As you research ways to incorporate AI into your life sciences commercial initiatives, watch this On-Demand webinar to learn:
  • What’s unique about applying AI in life sciences companies
  • How AI can strengthen your commercial strategy
  • Finding the balance between human oversight and automation
  • Where to begin and how to prepare



Webinar Transcript

Yigal Uriel

Thank you for joining us for today’s webinar on achieving commercial excellence in life sciences using artificial intelligence. I’d like to introduce two of the leads from our AI group. Gil, who heads up our AI product management, and Matthew, who’s in charge of AI strategic services. Gil comes to us with a wealth of experience in product engineering, cloud computing, and business intelligence. And Matthew comes to us with a background in strategic commercial optimization across pharma and medical device companies. Without further ado, I’ll pass the presentation over to Matthew.

Matthew Van Wingerden

Thanks Yigal, and thanks to everyone who’s joining us today. Excited to have you with us, to talk about AI. Today, just at a quick high level, we wanted to cover some of the contexts for using AI in life sciences and get into a few specific use cases that we’ve become familiar with in our experience. Then hopefully by the time we’re done, leave you with some practical tips and tricks on how you can start to incorporate AI more, either into your organization or into your personal work.

Before we dive into the details of AI at life science companies, I think it’s worth taking a bit of a step back and looking at the history of AI. Because while we often think of it as something that’s very new and only been around for the last 10 or 20 years, the foundations and the basic technology has really been in place for almost 70 years. We’ve really had AI, or at least the foundational technology of AI, since basically the time that we’ve had computers. Even some really advanced things like deep learning have their roots in as early as the 1950s, when people were already building technology that could recognize visually the difference between a circle, a square, and a triangle.

Now, throughout the 60s, 70s, and 80s, there was a steady progression of I would say proof of concept use cases that were really interesting and kind of going into new areas. Again, based on some of the same technology that had already been started in the 40s and 50s. We all recognize that, in the last 20 years, we’ve had an explosion of use cases of AI. Some of them very cool, very well publicized, that we’ve seen in the news, on TV, whatever it is.

It’s worth taking a second just to understand what’s different today from the 40s and 50s. I think we’re all aware of the fact that there have been advances in the low cost production of really fast computer chips, but it’s not just that. It’s also we’ve been able to get better at distributing computing load, which is a little bit of a technical detail, but it’s actually made our ability to do AI speed up a lot faster than you would just expect based on kind of the advances in chip design and things like Moore’s law, which many of us are familiar with.

Because of all these advances, we have had these really interesting and well publicized use cases, but also very ubiquitous ones that are in our day to day lives. Whether it’s something as simple as spam filters in our email or product recommendations in Amazon, I think we all engage with AI more often than we give ourselves credit for, or more often than we realize. There’s also more that happens behind the scenes, that we wouldn’t even have visibility into, such as a lot of complex use cases for image recognition, that are more B2B rather than customer facing. Then also security technology, which allows us to predict and prevent security breaches before they happen. These are all very important and they impact us day to day, just in ways that we don’t see.

There’s a lot of use cases in our day to day life across industries, but it’s worth taking a minute and talking about the advances in life sciences. I think the area where we can highlight the most use of AI has been used across the board in life sciences, but really in R&D is where we see the coolest, most interesting, and probably most well publicized advances. Whether it’s for optimizing clinical trials, or doing predictive drug discovery work, increasing the speed to regulatory approval, AI has really been used very well in our own industry.

It’s obviously been used a little bit less in the commercial side, and I think there’s some good reasons for that, but as we’re going to get into, I think the stage is really becoming set for the use of AI in commercial excellence, not just in R&D or manufacturing like it has been in the past.

When we look at what we need to have available or have set, for us to be able to use AI in commercial excellence or in commercial, in life sciences, first of all we need the technology. I think as we’ve already discussed a bit, it is certainly available. It’s not only that AI is becoming available as a standalone technology or as something that needs to be added on, it’s becoming integrated into a lot of existing tools, whether it’s analytic software or cloud computing, AI functionality is being, slowly, over time, introduced more and more as almost a core component.

So we have the technology, and we also have the data. This is just a little bit of a high level metric here, about how much is being spent on building data in healthcare, in the healthcare industry, by 2020. Obviously it’s a very large number, but we can all sort of anecdotally say that we have a lot more data available every day to us, right? The life science industry has been talking about building up big data warehouses and data lakes and all that, for probably 10 years now. But in the last 10 and even five years, I think we’ve seen a lot more advances getting down to the level, in some cases, of individual patient identification, identifying when there are new patients with certain indications, making out the patient flow in a very detailed way. There’s just a ton of data that’s available on the market out there.

