Market Segmentation with Novel Machine Learning

Even armed with traditional market segmentation research and past sales experience, a pharmaceutical field sales representative usually can expect to make 6–8 visits with a healthcare provider before securing business. That’s in line with the oft-cited general marketing “rule of 7” in which a company must get a brand or product in front of prospects seven times before a customer engages.

An innovative approach to market segmentation can help break those rules. Machine Learning attitudinal segmentation merges the best strengths of traditional primary and secondary research-based approaches, and it combines an attitudinal focus with the statistical power of big data to help companies better understand their customers and enable ever-more precise targeting.

Glean more robust, objective insights

While pharmaceutical marketers have long used attitudinal and behavioral segmentation approaches to identify potential customers and tailor marketing activities, traditional methods lack utility for healthcare-level commercial operations and tactics. Pharmaceutical companies commonly use behavioral segmentation and survey-based segmentation, both of which carry weaknesses:

Behavioral segmentation uses administrative data to segment physicians. So, for example, you might learn which physicians are early adopters, who prescribes what treatment options most, or whether a physician is based in a hospital or clinic setting – but segmentation factors are ultimately limited to data constructs available in administrative data. These data sources provide an accurate representation of certain specific behaviors, but cannot provide insights into motivations or triggers of their behavior, meaning the “why” and/or thought processes remain indeterminable. In addition, results are not always data-driven, because researchers often inject their own personal biases when defining healthcare provider characteristics and segments.

Attitudinal segmentation uses surveys tailored to the exact business need — measuring such healthcare provider characteristics as peer influence, industry friendliness, perception of safety signals, mechanics of decision making and therapy choice, or receptiveness to channels of communication, and treatment selection making, for example — to understand what messages or information are most likely to resonate with a physician. While analysis can produce interesting insights, survey samples typically lack random sampling, and often are skewed toward a homogenized pool of providers that self-select into survey participation. This means results cannot be reliably extrapolated to a broader healthcare provider population or context. Furthermore, primary research quantitative surveys are typically anonymized and only cover a small portion of the overall provider population, so traditional segmentations based on these data are incapable of assigning specific, real-world providers to segments.

Thankfully, innovations in machine learning and life sciences data analytics can solve for these limitations. Machine learning attitudinal segmentation is not limited to existing secondary data constructs like behavioral segmentation or a small universe of non-randomly-selected respondents like survey-based attitudinal segmentation.

Machine learning enables the ability to extrapolate segments to the entire healthcare provider universe, and modeling of the whole universe accounts for respondent selection bias. So, provider attitudes can be surveyed and projected through the use of secondary data across all healthcare providers to reveal the most likely motivations of individual physicians. Because the machine “decides” what the segments are, based on comprehensive data analysis, the results are objective and more purely pragmatic, while helping pharmaceutical companies identify the right people faster with better targeted messaging and content. The results also can help uncover opportunities.

These primary steps are taken to generate insights:

Quantitative data analysis: Rigorous, sophisticated analysis is used to develop healthcare attitudinal segments and extrapolate the provider universe. In this step, machine learning processes assign each customer in a comprehensive provider database that represents the entire provider universe to a segment. The final model results in significantly improved predictive performance.

Segmentation implementation: Now that the segments are coherent and stable, they can be used to understand customers and plan engagement activities. So, sales and marketing strategies, messages and tactics can be developed and tailored accordingly. In practice, when calling on new prospects, this also means field reps will be better equipped to lead with the most likely trigger (e.g., innovation, clinical outcomes, reimbursement, patient services, etc.), and cut to the chase to save all involved time and hassle.

For example, one company grouped physicians as “data-driven” (treatment selection is driven by the drug’s data), “peer followers” (those who look for the rest of the physician community for guidance), and “practice-centric” (those who prioritize patient volume, throughput, and ease of reimbursement). The sales and marketing leaders prioritized the groups, creating a multi-pronged engagement strategy that emphasized the three groups’ key motivators.

Typing tool generation: Once segmentation is achieved, a “typing” tool can be created through a similar machine learning process. Intended for deployment to field reps during customer interactions as well as for market research purposes, this tool helps reinforce and validate model predictions, and serves as a feedback loop to continuously improve the model’s accuracy.

Empowering engagement

Attitudinal segmentation can serve to enhance existing segmentation efforts, and the combined segmentation framework can reveal hidden pockets of opportunity, and inform efforts to drive the desired behavior. For maximum impact, pharmaceutical companies should combine attitudinal segmentation with more standard Rx volume-based approaches. The following is how this looks in practical applications.

In a recent product launch case scenario, the sales and marketing leaders of a global pharmaceutical company with a novel drug for a rare disease planned to focus initial outreach efforts on prestigious physicians practicing at academic centers of excellence. They “knew” from past experience working with key opinion leaders (KOLs) from these centers that these were the most innovative early adopters.

However, machine learning attitudinal segmentation revealed those physicians as a whole — outside of the few frequently consulted KOLs — are late adopters who wait for comprehensive clinical evidence before prescribing new drugs. In fact, research and analysis showed specialists in private clinics presented the best launch bet for the novel drug therapy. The leadership team switched gears to prioritize private practice physicians initially and roll out the new treatment option to other groups as more clinical and economic evidence became available.

In another scenario, a segment of healthcare providers who are highly receptive to efficacy data and generally described as early adopters was identified. However, the Rx data showed they were not prescribing certain novel drugs as frequently as they could or should. This revealed a previously unrecognized growth opportunity for the company to prioritize and develop outreach strategies accordingly.

Even in cases where all the relationships and patterns are found to align with the intuitive and/or experiential forecasts of field and marketing leaders, this innovative approach demonstrably quantifies the market, providing the relative size of the segment along with contact information for physicians in those segments. Machine learning attitudinal segmentation presents pharmaceutical companies with a powerful new opportunity to crack the traditional code and gain ever-more precise insights to engage customers.

Tahel Noy, Ph.D. and Bill Woywod, MPH, MBA are associate directors in the Life Sciences practice at Guidehouse.