Mental Models for Health Entrepreneurship

Of course. I'm a neurologist, and I practised for 10 years in one of the main hospitals in Madrid. But I was always super enthusiastic about data — even way before I knew the concept of big data. AI, for me, was something in the movies, but I thought that big numbers tell a lot of truth. That was a kind of background to my career, and I was super interested in mathematics and statistics. So I decided to take the entrepreneurial way of doing things. I paused my career as a neurologist and went to Silicon Valley to do a program where I learned how to start a company. Actually, I started two of them, both in AI. And that's how everything began.

What I've seen is that when tech people try to jump into healthcare and create tools and devices for doctors, normally they don't work — even when they're super advanced technologically. The reason is probably that contact with the patient gives you some intangibles related to counterfactual thinking, emotional thinking, contextual thinking. You learn to distinguish things that are important when you're jumping into artificial intelligence.

I like to say that in machines I rely, but in humans I trust. A machine cannot deceive you, cannot disappoint you. And one of the reasons is that it's not conscious. It's intelligent, because it solves problems, but it's not conscious — because consciousness is related to morality. That's why in medicine we talk about clinical judgment. Machines cannot judge. They can take decisions, but there's a huge gap between making decisions and judging.

So that ambiguity — that binomial way of thinking: judgment and decision-making, reliability and trust, consciousness and intelligence — all those things are, with all my respect, very difficult to understand for pure engineers. You need to be in front of patients to understand what it means. When a 75-year-old comes into practice, not necessarily literate, not understanding all the fancy concepts we talk about, you have to choose the right words and touch the right way. Those very emotional things are the ones that help me in the way I approach AI every day.

When you look at many of the applications that have failed, they assume that the process of giving care to patients is A plus B plus C equals D — a process where you get signs and symptoms, then an MRI, then this and this, and you get to a result, which is also algorithmic. Then you apply that result and you get a treatment and a cure. That's a model of disease that is absolutely not true. It simplifies enormously how health and disease work, which is a very complex model that also has to incorporate the social standing of the person, the economic one, how literate the person is, how you can communicate, the emotional part, the psychological part. Those take big roles — they're not small fractions compared to the scientific part. They're even bigger than the scientific part. So combining evidence generation with all those attributes of the person is absolutely needed in the way you create the tools.

Yes, of course. I think we doctors are biased towards a few mental models that don't help the creation of new tools in a software or technology company. We're maybe too biased towards perfectionism. That's one of the main problems of being a doctor — this concept of improving a bit every day, this kaizen way of doing things, is something we don't necessarily understand. For us it's either very good or it's terrible, which doesn't help at all when you're dealing with a team of engineers trying to create something. They have this much more incremental, improvement mindset, which is needed. It took me a lot of time to understand how things work in that regard, and my transformation into the light side is not yet finished. Today I still struggle to say, well, today is a bit better than yesterday, that's enough for today, tomorrow let's have a new approach. I still take it too personally, too emotionally, when the tools are not good enough. And I think that doesn't help.

Also, we doctors are very good at what we do, but we don't have a lot of cultural understanding of how other things work — the financial world, the economic world, everything that surrounds companies, human resources. I've had to catch up very quickly. Sometimes I notice I don't have the background. I didn't do an MBA either — I just learned from practice. So sometimes I miss these things and I have to rely a lot on my partners, which is good. But sometimes I wish I had more understanding of the world outside the door of the hospital.

If you look at the way we do things, there's a trend you see within our company. It's a hundred-person company, with doctors, engineers and linguists — because we do a lot of natural language processing with medical records. And there's something we usually find funny: when you give a task to us doctors, we normally respond to it immediately, that same day, sometimes that same hour, instead of scheduling it, prioritising it, and giving a result later. I think that relates to the way we work at the hospital. When we get trained as doctors, we try to have everything solved in case an emergency happens. The list of patients needs to be clean as soon as possible, because then something could happen — and if it does, I don't want a lot of people waiting. But that's super counterproductive for generating pieces of software or methodologies that need a program or a schedule. So yeah, that works for me, but it's always something to improve.

