14 Sep 2018 Shaping Effective Segmentations: Some Key Principles
I still remember the first time I saw a colleague pitch a segmentation solution to a client. Watching that, and hearing mystical (in those days) words like “algorithm”, felt a little like witnessing sorcery, and was a transformational experience for me in terms of the research I wanted to be doing. I’ve spent a lot of my time since then on segmentation research, and played the role of statistician, methodological consultant, salesman and account lead.
I joined Decidedly (formerly FreshMinds) three months ago. In that time, I’ve worked on and seen some brilliant, innovative projects. We blend traditional and innovative research to get closer to consumers – and segmentation research is a hugely important piece of our toolkit.
In this new role, and previous roles, I’ve had a lot of conversations with clients and colleagues about segmentation research. In these discussions many of the same themes have cropped up again and again. So, I thought I’d put pen to paper to reflect on some key principles that underpin planning, designing and implementing a segmentation study. Segmentation is a huge topic and this does little more than scratch the surface. If you’d like to know more about our approach to segmentation research, we’d love to hear from you.
1. You Don’t Find Segments, You Create Them
Research briefs often say “we want to know how the market is segmented”. Usually, however, the right question to ask is “how should we segment the market?”. These are two very different questions. The distinction between them has major implications on how to design, analyse and position segmentation research.
The idea of allowing subjectivity into a segmentation makes a lot of people feel uncomfortable. These are big studies and thinking of them as providing an objective truth is naturally reassuring when making that investment. But there is no such thing as one single, objective, truthful way to segment a market.
As such it’s critical to be thinking about the end outcome from the word go – specifically, what decisions will the segmentation support, and what commercial applications are most important. A segmentation that is great for helping to shape brand positioning may be less suitable for developing customer messaging or for product optimisation.
These needs may often be in conflict. It’s important to prioritise them in order to craft segments that fully meet the main objectives, rather than dilute the segmentation by attempting to cater to too many conflicting demands.
There are dozens of decisions you need to make across the segmentation. And all of these are about working towards a structure that will work for your specific objectives. The best mathematical model in the world will fall short if we don’t have an idea of what we’re aiming at.
A useful trick is visualising some segments early in the process, at the scoping and even proposal stages. Whilst you don’t want to pre-ordain all your segments, getting a flavour of a useful segment or two at this point will help clarify thinking and can enable stakeholders to make the necessary trade-offs at the right time.
2. Clearly Agree What Success Looks Like from the Beginning
I’d encourage everyone involved in a segmentation study to agree definitive success criteria at the outset – a set of rules where you need segments to differ from one another to make meaningful decisions based on your segmentation objectives. Set them in stone, before anyone even thinks about a discussion guide or a questionnaire or a data set.
When you’re designing and evaluating a segmentation solution, have these up on the wall. If you’re not constantly referring to these success criteria there are several possible negative outcomes, chiefly:
- The agency spends hundreds of hours experimenting on solutions which drift away from the core objectives
- The segmentation is accepted just because it looks sensible and intuitive at a descriptive level
- Clients are enticed into agreeing a solution because it’s spun well, has glossy pictures or poses as the one single truth
We’ll often need to cater for a range of stakeholders’ needs when designing a segmentation study. The more we can incorporate this range of needs the better. But they must be distilled into agreed success criteria to avoid conflict down the line. It’s important not to underestimate the amount of stakeholder management this can require.
3. Be Wary of Oversimplification
Segmentation is always about simplification, but oversimplification is a major reason why segmentation studies can fail.
Sometimes clients get asked at the kick-off stage: “how many segments do you think you can handle in your business?” Almost without fail the answer is six.
Six segments to describe the combination of rational and emotional drivers behind, for example people’s holiday choices is not very many at all.
When you oversimplify (or under-segment, to coin a phrase) your segments will, by definition, become more similar – and may be defined by relatively minor skews. You might find that 30% of your ‘Innovation Seekers’ fear change as much as your average person. There just aren’t enough segments to go around without stirring up some very muddy waters. And the implications of this on segment clarity, accurate opportunity sizing and ROI estimates are enormous.
If a client feels limited to relatively few segments from an operational perspective (and there are a lot of good reasons why this may be the case), then don’t focus on more than that in terms of detailed collateral, recommendations and so on. But that doesn’t mean you always need to cram 100% of the market into six segments.
4. An Arm’s-Length Analyst is a Recipe for Disaster
I see many proposals, creds pieces and website blurbs where agencies talk about “integrated analytics teams”. What this often means is that a senior analytics consultant will come to a few meetings, while the actual analyst sits in another office running a conjoint model for another client while he or she waits for the segmentation data to come in.
If you go into a segmentation believing everything I’ve said above and then, at the back end, give the data to a statistician who’s had minimal involvement in the broader process, you’re likely to run into problems – no matter how skilled they are.
From a more technical perspective, advances in computing power and the ever-increasing proliferation of more sophisticated analytical techniques (including machine learning tools) has changed the narrative about how segmentation data is physically analysed. I’m not disputing the statistical power of these tools. But I would caution against these being necessarily good just because they’re cutting-edge. These tools tend to lead to a “black box” mentality and reinforce a misplaced belief in an objectively right answer.
More traditional cluster analysis techniques allow for a great degree of flexibility and allow the analyst to play a more active role in shaping the outcome, encouraging a happy medium between a solution being algorithm-led and analyst-led. In short, the analyst is more important than the software or the techniques they’re using to analyse the data.
5. Get Back to Basics
It’s not particularly glamorous to talk about process. However, the success of any study hinges on the data we create. And as the industry moves towards cheaper, faster solutions, it’s also slightly moved away from making sure the basics are right.
Always run a live test. You don’t have to do a full-scale pilot, but at least pause fieldwork and have a look at the data for a day or two. You’ll know very quickly whether you’re going to be able to achieve what you set out to. If you find your data is working as it needs to, you’ve gained peace of mind for the price of a couple of days. If you find it isn’t, you can make the necessary changes, saving a fortune and months of time.
And if it’s a multi-country study, don’t skimp on translations. Go old-school with proper blind back-translations. Yes, it will cost a little more, but when your French respondents think that “let off steam” means putting the kettle on, they’re not going to appear in a ‘Party Hard’ segment – no matter how much they drink.