Reuters Events | Pharma

Thought leadership and innovation for the Pharmaceutical Industry

Machine learning is a boon for customer engagement. How can pharma best use it?

By Adam Chapman on May 3, 2018

The ability to read other people’s minds has long captivated the popular imagination. The conceit behind 2001’s What Women Want sees Mel Gibson acquire the power to read women’s minds. Gibson quickly gets to work undermining his office opponent (Helen Hunt) in a bid to steal the promotion she beat him to fair and square. Of course, things get messy when he actually “listens” to her thoughts and falls in love with her.

Fast forward to 2018 and mind-reading seems closer to science than superstition. Advanced algorithms continue to crunch the data generated by our digital footprints, tailoring content to our specific interests and increasingly predicting our next move.

This may sound Orwellian, but customers are embracing it. The trade-off is immediate and effective engagement. These tools may have come of age in the consumer industry, but they are now trickling down into healthcare.

Machine learning is already shaping pharma’s customer-engagement strategies, with traditional supervised learning gradually shifting towards more unsupervised deep learning, says Björn Van Loy (PhD), Global Head of Advanced Analytics, Trilations.  

Supervised machine learning is when you have a theory and you can implement that theory. Marketers are currently applying this approach in a bid to better classify which doctors to visit and when.

This is coupled with “regression” analytical techniques, which enable companies to peer into the future and predict the patterns of different doctors, gleaning valuable nuggets such as their prescription behaviours, says Van Loy.

Then there is unsupervised learning. This is where the machine is “deep learning, which is the blackest box because you don't put anything in there, you let the data speak for itself,” he says.

This is the stage before targeting — when you segment your doctors. The industry is looking to segments of larger groups of doctors, but Van Loy believes strides in AI tools can even segment to individual doctors, enabling pharma companies to devise a specific strategy to meet their individual needs. This will be the challenge of tomorrow, he says.  

Van Loy believes HCPs have a growing appetite for AI-led strategies. It’s ability to “aid drug discovery, early diagnosing, easier access to medical education and to track everything that is happening with the patient” is an attractive prospect to the physician.

Societal attitudes are also helping to turn the tide as the proliferation of smartphones and wearables continue to shift the boundaries of trust versus privacy — people are increasingly seeing the value of sharing their data, he says.  

Agnieszka Wolk, Director Data Science, IQVIA, shares Von Loy’s sentiment that advanced algorithms are already making their mark in pharma’s customer engagement space:

“Fuelled by rich data assets, enabled by new technologies, and empowered by machine learning and artificial intelligence algorithms, advanced analytics can uncover patterns we haven't seen previously and provide us with deep insights that can answer pharma’s most challenging questions. Advanced analytics has huge potential to impact healthcare significantly,” said Wolk. 

The result? Pharma’s targeting tools are becoming sharper and more sophisticated, which is essential to HCP engagement. The vast proliferation of touchpoints available to the HCP, combined with restricted access, calls for an ever more precise and meaningful marketing message, she says.  

AI provides the tools to deliver tangible impact by gaining insight about HCP’s communication needs, says Wolk. This is not achieved through generalized information based on market selection, but rather through ascertaining their preferences, anonymous patient cohorts they have on their books and the best treatments for those populations.

“What type of knowledge does this doctor have? What content would the HCP be interested in — adverse effects, clinical research or price reimbursement system? What would the HCP be happy to read about, what would the HCP need to know to make a treatment decision? How often does the HCP want to have touchpoints and information from us?

These are critical questions that machine learning and AI can inform,” said Wolk, facillitating a more efficient engagement with HCPs while whittling down irrelevant touchpoints and content, reducing the burden of information overload.     

This rigorous line of inquiry is prompting a more proactive approach to targeting. IQVIA is employing machine learning to predict specific events on the patient journey. The company crunches reams of data to calculate the probability that a specific patient type, will have a specific event, and then designs triggers actions to address that risk.

“For some cancer types, physicians diagnose but decide not to put patients on a treatment immediately. They prefer to wait and watch before they set up a course of treatment,” said Wolk. “We are using deep learning to identify features and feature combinations that differentiate and set aside patient types with high probability of transitioning to first line treatment.”

Identifying and grouping different priority areas and aligning specific treatments to them is a powerful tool of engagement. Pharma companies can alert HCPs to early warning signs and provide them with tailored solutions to support them in their decision-making process.

Employing advanced analytics to allocate resources has the potential to boost returns on investment by calculating the best avenues for profit, says Van Loy.

Salesforce teams can also be organized more efficiently, he says. Evaluating their performance in a specific disease area could prompt companies to reassign team members to other disease areas that better suit their strengths, yielding handsome returns.   

In addition, advanced analytics can help commercial teams determine the most effective omni-channel mix for communicating to doctors, says Van Loy. “Each channel that you use has a certain impact and a certain cost. It has an optimal use and if you combine everything together, we can use algorithms to actually predict what would be the best omni-channel mix.”

Trudge before you can run
To maximize machine learning’s potential, there are some pragmatic and philosophical questions that need to be addressed first, says Van Loy. “If you have these data lakes, you want to know:how deep is your lake; what data do you have available and what is big data?”   

Pharma companies might be swimming in records, but this doesn’t constitute ‘big data.’ Rather it is the “velocity, volume and frequency” of the data that is important, he says. The value will come from combining these three elements, which remains a challenge, but the industry is striding towards solutions, he says.   

The quality of the data must be impartially assessed and scrutinized if it is to yield genuine insights, says Van Loy. Data sets are susceptible to bias, and the source must be questioned. He points to the possibility of sales teams skewing data to advance their interests and doctors distorting their answers in surveys.

Wolk shares this sentiment, stating that data quality is crucial for success of advanced analytics. Insights are only as good as input data. 

“You need to try to normalize, standardize and optimize your quality of data,” says Van Loy. “Then see how many data points you have left. Is it still looking like big data?” 

An AI-led approach requires companies to do some soul-searching before stepping out. They must understand the “potential noise of their data” and have a long-term vision, cautions Van Loy. He underscores the need for companies to honestly assess the tools and techniques they have at their disposal. A lack of technical know-how will prove particularly problematic in the unsupervised area of machine learning. 

Finally, companies must combat internal resistance. It starts with educating your people, says Van Loy.  The virtuous circle of data entry and learning underscores the need for everyone to muck in to make the pool richer and more varied. This may require an organizational transformation to get the right people on board. The traditional role of the business analyst (BI) may even evolve to become more data science orientated.