It's a bit weird, because as you said, we're good at dealing with uncertainty. We're good at making the best possible decision with just a bit of information — and we neurologists are especially good at that. But what I've noticed is that, amazingly, at the same time, when we confront technology we judge it very thoroughly, very hard. If what they offer us as clinicians is not exactly what we have in mind, we tend to reject it instead of understanding that it's a continuous process. So we're good at being flexible with the biological world we deal with every day, but when we come to the human-created world, the tech world, we get much more strict. I don't know the anthropological reason why that happens, but that's what I noticed, compared to other specialties.

The mantra of Silicon Valley is move fast and break things. The problem is that here you cannot do that, and it has a big impact on innovation. Because innovation — this innovator's dilemma — is all about understanding that you're going to be wrong three or four times until you have something useful. The problem is that in healthcare you normally don't find that patience. The managers of the hospitals, the clinicians, even the pharmaceutical companies get super nervous when things are not ideal on the first day. They all claim they want to do innovation, but it's not really true. It's not because they're lying — it's because they don't understand that innovation is about failing. So I noticed that, I suffered that, and I learned from it. I learned to tell the right messages on the first day, not to raise expectations too much. That was probably my biggest mistake — I raised expectations too much, not because I was lying, but because I thought things were going to be faster. Then you confront reality, so you learn how to bring these other stakeholders to their reality: that it's going to be a journey where we learn together, and that's okay, that's the way it should be. And what I also learned is that if you do it right the first day, things are smoother, and the team trying to innovate is happy with it.

I learned that the good leaders are normally the — you know, in this division of red, yellow, green and blue — the red type of people. The red ones are the strong ones, the ones who don't give a lot of time to conversations but just make decisions and look at the pragmatic side of things. I tend not to be a red; I tend to be yellow, the person who wants a better world, not necessarily by looking at people but by looking at ideas and technologies. That's more the vision side of things. But then you need the numbers, and the numbers fall much more on the red side. So I learned that leaders need to have at least a strong part related to that. But at the same time, leaders need to be people the rest want to follow by themselves. So I'd say it's a combination. I don't think leaders have to be disgusting people. It's probably about making hard decisions — but the way you communicate to your teams can be a nice way. It's a balance. And it takes a lot of experience; you don't get that in the first five years, but it's something I've seen in some bright minds around me.

I learned a few things. Some of them you'll find in the typical Silicon Valley entrepreneurial books, which are useful — and sometimes it's better not to read, and just do the work and learn by yourself. That's also something I learned. And by the way, they don't always apply to Europe. That said, there are two things to consider. One is that the hunters you need for the first phase of your startup are not necessarily the ones you'll need in the second stage. The first ones are much more generalist. But in the second phase — one to a hundred, a hundred to infinite — you're going to need people who are more focused on one specific task and at the same time are more thinkers, with more strategic thinking, who don't necessarily jump on solving the task the first day but have the ability to reflect and create a plan. I learned that that transition is difficult. Things go very well at the beginning with the first ones, but then they start struggling when they have to stop doing and solving and instead sit down, plan, and recruit the right people to do it. That mindset needs to be changed as soon as possible, not the hard way.

The other thing — this was my Christmas message to the company, and it's something I tell everyone when they join — is: avoid politics. Politics appear way sooner than you'd think. You'd think that in a 150-person company... because at some point we were 160, but then we automated many processes and could go back to the magic number of 100. When we were 160, we discovered how incredibly fast politics grow in an organisation that's actually quite small. Managers start hiding what's happening in the company, because it's better not to be the one who brings the bad news. And that does a lot of harm. So that's one of my main lessons: be very cautious, and let your people know that politics are the perfect way to disaster.

