AI Special Feature: Harnessing the ghost in the machine
While a number of factors are constraining the usefulness of AI tools for pharma marketers, experts believe that the use and sophistication of these tools will evolve.
When it comes to the use of artificial intelligence, AI – as it has been classically defined in science fiction – is not truly here yet. In 2019, there are no self-aware robots or bodiless nonhuman intelligences prowling the internet. What we do have are Siri and Alexa, Google’s “Computer,” consumer chatbots, and little machine learning algorithms such as DeepLab and SPADE that can create weird, hilarious, and somewhat disturbing names (often all three at the same time) for craft beers, kittens, and even burlesque shows.
But pharma has been getting into the AI wave. Mostly it has been on the clinical research side, using machine learning to plow through mountains of data to narrow down R&D targets or compile evidence.
In June, Sanofi announced a partnership with Google to create a virtual Innovation Lab with the goal of changing how Sanofi develops new treatments.
“We stand on the forefront of a new age for biology and human health, with the opportunity to transform healthcare through partnerships with pioneering technology and analytics companies,” says Ameet Nathwani, M.D., chief digital officer, chief medical officer and executive VP of medical at Sanofi. “Combining Sanofi’s biologic innovations and scientific data with Google’s industry-leading capabilities, from cloud computing to state-of-the-art artificial intelligence, we aspire to give people more control over their health and accelerate the discovery of new therapies.”
The collaboration focuses on three key objectives: to better understand patients and diseases, to increase Sanofi’s operational efficiency, and to improve the experience of Sanofi’s patients and customers.
“Life sciences companies are looking to data driven, digital innovation to help fuel the creation of accessible healthcare solutions,” says Thomas Kurian, CEO, Google Cloud. “We look forward to collaborating with Sanofi to help accelerate the cycle of healthcare innovation to populations throughout the world.”
What about marketing?
On the pharma marketing side. AI progress has been much slower.
According to Justin Chase, executive VP of innovation and media at Intouch Solutions, the agency had been trying to convince clients last year that AI could do a lot for them. “We were trying to sell in these artificial intelligence powered ecosystems” such as AI powered patient support programs, receiving data from wearables and other sources,” he says.
While clients understood that value AI could provide, they were cautious. “They were not ready to go from 0 to 60 in 2.5 with this whole AI thing,” Chase told Med Ad News.
Then there were the promises that IBM was making with Watson, which was supposed to change every aspect of the pharma industry. “They overpromised and underdelivered,” Chase says. So trying to sell AI solutions in the wake of Watson’s failures was difficult.
Paul Balagot, chief experience officer at precisioneffect, says Silicon Valley tech companies are trying to close the gap between the application of their technology with pharma’s needs. On the flip side, pharma companies and biotech companies are looking for strategic partnerships to leverage many of these technical innovations.
This means even with the disappointment stemming from Watson, pharma companies started putting their own solutions into play. Novo Nordisk has a chatbot called “Ask Sophia” in which patients can ask very specific questions about Novo Nordisk products.
Although Ask Sophia seems like a sophisticated step forward, according to Ritesh Patel, chief digital officer at Ogilvy Health, the chatbot is a less than ideal example of AI.
“Have you played around with it? You should, I did. I got a glass of wine at 1 in the morning and pretended I was a patient suffering from diabetes,” Patel says. “I logged on and said, ‘Hey Sophia, I’ve got diabetes, will I get gout?’ And the response was, ‘I can’t help you with that right now, please call this number.’ And then I said, ‘Hey Sophia, I’m fat, I’m overweight, am I susceptible to diabetes, do I have to exercise?’ And the answer came back, ‘I can’t help you with that right now, please call this number.’”
According to Patel, what Novo Nordisk has done “to much fanfare, though the execution is not that good” is that it took the most frequently asked questions being received at the call center by MSLs and put them in a chatbot.
Ultimately, Ask Sophia is not really AI. “It’s just a guided conversation,” Patel says. “The machine’s not doing anything, it’s just guiding you through conversation trees.”
