The art and science of segmentation

Part 3 in our blog series: Not everything that makes you better is a pill

By Stephen Potts, Director, Purdie Pascoe and Gary Bennett, Founder, The Stats People

In this blog series, we have already looked at how pharma brands are increasingly using delivery devices to differentiate and deliver better patient outcomes. We also provided our top 5 tips for developing delivery devices that make a difference.

We are now going to dive into some of these tips to provide very practical advice on how to deliver an optimal market research programme to your organisation.

One of our Top 5 Tips focused on identifying the right stakeholders. For our third blog in the series, I have been joined by our friend and partner, Gary Bennett, from The Stats People, to talk about targeting and segmentation, and more specifically how to improve the effectiveness of your marketing through a segmentation approach that works.

As an industry, most of our segmentation models have been behavioural. Salesforce effectiveness approaches have identified ‘high potential’ prescribers based on their workload and what they claim to prescribe. As a consequence, our segmentation has given our salesforces a way of targeting the right customers, but with a one-size-fits-all message. We have been guilty of delivering the messages that we want to push rather than engaging in a dialogue that our customers want to have.

We have veered away from attitudinal or psychographic segmentation as a way of understanding the varying needs of our customers and adjusting our offer to each in order to be more successful. In the past, such complex segmentation approaches have been difficult for salesforces to implement, but CRM systems have now developed to such a degree that we can deliver different content to different customer segments through different channels. If we are not tailoring our approach, we are missing a huge opportunity.

Today, in a world where it has become of utmost importance to engage customers effectively through digital channels, it is more important than ever to segment effectively. Here are some of the principles that we hold dear when developing a segmentation model for a pharmaceutical product or medical device.

1. Content is King

A well-drafted questionnaire designed to measure patterns that we want the segmentation to represent is the best foundation for strong segmentation models.

Segmentation can be thought of as the ultimate form of data reduction. It takes multiple data points and creates a simple category-scaled typology, such that by knowing what category someone is in we can make predictions about their attitudes, behaviours and needs.

The biggest challenge with segmentations is identifying which questions we want the segmentation to represent. It helps to start out with some hypotheses about the patterns we hope to identify, with a balanced set of questions around key dimensions, without one theme being allowed to dominate.

The principles of good questionnaire design prevail. Using a mix of question scale types, mixing up positive and negative questions and breaking up large batteries of statements keeps the questionnaire engaging to respondents and helps discourage unhelpful patterns of response such as flatlining, a tendency to answer high or low, or at the extremes or centres of scales. This allows the segmentation to focus on genuine response patterns, rather than on artificial patterns determined by the questionnaire.

Techniques such as factor analysis can be used at the analysis stage, prior to running the segmentation algorithm, to ensure that the themes represented in the data are well balanced, enabling us to drop question items if one theme is too dominant.

2. Good Statistical Algorithm (The Science)

It is essential to use strong underlying statistical algorithms. We typically use Latent Class Analysis, which allows strong models to be built with any type of survey question, using appropriate predictive models for each scale type, rather than forcing artificial measures of distance between cases.

A key benefit of Latent Class Analysis is that it also allows us to combine attitudinal, behavioural and psychographic data, resulting in a nuanced segmentation and not simply a targeting tool. A good algorithm should not force the researcher to just adopt one type of question scale. It should allow for the blending of attitudes and behaviours, using a mixture of attitudinal scales as well as single and multicoded behavioural questions. Our segmentation analysts are involved in questionnaire development from day one, so that the questionnaire is optimised for a successful outcome.

It is also important that the algorithms have strong methods for neutralising, where possible, remaining response scale patterns (high vs low or extreme vs centre ratings). The algorithm should also provide robust measures of ‘data-fit’ and ‘complexity’ to enable us to settle on a relatively simple, but good fitting model.

3. Selecting the Best Segmentation Solution (The Art)

Choosing the optimal segmentation from a selection of good fitting models is an art, as much as a science. We pressure test several good-fitting models to ensure that they meet the following requirements:

a)     Discriminates the underlying questions well

b)     Provides a good fit to the data, within the simplest possible framework

c)     Is well defined, i.e. most cases are easily assigned to a segment

d)     Is well distributed, i.e. no segment is dominant (>40%) or too small (<5%)

e)     Has a low correlation with response style

f)       Internally consistent and makes sense

g)     Tells an interesting story that helps the client organisation achieve their objectives

h)     Can be estimated with a subset of questions (a typing tool)

By examining several statistically robust segmentation models through these eight lenses, we ensure that the ultimate segmentation that we recommend is fit for purpose. 

4. Bringing the Segmentation to Life

Ultimately, it is essential that the selected segmentation solution is adopted across your organisation. The segment names need to become part of everyday parlance and a common understanding of the segment profiles needs to be established.

To achieve this goal, the segmentation report from your agency needs to be engaging and easy to understand. We include segment snapshots, imagery and video footage to bring the segments to life. You may also want to involve representatives from your sales and marketing teams when naming the different segments to ensure internal consistency and buy in. Workshops to develop a common understanding across sales and marketing teams, exploring which approaches resonate with each segment, are also very helpful.

In summary, there are a number of different elements that are essential when developing a segmentation that will enable you to tailor your marketing and sales approaches. It starts with a well-crafted questionnaire, focused on the most salient types of information. The science must be able to synthesise attitudinal, behavioural and psychographic data before the art of selecting the most appropriate solution is applied. Finally, to become a living, breathing part of the business, the segmentation must be embedded in the organisation. Only then, will you be able to grab the opportunities presented to you and provide the right information and services to the right customers at the right time.

In our next blog, we will look at how best to understand the value of innovation, when combining pharma products with delivery devices and how conjoint approaches can help.

If you have any questions regarding this blog, or if you would like more information on how to develop successful delivery devices and beyond the pill solutions, please contact Stephen Potts at stephen.potts@purdiepascoe.com