I haven't written a book, but if I ever do about my entrepreneurial adventure, the title will likely be something close to Innovating Despite Europe, or maybe Innovating Despite Spain. Maybe the UK is a bit of an exception — we've probably had our best experience in Europe with the UK. But when you go to the rest, to Southern and Central Europe, things are very difficult for entrepreneurs, because decision-making is not based on rules. It's based on psychology. What drives the decision is not "this is A, B, C, you fulfil it and you're good to go." It's much more about a committee deciding whether they like you or not, whether they trust you or not, whether this is culturally acceptable or not.

It's so difficult to explain to the Americans — to our American investors — when we have the opportunity. It's not clear what you need to do to get an approval. For example, when you're trying to get permission to read the medical records of a hospital — and I'm saying read, not own them, just reading them in an anonymised, de-identified way, which is fully legal and GDPR compliant — in the US it's straightforward. You know what you have to do: you de-identify this way, you pay this amount to the IT staff who do the work, and they're good to go. But here in Europe, in some countries, it's very different. You fulfil all the criteria and then you get a no. And what's the reason? Well, the committee considers that this is not interesting academically. But what does that mean? It doesn't mean anything, except that as Europeans we're not dreaming towards innovation the same way they are on the other side of the ocean. So it's definitely something to consider when you launch a startup.

Half agree and half disagree, and I'll tell you why. I think you're very right, and that's the type of mindset you have as a doctor — understanding that a diagnosis that is not actionable, in the majority of cases, is not really worth it. It can even be counterproductive. It's difficult for people outside healthcare to understand that. I think it's the Cochrane phenomenon that said this. Even when it's very counterintuitive, sometimes knowing more can be less. That's the reason I never took a genomic test myself. When we were at Singularity University, we were about 80 people in the classroom, all super techie, incredible people. 23andMe and these kinds of services were super trendy at that moment, and everybody took the test except three people — and the three were exactly the physicians in the room. That tells you a lot, because we understand this actionable/non-actionable duality, which has to be taken into account. So I agree with you in that regard.

Now, there's another type of predictive modelling in healthcare that's not the future — you can do it today: treatment selection. What if I create a predictive model that tells you that in this particular patient it's going to be better to prescribe ipilimumab than the standard of care, and in this other population it's not? Things change if you have that type of cancer. And things change if you're the manager of the pharmacy, who's going to spend the money only on the patients where it's worth it. So that's where I see this happening sooner.

The way I see it, we already did something remarkable. If I die tomorrow, I'm okay, because I already brought something to the world that didn't exist: the concept that you can do clinical evidence generation, that you can create research studies by automatically extracting information from the medical records of hospitals at an international level. This didn't exist. There was NLP applied to medical records, but there wasn't the concept of having an evaluation methodology by which you can check that the variables are good enough — how it fits for purpose for research studies in, say, multiple myeloma at an international and multilingual level. So we created that. And by doing it, we created an ecosystem of companies that replicate our model — today you'll find six, seven, eight companies doing the same in different parts of the world, obviously less advanced because we started before, but they're good followers. So as a mission, that's great. And it's something I can tell my son.

Now there's a second stage, a second moment: I didn't create the concept, but I took it to every point in the world — to every hospital, every site, every sponsor. In other words, it's not that I created a new way of doing things; it's that I made the old one disappear. That's my next mission: to do things so that no one looks at data manually anymore, no one populates an Excel sheet with patient variables anymore, because it's taken for granted that the machines do it. That's where we are now. And growing is where the incredible business opportunity lies, because the bigger the network, the bigger the studies, and the bigger the pharmaceutical companies that get involved. The opportunity is massive, because at the same time, real-world evidence — observational studies as a concept — is growing massively. So we're coupling the growth of a company that invented something very innovative with the growth of the market we're in: the real-world evidence part of things, aside from clinical trials.

In our opinion, that's precisely not our biggest threat. It's actually an advantage. Because these large language models — you can chat with them and ask, "what are you?" — they'll tell you: I'm a generalist language model and not an expert in any field, and I don't intend to be. I intend to be the API to which you plug your expert database, and then you use my incredible linguistic engine. So it's not a threat; it's something that's helping us. Actually, we've been using the same technology that lies behind ChatGPT at Savana for years already — the transformers and BERTs and all these types of linguistic AI. So that's definitely good news for us, not bad news.

