Every marketing and sales professional would love to be able to peer into a crystal ball and see where the market is headed or what their customers are expected to do next. And now, thanks to advancements made in artificial intelligence (AI) and machine learning (ML), it is becoming possible to take historical data as well as the data life sciences companies currently collect and use predictive analytics to make—well—predictions. Predictive models can help companies to find patterns and behaviors in that data that suggest what might happen next, which is exactly what marketers want to know. And for those who want to know more about how this works and where the industry can benefit, we asked 12 experts:
- What are the best use cases for predictive analytics, AI, and/or ML that life sciences marketers can apply today in order to improve, enhance, or optimize their ability to reach and engage with their audiences? How can any of those three technologies be used to help marketers ensure they are making better and smarter decisions?
- What are the biggest barriers currently holding life sciences companies back from more fully utilizing predictive analytics, AI, and/or ML? What can companies do to overcome these barriers?
- What tips can you offer to companies or marketers looking to apply predictive analytics, AI, and/or ML to a specific problem or issue they are trying to solve? How can they determine if any of those strategies is right for the situation, if they have the right data, what approach they would have to take, etc.?
- Some have described our industry as being in the “peak hype cycle” of AI/ML. Do you believe these are just buzzwords at the moment or hold actual promise for the industry right now? Do you have any examples of applications of AI/ML by life sciences companies that have produced results? Or, leaning into the idea of hype, what kind of impact do think AI/ML can make further down the line for life sciences marketers?
Marketers should start thinking about predictive analytics, AI, and ML not as future-facing technologies, but as tools they can integrate into their existing processes to improve their current operations today. In my experience, the most successful implementations of predictive analytics, AI, and ML are designed to be activation-ready from the start, meaning they are developed and deployed in response to clearly articulated key business questions (KBQs).
High-value KBQs are lines of inquiry that have a high potential business impact and are highly actionable. For marketers, honing in on these KBQs and developing an understanding of how a predictive analytics, AI, or ML tool would fit into their pursuit of answers is a necessary condition of success.
Determining if Predictive Analytics is Right for You
To date, the most effective use cases for these tools in pharma marketing have included predictive physician targeting and segmentation and AI-informed patient support services. Predictive analytics, AI, and ML are powerful tools in enabling marketers to develop highly effective communications and services strategies based on precise predictions of future behavior. In other words, the insights produced by these implementations of predictive analytics, AI, and ML enable marketers to not only answer KBQs, but take concrete steps that drive real business value.
As such, the most important thing a marketer can do to determine if a predictive analytics, AI, or ML tool represents an appropriate solution to a specific problem they are trying to solve is consider whether the tool can be integrated into their existing processes. These tools are most effective not when they are treated as addenda, but when they are woven into campaigns and services from the earliest stages of the planning process and used as a means of developing activation-ready solutions to solve KBQs.
Today, the amount of health information is doubling every three years—by 2020 it is estimated to double every 73 days. This growing body of data may hold the answers to many of the world’s most enduring health challenges (https://ibm.co/33yhY1C). This drastic advance in level and amount of research itself means a higher chance of curing diseases and saving lives.
The bad news: Rapidly growing amounts of data means an increase in unstructured data. The speed of data growth causes significant problems in terms of data extraction and analysis. This is where AI creates value: Used in the right way it can accelerate research and drug discovery processes and enable innovation.
AI methods such as Intelligent Web Crawling, Computer Vision, and Natural Language understanding can optimize the work of a data scientist providing a structured overview of most of the data. However, this alone is not the end. It is at least as important to understand the language of life sciences to draw the right conclusions. This emphasizes the importance of going beyond NL and modifying it to Life Sciences Language Processing by creating a self-learning life sciences ontology to enable usable data analysis.
Developing a Life Sciences Ontology
An ontology acts like a connected dictionary which includes not only the definition of a concept but also the relationships to others. It can help in identifying unknown connections between genes, targets, and diseases, accelerating the identification of new biomarkers and drug candidates or automatically gives you the top KOLs in a specific domain.