Then finally we have a very high complexity in our commercial situation. This may seem at first as a detractor, but as far as AI goes, it’s really an opportunity area. In situations where we have a lot of data and we have a lot of technology but low complexity, AI is almost not needed. However, in life sciences, we have very complex decision-making processes, with many different stakeholders, including patients and caregivers and doctors and pharmacists and other doctor’s office staff, and so on. We have a lot of different ways that we, as a manufacturer, may interact with them, either through direct to consumer marketing in the US or in-person business to doctors around the world. A lot of different sources of data, whether it’s our sales or something like IMS track share or volume data. We have a lot of different pieces of information scattered across a lot of different channels, about a lot of different stakeholders. This type of rich complexity is exactly where AI can really make the most impact.

So we are starting to see, in the life sciences industry, the ability to do it. More importantly is, is there appetite to do it? This is just sort of a high level report put out by PwC about sizing the prize for AI gains. They predict very large gains in GDP due to AI. Well, what we can tell you more specific to the life sciences space, but a bit more qualitative, is we are in a lot of discussions with our customers about how they are going to use AI to do one of two things. It’s I guess what you would expect. One is either to use their existing resources to drive more growth in their brands or in their devices. Or two, which we’re seeing quite a bit as well, is trying to maintain the same level of commercial effectiveness with more cost effective organization. So shifting resources from costly in-person channels to digital channels, but doing it without the loss of any quality. That’s where AI can really come into place, to help people do that in a very careful and judicious manner.

Because we have this foundation of technology and this data and this complexity–and I think people are really starting to see this ability to make practical applications of AI, which Gil is going to get into in a minute–we’re finally seeing a shift in the commercial side in life sciences, from sustained interest, which we’ve had for many years in using AI, to actual investment, where people are trying to find concrete use cases and shift from just sort of it being something we will do in the future, to something that we are doing right now. With that, I’ll turn it over to Gil, who as I said will get into a little bit more specific ways to use AI.

Gil Blustein

Thank you, Matthew. When talking about AI in life sciences, especially in the commercial side, we usually talk about the usage of AI in order to shape or impact the brand strategy. In most cases, the brand comes up with the strategy to define the targets, the channels, messages, the cadence, and the timing to engage with doctors. Usually they’re doing so while bearing in mind the costs of each of these channels and trying to reduce that cost, as each channel has a different implication. This strategy is the plan that the brand puts together to execute their marketing material, while using the available marketing tools towards the target audience.

If we can see that the range of marketing approaches from mass marketing, where we treat all our HCPs equally and there is no personalization, basically each HCP receiving the same campaign, regardless to their experience, their practice, their location, age, and so forth. Through campaigns where we apply some segmentation. That segmentation applies some personalization, but on the aggregated level. It means we apply HCP to each segment by credentials, by specialties, or whatever we define as a reason to cluster. Each of these segments we basically treat differently, and we apply some personalization of that segment level. As a result, HCPs that belong to that segment are benefiting in that campaign. So traditionally, that is the approach that brands are using to run campaigns.

However, the other approach is missing the ability to treat each HCP separately. All the collective data that we have obtained through our marketing tools, through our sales tools is being ignored, and it’s not being used. Whenever we’re using it, we’re using it at a cluster level. Me might move HCPs from cluster to cluster over time or create new clusters. That happens, but we’re not treating HCPs in that cluster individually. We’re basically treating the whole cluster from the HCPs.

This is where AI is changing the language of how we interact with our target, with our HCPs. With AI, the concept of a segment changes from the aggregated level to the individual level, the HCP itself. We’re using HCP specific data, their characteristics, their history, to generate a campaign that is targeted for that specific HCP. So with AI, we’re actually changing the language and moving from a segmented marketing to a one to one marketing, and it’s fully personalized.

Now that we understand the benefits of AI, the question is: how can we apply AI, and where can we use it? There are a few types of areas where AI can help us. The first question you might consider using AI is regarding your investment. Where should I spend my next dollar? This question becomes very important when sales force and marketing departments have more tools available to them across multiple channels. It’s not just a single channel dilemma anymore. The traditional approach of a response curve diagram of the channel or sales or prescription does not really represent the reality. It doesn’t really work.

In a multichannel world, the problem is much more complex. Here is how AI can help you. For example, we have a few customers that we have worked with in the past to identify optimal combination of emails and in-person product details, in regards to sales. Or in other words, how can we predict the sales as a response to email sends and in-person product details? What we found are some very interesting results. First, we did this on HCP level data, where we have HCP level sales, such as market in the US or in Japan, where we have plenty of HCP levels, or single HCP facilities that we can actually use their sales data. What we actually noticed is, when you’re sending an email, it’s always better than not sending any email at all. We also noticed that sending three emails, for example, doesn’t really impact the sales, because we already reached the top sales or the top predicted sales with two emails, so the third email or even more emails doesn’t really change anything.

