Data science: a term that always sounds cutting-edge. While discussions of personalized medicine and the related science of gene therapy and similar advances become more science than science fiction, pharma marketing is beginning to step into the individualized, independent experience economy, as well.
What is the experience economy? It’s not just the product or service but how the experience of getting that product or service makes you feel … i.e., “Am I being treated as the individual I am?”
Practically speaking, the experience economy is exemplified by companies like Netflix and Uber, which have changed their industries by focusing on addressing the problems associated with getting, growing, and keeping customers. Netflix retains subscribers by constantly improving the personalized recommendations of things to watch; Uber made pay-for-service transportation better by reducing the cost of entry to participate, offering up-front pricing, and eliminating fare control.
Data science is at the core of each of these industry improvements: each used it to find the hidden stories and patterns in data that can create efficient, valuable experiences. Data science, as a practice, uses historical data and algorithms to teach computers to think, which in turn helps us make smarter business decisions. While terms like artificial intelligence (AI) are thrown around as the solution for almost everything, the opportunity for making the experience economy real is about combining business expertise, defining the core strategic problem(s), and using data scientists, technology, and machine-learning models to solve them.
Machine Learning: Powered by Data Science
Let’s use the example of driving to demystify machine learning: how do you make the decision to stop or go through an intersection when the green light turns yellow? Your brain considers several similar historical situations and applies those experiences to the data that’s presented to you in the moment: How much of a hurry am I in? Is there a red-light camera? Is there a police car in sight? How fast am I going? Is my family or kids in the car? Is there a car ahead of me that’s going to go through? Is there someone turning left in front of me? Is the sun inhibiting my view of the intersection?
The yellow-light decision is very personal, and potentially very dangerous. But it’s one that many of us make almost every day. Now imagine that a car is making that decision for you, and you have a pretty good grasp on how self-driving cars work! The improved driving experience might be summarized as reducing the physical and mental requirements, making our use of time more efficient, and improving the safety of personal transportation.
Personalization: The Key to the Experience Economy
The online experience economy is based on two factors: independence and individuality. We want what we want, when we want it and how we want to get it, and we want it to be specific to us individually. Today’s experience economy is about personalization; no one wants to be part of some huge, amorphous group that is summarized by saying, “All of these types of people do or want this.”
This is where pharma marketing is changing, much the same way medicine is becoming more personalized. Let’s outline three areas where data science can improve the process of getting the right treatment into the hands of providers who improve the lives of their patients, one of our most personal human experiences:
- Segmentation: How do we better identify the providers we want to reach? Instead of selecting large groups based on specialty or TRx decile, why not let the data define the groups? Machine learning techniques such as clustering can define not only how many groups the data can support but also differentiate the similarities within and the differences between those groups. This improves targeting by highlighting the important group variables that can influence messaging and channel selection, for example.
- Targeting: How do we determine which providers to spend our marketing dollars on? And of those we choose, how should we reach them — rep visit, email campaign, media — and when or how often? Data science-driven targeting can reduce waste, improve effectiveness, and inform messaging and channel use to improve overall return on investment (ROI).
- Measurement: While ROI is important, you may be interested in understanding which channels drove success or how much of each part of a campaign was influential. New algorithmic solutions can optimize marketing spend through attribution modeling and marketing mix modeling. The power of these models lends itself to a more dynamic understanding of how change in spend in one or more channels affects the probability of success.
Following are six mini case studies that illustrate ways data science can help pharma marketers do more, faster and better, with less wasted time and money.
Making Complementary Prescriptions More Interesting to HCPs
The problem: Providers unresponsive to non-personal promotions (NPP) for products complementary to those they already prescribe.
The solution: Optimize NPP targeting. Provider data for more than 100,000 specialists were clustered into four groups: “All In,” “Potential Switchers,” “Disconnected,” and “Idle Prescribers.” The cluster analysis is important because the client was interested in optimizing their marketing spend for NPP – in this case, email, in combination with rep visits.
The results: 60% reduction in targets, with a slight lift in overall prescriptions. “All In” and “Potential Switchers” segments (~50K) were shown to have opportunity and were used for NPP marketing. After over a year of tracking prescriptions, we confirmed that those segments continued to make up nearly all scripts, proving that targets were identified correctly.
Improving Email Open Rates
The problem: Email open rate below benchmarks, and with higher-than-average “marked as abusive” rates.
The solution: Optimize non-personal subject line and body content. A test campaign targeted thousands of physicians over a month to benchmark an open rate of 4%. Analysis uncovered the words and phrases that were significantly more likely to result in opens, click-throughs, or in emails being marked as abusive.
The results: Clickthrough rate up 25%; prescriptions up 7%. A campaign developed with these findings resulted in 25% increased click-through, and a reduction of “marked as abusive” by 70%. Treatment rates improved by 7% for the providers in the new campaign (though a variety of factors could have impacted behavior).
Handling a Changing Formulary
The problem: Changing formularies put a treatment at risk in the market.
The solution: Tell national account managers (NAMs) which payers were likely to make changes when. Seven NAMs handling more than 500 payers nationwide were able to learn from predictive models not only which payers were most likely to make formulary changes in the next 90 days, but also how different payers influenced each another.
The results: Accurate to 80%; protection of $120-$200 million possible. Correctly predicted 80% of payers who made changes in the first month, with 62% overall accuracy over the year. Had the models been live between 2016-2018, an estimated $120-$200 million in revenue would have been protected.
Figuring Out Which Markets to Target, Which to Protect, and Which to Drop
The problem: Where should we target our new treatment? How can we protect our market share with new entrants? Where should we target a competitor?
The solution: Market and competitive analysis. Using provider-specific TRx data from providers like IMS or Symphony, it’s possible to avoid the pitfalls of focusing on the wrong competitors or geographies.
The results: Depending on the size of the target list, existing prescribers list, or predicted potential new prescribers, we have seen up to 25% savings in overall marketing spend and 10-20% increases in TRx versus control groups
The problem: What are our target patients like?
The solution: Patient persona development. Going beyond the common personas found through third-party vendors, these personas highlight geographic, demographic, psychographic, behavioral, and engagement variables to understand and identify potential patients and develop the right messages that appeal to them.
The results: In one instance, we were able to reduce the target MSAs by more than half where the potential patient base was not of interest based on gender, income, and engagement channels.
Predicting Which HCPs Will Stop, Start, or Increase Prescribing
The problem: Which providers are going to change how they prescribe a treatment?
The solution: Classification models. These predictions make it possible to determine whether an individual in a population falls into a small subgroup – for instance, providers likely to start, stop or increase prescribing a drug.
The results: For one client, we reduced a target list of 9,900 existing prescribers to around 4,600 with the highest probabilities of dropping or switching treatments and then offered a list of 1,400+ potential new prescribers to target. In the six months since the predicted new prescriber list was developed, almost 200 new prescribers from our model began prescribing and have similar TRx patterns as current high prescribers. Low decile prescribers in the 4,600 existing prescribers have increased TRx by 11%, with most reducing or eliminating two competitive treatments.
Data science can make a real difference in continuing to evolve pharma marketing: throughout a brand’s lifecycle, throughout a patient journey, and throughout your to-do list. Talk to your Intouch team about ways that our data science experts can put tech to work for you!