AI can not only perform repetitive and labor-intensive tasks faster, more efficiently, and comprehensively, but can also help in deriving the right conclusions, previously hidden in the vast data ocean. It empowers individuals to spend their time on higher cognitive tasks instead of lower cognitive tasks. And ultimately, speed up the time to market for life-saving drugs.
For many business users within life sciences organizations, the data warehouse has traditionally been a “Black Hole.” Data goes into the warehouse and access becomes a challenge for them. That’s why the business needs to partner with the Information Technology Team to identify the right AI vendor for the organization to ensure users have access to the data they need.
Preventing Data from Falling into the Black Hole
AI technology must have the ability to connect to CRM, data warehouse, and other organization data sources (inhouse or outsourced) to gain access to sales and marketing activities combined with outcomes data to produce actionable and insightful recommendations. Before embarking on the AI journey, ensure your data warehouse has the required data (structure and unstructured) to support your marketing and sales strategy. Identify an AI partner that has industry life sciences experience and is capable of consuming data from multiple sources, such as CRM, data warehouse, partner portals, and data files. Cloud AI Solutions will give your organization the greatest flexibility to modify your AI strategy over time and extensibility to expand the AI solutions across the organization and/or products.
Implementing an AI solution the right way in the organization will empower business users with improved targeting, increased reach, and effective pull-through from insightful suggestions and/or alerts in CRM. AI will also open non-personal promotion channels to reach new audiences in geographies that have been considered whitespace. Many companies already have the data needed to enable AI in CRM—removing the data access barrier will ignite the explosion of insights for the business users.
1. Which customers do I want to engage? To answer this question, the use of sales data is essential. ML, combined with domain expertise and information about a potential customer base, has the power to understand which customers are driving a company’s current business and which have the potential to drive it in the future. Gone are the days when marketers had to rely on “rules of thumb” or high-level customer tiers and segments to prioritize their customer base. This is all true even as the lag time, granularity, and coverage of sales data changes.
2. What do I say to each customer? ML can help marketers develop personalized engagement approaches, essentially providing “segments of one.” Before the age of the internet, personalized marketing reigned king. Each sales rep got to know each customer in a meaningful and sustained way, and could tailor their actions and messages for each. We’ve strayed from that personalized marketing approach amidst an industry-wide shift to mass marketing—but can get it back using data and analytics.
Today, intelligent engagement technologies are able to determine customer interests by tracking responsiveness across all channels. Using ML, these technologies can also predict what any one customer is likely to engage with based on data from customers with similar demographics and purchasing habits. With this information, marketers can determine which content resonates best with which type of provider. All of this information is then combined to build complex, multichannel marketing strategies for each individual customer that adapt and improve over time.
AI, ML, and other terms hold sway in the business community these days, offering potential tools to unlock competitive advantage or unparalleled efficiency. While there is certainly promise, today’s life sciences companies need to take a thoughtful, holistic, and learning-based approach before adopting these technologies.
Let’s first take AI and remove it from the discussion of life sciences commercial operations. AI implies real-time autonomous capability—think Tesla Autopilot. There is no place for this currently in our field and no need for it. Be wary of sales pitches that offer AI capability and challenge them on what use cases would be appropriate.
ML Offers More Value Over AI
However, ML is a product-market fit. ML is more deliberate and reactive—terms that may have negative connotations but, in this case, really make sense. Let’s take a practical example that we see deployed right now in the real world: Creating call plans for your field teams. You have limited and expensive field resources yet lots of potential customers to talk to. How do you optimize your approach? This is where ML can help.
Business, in partnership with data scientists, can create adaptive models that learn over time. As usual, start with your KPIs and historical performance to develop a call plan. However, this time, supply the feedback of how that next quarter went into this model. Enable it to learn from what worked, and let it apply that feedback to the next iteration. Soon you’ll be talking to the right customers at the right frequency, and seeing the results.
Of course, none of this works without good clean data. Make sure you’re making foundational investments in master data management and other data quality initiatives. After all, the meal is only as good as the ingredients that go into it.
As an engineer, marketer, and champion of health equity, I am heartened by growing sophistication in predictive health modeling due to the integration of data on social determinants of health (SDOH) into the EHR.