What’s more interesting was when we added the in-person product details. What we noticed is, when you actually add visits, that pushed the sales even further. The question we wanted to answer is what is the right combination of email sends and in person product details, to get the optimal sales without actually exceeding the costs or expenses? For example, if you look at the two areas highlighted on the chart right now, you notice that both of them were predicting the same amount of sales, but one is being achieved by sending two emails and having two visits, and the other one is being achieved by sending one email and four visits. Obviously the second area is much more expensive than the first one, because you’re doubling the number of visits. That is something that the brand team will take into consideration.

Another question where we can apply AI would be: what should we do with this next? In this case, you would want to use AI not just to tell you the number of times you should engage with your targets and over what channels, but also what is the sequence and set of actions you should take in order to interact with your targets? You can also use AI to try and come up with the optimal cadence and timing. It’s not just a question of what channels should I use, but what is the sequence of these channels and what is the timing? Should I wait one day, or should I wait three days? And so forth.

Last is the model can actually tell you what is the next best message to send. What do you need to do, in order to engage with your targets, with you HCPs, over what channels, what time, and what message? That combination of the channel, the timing, and the messages could be simulated in a similar approach that we have right now on the screen. What it actually allows you to do, it’s not only to understand what has already been communicated in the past, but you can also get a glimpse in what is being predicted to be the next steps. That is a big advantage.

Another benefit of this, this is a live model. Meaning the data that is being generated is being updated all the time, and that journey will be updated for every HCP once the data is refreshed.

The last question that you can ask yourself where can we use AI or how can we use it is, how do I get this done? What do I need to do in order to execute on the strategy, and how can I actually have AI to help me? At the end of the day, the main goal is to have your sales force team and your marketing team act on the strategy that you just implemented, but you cannot guarantee that your team will actually execute on this. This is where AI can actually help you.

Each of your reps has its own routine and has its own way to engage in their targets. For example, if you can predict your rep’s location and where they’re going to be, you can actually generate the optimal action plan or action route that allows your reps to engage with your HCPs within the strategy, in the most optimal way. At the same time, you can also predict the time that the rep and HCP should meet, based on the historical data that you collected, what day they should meet, what time they should actually get to meet, what day, what time should they actually send an email, and so forth. This information can help your sales force team to use their time wisely. It’s mostly trying to avoid wasting valuable time, where the rep is sitting at the HCP’s office, waiting for five minutes, waiting for the doctor to be available so they can actually present and talk to them, where they actually can be somewhere else.

At the same time, you can use AI to identify opportunities to reach nearby HCPs or high value HCPs that were not planned on the calendar. For example, if a rep has his own calendar planned for the next week, knowing where they’re going to be, you can also present opportunities that the rep can engage with those doctors that are nearby and are still valuable to your strategy.

Now that you understand where we can use AI, the question becomes what do we do with this? How can we actually act on this generated data? What we’re seeing from the models is that there are several approaches that we can actually implement. You can choose to take either one of them. It basically depends on the use case and where you want to land.

The first approach is what we consider the manual approach. This is where you use the output of AI as a consultant. What it means, the output is generating a report, and you consume that report and decide which part of the report you are going to implement on your strategy. This is the approach you pick and choose your information to see how you fit it into your strategy and decide however you want to implement on your strategy that you have in place.

On the other hand, there is another approach, which we consider the automated approach. This is where you would integrate the output of your models into your marketing tools. You let the model basically pick the strategy automatically, and that is automatically integrated into your sales force team and your marketing tools and eventually change the strategy as it generates a new output.

The main difference between the manual approach the automated approach is really depending on the use case. The manual approach is usually suitable for use cases where you can implement small changes. For example, if you are operating on a specialty pharma, you might have fewer targets, and then enforcing the output manually might make more sense, since you already did it with a small number of HCPs and it will be easier to actually implement. The other way, if you’re working in primary care, permutations of messages, doctors, segments, and so forth is so large that it would be almost impossible to implement this manually, and you would need to find the solution. This is where the automated approach can actually help you.

The third approach is where we usually find our customers feel more comfortable and what we usually recommend our customers to implement. That is what we consider the assisted approach. What it means is, the models are integrated into your marketing tools as the automated approach, but at the same time, you have the tools that allows you not only to visualize what is predicted to happen, but also you can control that output by applying constraints or applying new logic on it. You’re becoming a part of this integration, a very integral part, and you’re confirming and changing this output the way you desire.

All right. For example, going back to the question that we discussed earlier, on what should we do next, where we show the simulated route or journey of an HCP or a segment. With the assisted approach, not only can we visualize what’s going to happen next and we provide a glimpse into the future, we can also allow the brand team to control and mold this output with some logic on this constraint. For example, if they decide to switch messages or to replace messages and provide some logic for that to happen, they can do that. If they decide that they want to space out the cadence and replace this two days, for example, with five days, they can do that as well. This approach allows you to control your strategy, based on information that you provide, and change whatever the output of the model is.

