I’ll show how a Segment Set can provide the framework for a fundamentally different kind of KPI report.