Up to 50% of an individual’s health risk can be attributed to factors such as housing stability, food insecurity, transportation and mobility issues, social support, and other measures. Yet the vast majority of U.S. healthcare dollars and efforts are spent on clinical factors instead of addressing the underlying socioeconomic and behavioral factors that greatly impact patients’ health. And with contracts being measured on patient outcomes, ignoring the role of SDOH is bound to affect value-based reimbursement.
Providers and payers are increasingly aware of the need to capture data on the availability and access to resources that will enable a patient to meet daily needs. And they are moving to collect and use it in actionable ways.
While many providers within the social net have long captured SDOH data, it lived outside of the electronic record. Fortunately, it’s becoming commonplace for technology vendors such as Cerner to offer social assessment forms in their solutions, automating the downstream integration of a factor such as whether the patient lives in a food desert, lives in an unsafe environment, or an underserved rural area.
Making SDOH data shareable across the continuum of care still faces notable challenges. However, we are now at the point where we can integrate this data as variables in all predictive modeling of a patients’ future health, or reimbursement risk.
What I hope to soon see is the integration of SDOH in the pharma industry’s modeling, as we progress in leveraging Health Economics and Outcomes Research to support precision medicine.
Every year, thousands of speaker programs train hundreds of speakers, but because of certain inefficiencies in recruitment and planning, the results of these programs may fail to deliver reliable outcomes. Often, low-impact speakers are selected by subjective criteria, which leads to low audience attendance and engagement.
However, applying predictive analytics allows for objective measures to inform the speaker identification process and can leverage HCP networks and connections for audience selection. In fact, the profound impact of incorporating AI into promotional program planning is evidenced by the fact that matching clinical leaders with attendees in their network can yield dramatic multi-fold increases in ROI and incremental prescription volume, as compared with using traditional, non-data-driven methodologies, such as field force speaker nomination.
Barriers to Implementation
Several years ago, a photograph of a dress became a viral internet sensation when viewers disagreed about whether the dress was blue and black or white and gold.
This example of people seeing something completely different when looking at the same photo presents an ideal analogy for AI and ML. For example, why are HCPs who were once seen as highly influential appear less so when viewed through the lens of predictive analytics? How can a treatment paradigm, favored in advisory boards and market research, not quite align with the results of a claims-based clinical journey? Why do known specialists have undiagnosed patients for the rare and difficult-to-diagnose diseases they treat when AI can support HCP and patient activation of the clinical journey? These issues arise from the shortcomings of traditional processes and barriers to adopting new ones, such as:
- Fearing the new—rather than adopting more effective and efficient approaches.
- Operating in functional silos—only through the integration of data science with medical expertise and marketing strategy can we translate analytic outputs into actionable insights.
Marketers, data scientists, and analysts should work together from the start to ensure clarity on the problem they’re trying to solve, the datasets they need, and how to effectively test the insights they’ve drawn. I believe AI (Artificial Intelligence) still needs IA (Intelligent Assistance from humans) for successful outcomes. AI is very good at surfacing the patterns in the data that you provide, but we still need humans to decipher whether it’s meaningful.
The Growing Benefits of AI
Although there has recently been a large uptick of new datasets available globally, organizations have not been able to utilize them because they do not have the systems needed to properly ingest so many varied sources at once. As these systems are being developed, AI will be able to surface patterns that are normally invisible to humans. In healthcare, applying AI to anonymous patient datasets can provide insights into how therapies are working for patients. Further, with this knowledge, we can perhaps improve diagnosis, treatment protocols, and medication adherence for patients.
Data acquisition cost is the number one factor holding life sciences companies back from more fully using predictive analytics and AI for speculative R&D. Moving the needle with predictive analytics often requires large volumes of training data, from thousands to millions of training examples.
Life sciences companies are often working with unique sensors that do not yet have large publicly available datasets, and so they have to collect their own datasets. Collecting their own often involves human subjects, doctors, fulfilling legal requirements, and long processing times in labs with expensive equipment. Each data sample can cost tens to hundreds of dollars. That means companies must spend tens of thousands of dollars at least, plus plenty of effort, to collect a speculative dataset that is more likely than not, useless. This really slows down exploration and innovation. Companies that do manage to navigate the expense of collecting useful datasets do not share their data because it is such a competitive advantage. So the sector as a whole could benefit from companies both sharing the data itself and sharing lessons learned about what data collection practices are most likely to yield useful data.