However, what we’re seeing happening slowly in the market is brand teams are pushing more and more for fully automated integration, as opposed to the assisted approach. We believe that over time, we will get there, and we recommend that the brand will take a step back and learn to walk before we start running. Mainly because this is a new domain in life science, where we want to make sure everybody feels comfortable with the concept of using AI. Either it’s the brand team with the strategy that we have generation, or the sales force and marketing team with the execution. Until we get there, we recommend to the brand teams to use the assisted approach as part of the process and learn from it.

In the future, I predict that we will actually move to the automated approach with more brands. Since the market will become more mature, we probably will have even newer technologies that will emerge and support this even better, but until then, we would rather have brand teams walk before they run. With that, I’ll pass it back to my colleague Matthew to wrap it up.

Matthew Van Wingerden

Thank you, Gil. Before we end our presentation for today, we wanted to spend just a little bit of time trying to give you some practical tips for how you can make AI more reasonable or more practical for your organization or yourself to implement. There are three things that we wanted to highlight.

The first is the upfront time investment. We don’t want to undersell here the fact that AI can be very complex. The terminology is new, for a lot of us at least, and the capabilities required are quite high. So it is worth it for the organization to provide some training opportunities for people who are in an analytics function or commercial function. Also trying to conscientiously build a center of excellent, whether that’s a formal center of excellent or just adding AI specific mandates to existing parts of the organization, which might just be focused on innovation or new technology development, centralizing or creating visible points of capability for AI in the organization is really important. Also for you as the individual, making sure that you’re prepared to have these deeper level discussions can be helpful.

Again, this is not something where we all need to go back to school and have a lot of expensive education, but just doing a little bit of preparation of either reading a book that’s kind of introductory to the concepts of AI or just spending some time kind of learning the market and studying the different vendors that are out there, or what competitors are doing, just so that you’re able to speak with ease about the different concepts and the different advances that are happening in your state, can really set you apart from the rest of the group. I think all too often people are afraid to engage in AI. While it is difficult to engage with AI without any preparation, I think what I’ve been surprised at, at least in my own career, is how a little bit of preparation goes a long way here. It’s something that’s very useful, and people are really looking for the capabilities to engage in AI. But a lot of people just aren’t taking that time to prepare themselves.

The second part is being able to partner with the data science team in a different way than you would necessarily partner with the other parts of your organization or other vendors. The biggest point here that I want to make is the classic project plan of aligning on requirements upfront, doing the implementation or the development, and then checking at the end to make sure everything is right really does not work as well for complex data science based deployments or AI deployments.

The biggest thing here is there will be frequent iterations between the data science team and the commercial leaders, or the business leaders. One thing that you can do as an individual, as you’re going through these frequent iterations, is maintaining your own KPIs, your own sort of gut check, is what I always like to call it. One thing that will happen is the data science team will come up with something that makes a lot of sense, is very well validated statistically, but there are simple things that you know, as somebody who’s more familiar with the market, that are easy to check. The example I always use is when we do optimization of the number of calls, in-person visits versus emails, that Gil was describing before. Just remembering what is a practical number of times that a rep can visit a doctor in a week or a month is a really useful thing to pressure test the outputs that you’re seeing.

Then the last one here on the partnering with data science, again, as the individual, if you are able to translate the concepts that are being brought up in all these different iterations with the data science team and sort of help your colleagues navigate through the complex results that are being generated by the AI models, this can have a ton of value. It can really help both you and the data science team more quickly come to a solution.

Then finally, we just called it advocate here, but really this is celebrating success or kind of promoting AI within your own organization. I would say here that, broadly speaking, there are two conditions that are really important for any life science organization that we’ve seen be successful in deploying AI sort of ahead of the curve. One is there is kind of this well-defined network of AI focus colleagues. These are people in different organizations. They could be in IT. They could be in commercial excellence, they could be in some sort of center of excellence. Whatever they are, but it’s a group of people who are supportive of each other, understand AI just enough, so they’re kind of in that translator realm, and they, as I said, are a network that can help bounce ideas off each other and therefore aren’t in a vacuum.

Then the second thing is being able to create a clear sense of career progression, related to AI. There are almost never any AI specific roles that I’ve seen in any organization. In fact, that can be in some ways limiting, so I don’t necessarily think that’s the right solution. But just as an organization, being able to highlight AI success, either in things like performance reviews or business reviews or in annual innovation awards, this can be really helpful. For you as an individual, it’s really important for you to view this as kind of an opportunity to advance your career. It’s not necessarily about the great work that you’re doing, which of course you would want to highlight, but it’s also about highlighting the advantages of AI in general. It’s kind of evangelizing about the concept of AI and how it can be made very practical in your organization.