There are more sophisticated chatbots around, such as Conversation Health Guides in Toronto, and Colgate’s Brush With Me. Both use machine learning and natural language processing to discuss teens’ health concerns and HPV vaccination (Conversation Health Guides) and help moms teach very young children how to brush their teeth.
Both are “learning” programs, taking the responses and using them to figure out what information questioners were seeking. The Colgate program uses Google’s assistant, which had 400 kids teach it “kidspeak” before it was launched.
In less specific ways than Ask Sophia, AI has impacted pharma marketing in three areas, according to Pratap Khedkar, managing principal, ZS Associates. These areas are enriching insights, optimizing decisions, and enabling actions.
For example, AI is being used to improve physician engagement by doing marketing promotion in a better way. “Currently physicians get inundated with promotion – from reps, emails, alerts, all the internet channels – to the point where a high value doctor may get hit 2,800 times a year by pharma as a whole,” Khedkar says. “None of this promotion recognizes the individual physician’s preference for certain channels, or certain content and messages. The touches are uncoordinated and pile up on top of each other without the right cadence. AI is being used to streamline all of this – predicting which physicians will engage with certain channels, predicting the content affinity of a physician, designing the right sequence of touches to maximize an individual’s engagement and eventually prescribing behavior.”
According to Khedkar, this leads to AI creating a “Next Best Action” with each physician every day that the rep needs to do. “There is enough data now to do all this, as well as automated software systems to enable the actions at a micro level,” he says. “The intent is to personalize and harmonize all the contacts to every individual physician.”
Measurements have shown the impact on engagement and prescribing behavior is substantial, with a 6 to 7 percent increase in sales. “Prediction has also been applied to which is the right customer, based on their propensity to write or switch in the near term, and this leads to dynamic targeting using AI,” Khedkar told Med Ad News.
According to Patel, there have been trials in using AI to serve up ads and training them using machine learning.
“You’d create an ad, which will then be broken up into image, text, background, call to action, format, and size,” he says. “You then service it up and let the machine learn which pieces work really well and what’s resenting with whom. The machine then compiles the ideal ad unit.”
Although there have been pilots of this AI use, “then you have regulatory in the way,” Patel explains.
“Everything has to be preapproved so how can you dynamically create an ad unit that works really well?” Patel says. “You need to come back to approvals. You could say, ‘OK, here is the ad unit that works really well, can you please approve it,’ and then you can serve it.”
These medical and regulatory concerns about messaging may have limited Ask Sophia’s capabilities, according to Patel. “I bet medical and legal said, ‘I want to know and preapprove this conversation.’” He also guesses the chatbot was not trained with patient answers, but with the call center agent input. “It’s like taking the FAQs from a website and making them an AI.”
Intouch Solutions has built its own AI engine, called Cognitive Core, which was featured in Med Ad News’ December 2018 feature “Ad-ventures in Marketing XI.” The offering powers a variety of AI activities for Intouch clients – chatbots, patient adherence programs, Veeva digital sales aid rep interactions – and that list is rapidly growing as brand managers seek out new ways to harness AI.
One area where AI can be a great deal of assistance is the SEO process, which Chase called “wildly inefficient.”
“Roughly 50, 60, 70, 80 percent of this process is back office, for most brands there are 10,000 search terms or keywords, and each one of those keywords needs to be classified,” Chase explains.
“What category, what theme, what is it related to, is it related to treatment, to research, to clinical trials? They’re going through and classifying and characterizing and analyzing all of these keywords.”
That process could take 100 hours per month, or maybe the SEO team has 100 hours budgeted per month, and 80 percent of those hours are spent on this back-office task, according to Chase. “It’s a task that the client has no visibility into, nor do they really care. They don’t care how the sausage gets made, they just want to see results.
“What they do want to see are better strategies, better optimization plans, better results, better ROI and better ROI presentations and methodology,” he told Med Ad News.
As a client test case, Intouch deployed Cognitive Core to handle the SEO terms classification process.
“The artificial intelligence does all of the processing, all of the analysis, all of the classification of those key words,” Chase says.
