Back to all articles

Pharma and Data Science: Knowledge is power

Sanofi’s Milind Kamkolkar is the first enterprise chief data officer in a major pharmaceutical company, and he has been outspoken about the need for pharma to adapt to the new digital world of AI, real-world evidence and prolific data. We caught up with him on the show floor at eyeforpharma Barcelona 2018

How do you see the current landscape for digital and data science in pharma?

The timescale under which we operate today is not the fastest. I bet two years from now the same conversations will probably be going on, just with different faces trying to make the same impact in different companies. You have to ask yourself how much of what we’re doing right now is truly impactful vs trying to marginally improve an already inefficient process.

What’s been even scarier is that we’re looking at digital being the utopian cure for everything when I actually think it’s the reverse – it’s becoming something like an anti-bacterial agent that’s developing its own resistance and pitfalls, and I think the companies that are going to win in this space with customers are the ones working on the antidote and scale. Because there will inevitably be fallout – and that’s ok. What’s missing is the ‘why’ are we doing digital and for whom does it benefit? For example, our knee jerk reaction is to outsource without thinking what we aim to achieve. We also don’t consider things like putting ethics into digital engagement driven by algorithm development.

What does being a chief data officer entail?

I had a pretty clean slate when I started in terms of what to focus on. Inevitably, this came down to a basic approach: do things better or do better things. As such, I first identified what the three things we needed to be excellent in were.
One is using technology or digital as an operating model – just making it the way we work, and embedding that new way of working in how we do our everyday operations. The insights and information we generate are assets with great value. So when I look at what a chief data officer should do, it’s really around helping companies not just be innovative but also be excellent in making compelling decisions at an operational excellence level.

That encompasses a number of things. It encompasses what I call the ‘oxygen of information management’, which is just doing the basic block and tackle of data identification, acquisition and management really well. For example, you should be able to answer the question, “How many people at Sanofi have engaged with this physician?” Sometimes we can’t answer that at all! No one knows. And that’s one of the basic questions you should be able to answer.

We also need to know where that data is. We don’t know who’s got it half the time, and we’re buying it over and over again. It’s not just about finding it but also then once you’ve found it determining who can access it, and then where it’s being used and what it’s being integrated with, and once it’s been integrated whether you can reuse it.

That’s the foundational stuff. I.e. do things better. The second thing we need to do well is the machine learning and big data space (do better things). How do we automate, are we even going to automate? Simply put, how can we store and access everything (where compliant and ethical) and analyse anything (where relevant and actionable)? That becomes a really important piece.
The last area is building up new capabilities in data science, engineering and software development. My feedback to the organisation was that hiring data scientists or classical statisticians or mathematicians is not enough. You need to build an ecosystem of players that has data science as a function not as a role. You need to have algorithm developers that know how to code, you need to have data journalists that know how to communicate the story behind the data, you need dashboard designers that are designing great user experiences through modern capabilities like haptic, voice etc.

What’s the hope for how these approaches will feed back to
patient outcomes?

We’re moving into an era of real-world evidence, and it’s going to be very interesting when we start moving from clinical endpoint to physiological endpoints to behavioural endpoints. All of those are an aggregation of different data that’s coming from all sorts of things. You can literally store anything now.

But I think what’s going to be key here is thinking about health outcomes rather than just patient outcomes, because it’s not always about the patient. It could also be something relevant to a caregiver looking after their parent or child. That’s equally as good. It’s about doing that work, knowing which customer you’re providing the right intervention for and delivering the right experience for, and progressing the correct behavioural changes necessary for that patient to get better – and yes, sometimes it will not include our medications.

What are the main barriers to this way of doing things?

There are two. One, not all physicians are ready for this. They’re getting better though. There are definitely some mavericks out there who are so frustrated with the current system that they just say “I’m going to build it on my own, I’m not waiting for anyone”. They’ve gone and learnt code and they’re building their own stuff, which is great.

Secondly, I think that when the timelines are what they are for us, the biggest issue is inertia and these artificial/business boundaries that we create. I’ll give you a very simple example. We have therapeutic area divisions within a company to help operations and impact, but you know who doesn’t care about that? Patients! But in our infinite wisdom we don’t give ourselves the opportunity for true and scalable co-creation and collaboration. Sadly, we end up repeating the same stuff over and over again without thinking that there’s a team just down the hall (even virtually) we could partner and learn from instead.

Are there any areas where approaching digital and data like this is already seeing some significant success?I think you’re seeing good applications of machine learning. The world of natural language intelligence is by far the most prolific application of machine learning in this space for us. For example using a speech-to-text interface for record-keeping instead of typing it out, which includes a risk of spelling errors. I’m not saying something like Alexa is perfect, but it’s probably a lot better than someone who’s really tired typing in physician notes at the end of the day. Mining through literature is another one. It’s still a complex field but we’re getting better at it, and through the use of outsourcing and open sourcing you’re seeing a lot of really great techniques that are free to purchase. They of course cost money to implement, but you can do it fast. I’m seeing that the organisations that have a software engineering team tend to be way more productive than those who outsource to consulting companies as the first step or are enamoured with startup without thinking through the business problem and so forth. It doesn’t mean you never use partners but rather leverage the right partner at the right time of problem statement identification and scale up. I think Sanofi has really gotten better at this and are evolving quickly to value internal staff execution as much as they value external partnerships. I call this the connected enterprise as opposed to the centralised or fragmented enterprise.

We’ve just got to look in a mirror sometimes and say, “Come on, what are we really doing here?” But it’s still hard. Throughout my career I’ve been told on numerous occasions, “Milind are you sure you really want to analyse that, because what if it’s negative?” Well, it might help someone’s life! But I find these conversations are tending to be much more productive now. There’s a desire to move. The skillset change is also not as costly as it used to be –these days people can learn data science in seven to eight weeks. We do that with our team now, I’ve got a bunch of guys who have been at Sanofi for almost 19-20 years who are asking how they can learn it and we give them the time to learn and participate in challenges to hone their ninja skills in data.

I think you’re slowly seeing this adoption. When you start getting leaders in the field who are walking the walk and talking the talk, and are not afraid to call out BS where it exists, that’s a good thing.


Published by  PharmaTimes magazine – May 2018

International digital analytics, data science or programmatics expert, and looking for a job?

Get in contact with us!