Failure and Opportunity with AI in Life Science

Failure66% of AI projects fail. That’s the bad news. The good news is that in life science we’re not really investing enough or working on the important problems yet. The opportunities are still there. So we’ve got that going for us.Severence McLaughlin is the CEO of DeLorean Artificial Intelligence. He has some thoughts about what is actually AI, how it’s being deployed, and the impact it could make if companies stopped doing science experiments (investing in little test projects here and there) and had a vision for achieving business objectives at scale.We’re pretty good at RPA (robotic process automation) but are we helping doctors and patients yet? The value of AI (as opposed to advanced analytics) is the ability to make a prediction with a probability assigned to it: When will a patient going from diabetes to chronic kidney disease as her primary diagnostic code? When a physician sees that, what are the next best steps for that individual patient?Opportunities We have enough data on individuals to move beyond the one size fits all approach. Not… “Hey, we're going to take this data and it says this, and we're going to apply it to everybody. One size fits all. Another colleague would call that primitive AI, which means analytics. But now the ability is to say, okay, you're a diabetic, but you're eating a rabbit diet, you're doing yoga...Maybe we can get your A1C down by going through lifestyle.”Sev, I'm one cheeseburger away from a heart attack, man. We need to have some heavy medication. So it actually goes out to the individual person and has that capability. So hyper personalization.Sev extended that idea to the sales opportunity. How do pharma reps build relationships with doctors? AI can inform the next best action to take based on what a doctor is likely to respond to.I have wondered whether pushing info to doctors will lead to prescriptions based on… (I don’t know what.) I challenged him on that and he pointed out that many doctors are too busy to keep up with the latest and greatest. Wouldn’t you want your doctor to have that info? To know about the therapy that is most likely to work for you? In the case of a rare disease, how do you find the one doctor that can help the 1200 patients in the US that have it? Or help a doctor correctly diagnose the patient in front of her?What about clinical trials? Could AI help select the patients most likely to respond to your therapy? That seems a bit of a chicken and egg scenario, but if that were possible, it would make marketing the drug a lot more cost effective following approval.Getting StartedMy last question was about data readiness. The data is never perfect and never will be. Again, the vision needs to be driven from the top by the CEO to get all the parties involved working together. For applications around marketing and sales, Sev thinks 60-70% clean is good enough. Of course, you’ll want to do better than that for clinical applications. The way to get better is to get started.If you have thoughts on any of this, please share them in the comments. I’m here to learn. If nothing else, tell me what you think about my intro music. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit cclifescience.substack.com