Advice for Getting into Predictive Analytics
Before hiring a full-time data scientist, hire a fractional data scientist with successful expertise across a variety of data types. They should also have experience in making the most out of limited quantities of data. Even if they cannot obtain results good enough for the targeted diagnostic of therapeutic product, they can often indicate the trends in the data. These trends can help to predict the amount of training data needed to improve results sufficiently. Similarly, they can predict how the data resolution might be limiting results and if there are options for collecting higher resolution data.
For life sciences and healthcare marketers, AI and ML are foundational elements for tailoring products to customers. The healthcare delivery system faces consistent challenges around quality of care and cost containment. Marketers use AI and data analysis to develop measurement strategies and build products that focus on patterns to maximize efficiency across products.
One Use Case Example of AI and ML in Pharma
Data services are used to clean and aggregate datasets (claims, patient, physician, and Rx data), which enables professionals to identify precisely which patient, treatments, and outcomes are linked to which physician process. At the center of the data service is a machine learning algorithm that enables the team to isolate patterns that reveal specific impacts on patient health outcomes. When the model processes incoming data from the various systems, practices ranging from drug prescriptions to hospitalization time are scored depending on how detrimental or beneficial they are in terms of cost and patient health.
The best use case we have found for predictive analytics is in helping to improve pharmaceutical sales force effectiveness and to bolster the activities of sales reps by using data-driven insights. While predictive analytics is widely used to estimate the number of calls to be delivered and several other tasks, AI is still very new on the commercial side. But one application we have found to be good use of AI/ML is for the rep’s next-best-behavior to be calculated and communicated constantly in real time.
When sales reps are operating in territories under similar conditions (the same quality of targets, managed care, etc.), but performing differently, AI-powered predictive analytics can provide clues on how to improve sales behavior.
Improving Sales Rep Performance
Reps do not have an easy way to know what they need to change to sell more. By using AI to track and monitor the process, we can analyze rep behavior within similar territories and show reps the best specific behaviors that will most likely improve sales results. Over time the algorithm continues to learn and adjust and improve upon best practices.
This information can be used by district managers during coaching sessions with reps and be used alongside primary market research to ensure the highest rates of sales effectiveness. Primary market research can uncover any overflow—any activities the best reps are doing that is not recorded in the CRM system—for further support in designing and communicating the sales process.
Using AI and analytics, reps can decide how to fine tune their call activity, how and when to drop off samples, which messages deliver the highest results, how to improve managed care pull-through, and more. AI really will help improve the relationship between sales reps and doctors and contribute significantly to pharmaceutical sales force effectiveness.
The biggest barrier from fully leveraging AI/ML: The way it gets used in an organization. Most organizations rely on data scientists for their AI/ML needs due to technical and knowledge requirements and then results are reported back to business users. This becomes a major barrier in multiple ways.
First, data scientists are very limited in numbers and spend more time cleaning ad-hoc business data then actually applying AI/ML, which in turn limits their bandwidth. They also tend to have limited knowledge of specific business needs and often face unrealistic ML expectations by the business user and report back their findings in a highly technical manner. As a result, only a very small percentage of corporate data ever gets used for ML/AI and the results often get lost in translation or not what the business needed in the first place.
Democratizing ML/AI in Businesses
The best way to overcome this: Provide data scientists with governed and certified datasets directly from business intelligence (BI) platforms so they can focus more on the ML/AI process and integrating their resulting algorithms seamlessly into modern interactive BI tools that are already in use by businesses. This would facilitate the democratization of ML/AI across the business by allowing tested and proven algorithms created by data scientists to be used many times by many users with different subsets of the datasets without having to go back to data scientists.
I truly believe ML/AI is not a buzzword but will be with us for a long time. The technology is there, but businesses do not yet fully understand how to properly apply it or to use the technology without a data scientist in the middle. Once these barriers are removed and businesses can incorporate ML/AI seamlessly into their daily decision-making processes, it will make a huge impact.