The result? “Those 80 hours a month are literally done in 10 seconds,” he states. “And then we spent an hour or two or three just checking the results of the AI. But it literally has saved us that much time.”
The agency can then apply the time they would have been helping to classify SEO search terms instead to strategy, the optimization of the
SEO plan, and more time on presenting the results, Chase says.
AI has also been used by marketers to predict patient events.
“Marketing teams have always asked questions about ‘how many’ patients – how many patients have the disease, ask for products, drop off therapy etc.,” Khedkar says. “AI is enabling them to ask this question at an individual patient level – Which patients will see a progression of disease, which patients are likely to drop off their therapy, which patients are not likely to fill their scripts? – these are all prediction questions based on a large number of personal attributes and past medical history. Having these micro-insights means that the right intervention and service can be provided, especially if the patient has opted in. In some cases, such algorithms are being deployed in conjunction with provider systems to get scale.”
The third area is automation of business processes. According to Khedkar, many decisions in marketing require data gathering and analysis to make a decision and this can be done by automating queries through intelligent chatbots rather than producing lots of dashboards that marketers often do not look at unless they need an answer to a question. AI is also being used to automate forecasts for many products and countries based on past data – making the process less human-dependent.
Although AI has been used for patient and physician programs, most of the use cases have been about physicians. There are two key reasons for this: the vast majority of pharma marketing activity is still directed at physicians rather than patients; and the data on physicians is identified and much easier to collect and integrate across sources than it is for patients. “As patient centricity takes off in pharma, we expect the patient-focused applications to gain more importance – for instance, the idea of optimizing interactions and services that is now being done with physicians,” Khedkar says.
One of the reasons AI movement has been faster on the R&D side than the commercial side is the size and availability of datasets that can be used to train learning algorithms. According to Khedkar, large R&D data sets can be found within companies, but for commercial data sets, “one needs to go for data across companies.”
“For example, if I want to predict whether Dr. Smith will open his email – and all I have is the 10 emails one company sent him, that is not enough data,” he says. “But if I pool data across the industry – as is done for prescriptions – there is a robust way to compute channel affinity for the doctor.
“Similarly, patient-level data – even if anonymous – provides a huge treasure trove of information on physician behavior as well: What is the treatment pattern of this physician across different morbidities, when do they tend to switch, do they have high rates of non-compliant patients, does the size of their practice have anything to do with that? All of this is possible because of industry-wide, third-party syndicated data – and marketing can leverage that.”
The disadvantage is that as data sources proliferate, many may not be available to pharma in the future even for purchase – and the ones that are available will necessitate an increase in investments.
According to Chase, AI can help shoulder the human cognitive load.
Cognitive Core is helping another client in the challenges with a depression drug launch. “They believe it’s going to be a blockbuster, at least $2 billion-plus,” Chase says. “But they believe their sales force is operating very inefficiently.”
One of the problems is that the sales reps are not writing their reports in line with the goals set by the company. “They’re not doing a good job of showing things such as, ‘In my last visit to Dr. Smith, I had this conversation, I emphasized safety over efficacy I talked about X, Y, and X things, and based on my last notes and coaching from my manager, I’ve evolved my talk track so I can hit on all the things I didn’t hit on last time,’” Chase told Med Ad News. “They’re not at all capturing their experience the first time. They didn’t do a good job at all of tracking the goals, and they didn’t capture at all what they needed to do in order to evolve the conversation and to hit on the points that the doctor wanted to hear next time.”
An Intouch team including Chase then went into the client and took a consultative approach in helping them map out the business challenge. The team took all of the field reports and fed them into Cognitive Core and used the AI to find instances where the sales people were not doing a good job of mapping to the goals, as well as identifying other opportunities for talking with the doctor.
According to Chase, this analysis process is expected to recenter and refocus the sales team as they are ramping up the marketing on the new drug.
