With the increase in the number of touchpoints for a Healthcare Physician (HCP), it has become more important for pharmaceutical companies to personalize marketing efforts at the HCP level. Segmentation forms the backbone of customized marketing and is a critical step in converting brand strategy into actionable tactics. A robust segmentation process enables effective targeting – a potential group of customers receiving communication through preferred channels from the pharmaceutical companies.
Before we look at traditional approaches and how ML-aided solutions can help in improved segmentation & targeting, let us look at how one can measure the advantages of a segmentation process. Following are the key factors that can be used to validate a segmentation process:
One of the most common approaches involves looking at the historical prescription by the HCP, for the brand and the market, and determining segments based on past prescription behavior. Though this is still a valid and widely used approach, it lacks a few key elements:
What is the expected behavior
in the future?
What causes some of the HCPs to behave differently than others?
How to inform segmentation with external factors like the current COVID-19 pandemic?
An insight into the answers to these questions can help the pharma companies make far more informed decisions on designing marketing tactics based on segmentation. In other enhanced approaches, additional elements, such as historic patient visits, HCP demographic information, etc. are included in the segmentation process. Yet, they are only focused on historical behavior and do not provide much help in customized tactic design and targeting.
ML methods provide a potential solution for the missing elements in the segmentation & targeting process. Augmenting some aspects of the traditional approach with an ML-driven solution provides an enhanced framework for segmentation that meets the validation criteria described previously. One such solution, described in the subsequent sections, involves estimating a valuation score for each of the HCPs to be considered in the segmentation process and using the score to define value-based segments.
Unlike historical approaches of creating multiple scores based only on the historical data and combining them to get a single valuation/potential score, the enhanced approach incorporates the predicted future potential. Supervised ML models provide HCP-specific estimates of next opportunity based on a multitude of predictors, such as writing behavior, HCP demographics, historical promotions, the impact of external events (for example, incorporating the effects of COVID-19 through the change in patient inflow, affordability, and insurance provider mix). The estimated opportunity, when combined with old prescriptions, provides a holistic estimate of the HCP's overall potential. For a new drug launch with minimal history to build ML models, lookalike patient identification algorithms are used as an alternative for quantifying future potential.
This valuation score is used to arrive at HCP segments. A higher value segment thus derived will not only include the HCPs who have prescribed the brand heavily in the past but also include medium volume prescribers who have a high probability of increased brand writing in the future. A potential target list for a specific promotion channel is identified by estimating the break-even valuation score given the cost of promotion.
With the recent advancement towards increasing "explainability" of ML models, we now have algorithms that help in understanding which predictors and to what extent do they influence the predicted potential of an HCP being higher or lower. We use these explainer algorithms along with the predictive ML models built previously to identified HCP-specific drivers of brand writing. These key drivers are then used to create micro-clusters within each of the already identified segments. The HCPs within each micro-cluster are motivated by similar factors, and micro-cluster level tactics can be designed to communicate with them.
At Axtria, we have helped one of the top 10 US-based pharmaceutical companies to enhance their segmentation process using the above approach and to define an efficient targeting plan customized for each of the various promotion channels available for the brand. Subsequently, a promotion response assessment was integrated into the solution to provide a recommendation on the optimal level of promotion for each channel.
Having defined a broader approach for ML-driven segmentation & targeting, let us see how this approach fares on the segmentation validation criteria described previously.
Successful implementation of this segmentation approach would also include training the internal brand and sales team on the key features of the new approach. Also, it is recommended to assess the performance of the new approach in initial cycles of implementation – one such way is to design test and control studies. A salient feature of the new approach is that it provides relevant solutions that anchor to the same segmentation schema for both sales and marketing teams and hence helps to bridge a common gap present at many organizations.