Nick is Head of Data Science at Evolve and is a master of new research technologies including text analytics and AI.
In market research, segmentation projects are some of the most rewarding projects.
I can particularly remember one such project involving a workshop with more than 50 client staff to present a new segmentation. We had tables littered with props, butcher’s paper and markers. The workshop was a great engagement exercise. It enabled connection to the research and formation of their own ideas about the segmentation. I have seen some projects follow segmentation analysis with focus groups. It is amazing to see a room full of ‘segment A’ and immediately see the contrasts to ‘segment B’ in the next group.
Other segmentations I’ve worked on have identified high value customers and provided a direct line to better return on investment. Being there to see clients have the ‘ah-ha’ moment when they intuit an understanding of the target group is very rewarding.
If you’re new to segmentation work, this article seeks to provide an overview. We will discuss what segmentation is, why we segment, and describe the phases of a segmentation project.
In marketing, we conduct segmentation to identify a target group for our marketing efforts. By trying to win with a target group, rather than with everyone, we’re more likely to get better return on investment. We can win through targeted media buying, promotional offers and communications tailored for the target group. Segmentations are particularly important for challenger brands, competitive markets and B2C in order take a piece of the market through a niche strategy – that is, serving a smaller part of the market with a very specific offer that meets very specific needs.
There are also operational type segmentations, where we look to identify high value customers. We can then prioritise and grow the relationship with high value customers, while attempting to reduce the costs of doing business with low value customers. Marketing segmentations are more common in retail, B2C and other mass market businesses. Operational segmentations are more common for enterprises with existing customer relationships such as finance.
Usually segmentation is a quantitative analysis – that is- it is developed using large survey samples of consumers or customers. As such, questionnaire design is very important to the success of the project. Segmentation questionnaires tend to be exhaustive, covering many topics including purchase and media habits, attitudes, behaviours and demographics.
There are many approaches to segmentation, but popular ones include demographic, behavioural or attitudinal segmentation.
For a demographic segmentation, we may define our target by age range and income. For behavioural, we may define our targets by recency, value and quantity of purchases. For attitudinal, we try to understand motivations and customer preferences. Attitudinal segmentations are rich territory for marketers, as we seek to stake our brand’s claim in the minds’ of our customers.
The analytic process of segmentation looks to create groups of respondents. We are ideally looking for a manageable four to six segments. Within these groups, respondents will be similar to each other. But between the groups, we want to see differences. There are many technical methods we can use to reach this goal, and this is a domain that is both art and science. Different statistical methods such as k-means clustering Latent Class Analysis, Q-Factor Analysis and Mixed Models can yield very different outcomes. At the end of the day, it doesn’t matter how derive the segmentation – what matters is its properties and the extent to which is supports your objectives.
We tend not to use every available question for segmentation, but instead focus on a few batteries of questions. As a result, we may analyse several different segmentation solutions. Consultants then look to profile the segments by running crosstabs and charts of other questions. This helps elicit the between-group differences. Assessing segments is actually quite subjective. We are looking to judge a good segmentation on many attributes; including groups that are differentiable, actionable, sizable and winnable (i.e. not completely entrenched with the competition). As mentioned earlier, developing the story of the segments is also a key to success.
When the client and consultant have selected the most appropriate segmentation model, it is generally necessary to enrich the model with profiles and summaries of each segment which can be used to bring the segments to life for a wider client audience. Depending on budget for the project this may include focus groups and workshops to increase depth of understanding, particularly focused on needs and underlying motivations.
Now the hard work begins; strive to win over your target segment! This can take a long time. Besides judging on sales, we may want to further survey customers to understand if our marketing efforts are working. But, we don’t want those respondents to have to answer large batteries of questions from the segmentation. To address this requirement, analysts are asked to produce a segment allocator. This is a cutdown of the questions asked in the segmentation that accurately classify new respondents into their most likely segment.
Development of a segment allocator has some competing objectives;
- As much allocation accuracy as possible
- Fewer number of questions
- Simple set of rules / model for allocation
This stage of the process is also both art and science. There may be an ideal subset of questions that produces the best accuracy. But, a more curated subset might make for a better survey experience (i.e. because of mixed rating scales). Generally, including more questions improves accuracy, but makes the surveying experience longer.
The segment allocator might use sophisticated machine learning models, or more simple models. Generally, a more sophisticated model will produce better accuracy. But these are more complex to implement and therefor more costly. We typically use simpler approaches such as decision trees (if/else statements) or logistic regression (y = ax + b). These can be implemented in an excel workbook or embedded in a survey process. Generally the sophisticated model deployments are reserved for more operational studies. This is because the extra accuracy directly translates to increased ROI.
From my experience on both sides of the fence as a consultant and as analyst, I can say that segmentation is both about the heart and the mind.
We may be able to generate the most mathematically precise segments. But much of the success of segmentation comes from actionability. For that, we need to be able to tell the story of who the segments are. The questionnaire design phase makes or breaks segmentations. Starting with a good idea of the differentiating elements is important. This process may be better informed by free-flowing focus groups. Additionally, getting all stakeholders on board with the objectives, responsibilities and expectations goes a long way to ensuring a rewarding, productive segmentation project.