Balagot also says some of his agency’s clients have turned to AI for their “maiden” drug launches. “In those situations, they often don’t have the full commercial infrastructure or customer support infrastructure built. So there are opportunities to use artificial intelligence to provide scalable 24-7 customer service to patients or HCPs.” Precisioneffect has actually deployed two AI chatbots for a couple of the agency’s clients to provide patient support when these companies did not have the full patient call center in place.
Looking into the future
What are the future possibilities for AI in pharma marketing? The first prediction, according to Khedkar, is how much more data will be generated, as data in healthcare has been exploding and will continue to do so, growing by 15 times from 2013 to 2020. “The more important fact is that it will be much more variety, not just volume,” he says. “Data on genomics, social determinants, individual behavior all will become available, integrated, and analyzable (once the appropriate privacy and ownership are decided, which can be a challenge).”
Additionally, the marketing mindset in pharma will change from being mostly “strategy” to focused on “execution”, which involves making decisions at a micro or individual customer level, Khedkar says. “Other industries already do this, pharma will learn to do it as well,” he says. “As the portfolio shifts towards large molecules and digital health – every patient becomes far more valuable – and it is worth getting each decision right, not just the average decision right.”
Individual customers – patients, consumers, and doctors – also will all demand a much better experience from pharma than they are getting today. “Getting a good experience involves not just a service mindset, but actual insight about your customer,” Khedkar says. “In 10 years, all the data will be able to provide it, and those that use it will have a competitive advantage. And AI is crucial to this, since it can automate the making of thousands of decisions – no human will ever be able to do the customization that is needed.”
Additionally, institutional customers – payers and providers – will demand even more value and evidence from pharma – so being able to predict the impact of a medication correctly will be a pre-condition to having it utilized well. “It will no longer be enough to say that your product works for 40 percent of the patients compared to placebo,” Khedkar says. “Providers will use AI to push back against pharma – so pharma will have to stay one step ahead of them.”
Balagot believes that Artificial Intelligence has the potential for improving medical communications to physicians, but there needs to be more done to perfect these programs. “I think by nature of the questions that come through in that arena, they oftentimes are much more complex and much more involved, so the natural language processing that AI affords, there might be some limitations in being able to provide the level of depth and detail of information to a physician in that type of dynamic,” he told Med Ad News.
Precisioneffect has also launched a branded Alexa skill to provide patient support. Balagot says this can be valuable for patients who are physically disabled or have motor skills challenges and cannot easily use cell phones or computers.
“Voice AI, in particular, provides a tremendous amount of opportunity to be able to give patients access to information in ways they may not have normally had,” he says. The AI can be used to search for information, schedule appointments and medication reminders, and order in medical supplies.
Of course, just because AI is hot, does not mean that every marketing program has to incorporate it.
At Ogilvy, Patel says the first thing his Innovation Lab colleagues ask a client is why they want to do AI in the first place, what are they trying to achieve, and then they try and find a use case.
“There’s a canvas that we try and use around what is the best benefit and use of a thing – is AI really the answer, or is it straightforward guided conversation? Are you doing something with images, which is a different thing? Or do you want to do something very sophisticated with data and images, structured and unstructured data? Do you want a dashboard and all sorts of stuff?”
There are a lot of things AI can’t do and shouldn’t do, and there are a lot of things it can do and should do, Patel told Med Ad News, “so we try and focus our clients on what do you think you should do, and what’s the benefit?”
Often through this analysis, they find out in most cases that what the client wanted AI for actually does not need it. Sometimes the client actually wants just a dashboard for data, thinking that AI could be plugged into that to do predictive modeling. “And then we tell them if they want that, we’ll need this much years’ worth of data, we’ll have to teach the machine over six months, the algorithm has to learn, it has to be tweaked, and then we can launch it in the marketplace,” Patel says. “And then the marketer says, ‘Actually, that’s not what I meant.’”
Rather than say “artificial intelligence,” what Patel and others at Ogilvy like to say is that they are providing “assisted intelligence.”
“What we are trying to do is give you the assistance you need, from a machine, that has some intelligence,” Patel says. “If you want artificial intelligence, it’s a whole different ballgame.”