I'd say we have two threats. One: I talked about the growth of the real-world evidence market, but the question, with a business plan in mind, is how fast is it going to grow? In five years, two years, or ten years? There's uncertainty about the pace at which the drug agencies — the FDA and the EMA — are going to start accepting real-world evidence seriously, not as the small brother of clinical trials, but seriously. No one knows. These agencies are super bureaucratic, so that's a risk. The other threat is related to what you mentioned. We create these incredible databases, not just because we want to save human effort, but for something bigger: with bigger databases you can create better predictive models, because you have more variables. That's the real reason we do things — it's not to save data managers. The problem is that it's not clear to me that the market for buying predictive models has been created. You can't really sell many predictive models to stakeholders today. In hospitals and healthcare systems it's no one's job to buy predictive models. They're starting now, in a shy way, in the US, the UK, China and Japan. So for now we have to apply our technology to more descriptive studies and outcomes-generation studies, which is good because we automate it — but we'd love to live in a world where people buy predictive models.

It's a mix. There's something natural — what they tell me is that since I was a child I liked public speaking. I enjoyed giving ideas. So there's something in me that's always been like that. But at the same time, I put in an incredible amount of effort. People would be surprised to learn it. I watched every talk of, you know, Obama — everyone who's a good speaker on YouTube. I read every book about public speaking. I remember after being on call for 24 hours, super tired, instead of going home I went to a school of neuro-linguistic programming — how to convince people with your eyes, with your voice tone. I've done this for years; there's a lot of training behind it. So it's never easy. It's a combination.

This school of thought basically analyses what super-convincing people have in common, and they discovered it relates to the way you move your eyes and your body — the body-language things. A bit of hypnosis, maybe. It's a combination of different approaches to communication.

The first lesson would be: don't go to a public-speaking course, because public-speaking courses don't teach you how to public speak. If you want to learn public speaking, go to theatre classes. Because public speaking is about playing a character that is not yourself — and once you're on stage, it lets you connect with the idea. When you connect with the idea, you forget about yourself, and when you forget about yourself, you start sounding natural. Everything is smooth, you're happy to be on stage, and the nerves disappear — because you're not thinking about your person, you're thinking about the idea you're trying to convey. Being able to merge with the idea is something actors master. So taking improv classes, theatre classes, would be the way.

I don't mind saying it: I 100% play a character when I'm on stage. Something that's happened to me many times — people want to meet me after being on stage, they come and ask a couple of questions, and they're super disappointed. They're like, wow, this guy has nothing special, he's super normal, his answers aren't really special, he's not really very tall. They're very surprised. It's what happens when you meet actors or actresses — outside their work, they're not that impacting anymore. I feel that's what happens to me. I grow when I'm on stage. I become another person. I'm a character — this doctor who talks about these things this way. If you look at the emails I send every week, I say a lot of bad-sounding words. I'm kind of an arrogant person. I'm not that person — I would never do that in my real life. It's just a character I created, that I invented.

That's part of the stories my arrogant character writes in the weekly newsletter. That's not me. I would never say it in public — but I did it. I went there once. It was an incredible palace in the surroundings of London, and I had a good morning and learned a few things. Of course, many bright people there. But I thought — with all my respect for their incredible CVs, which are way beyond what I'll ever get in my life; I was literally the dumbest person in the room — these guys are not really changing the world. They can spend a morning there talking because they don't have to pay 100 salaries, and because they don't have a list of 200 hospitals complaining about the technology you're deploying. And even deeper than that, they don't plan to write papers with predictive algorithms that are going to be implemented in real patients in the next five years.

So I cannot spend one more morning here. I have too much to do. When I'm older, and when my son gets older, I don't want to tell him I have an incredible curriculum in science because I'm the author of countless publications. I want to tell him I saved lives — and you don't save lives going to seminars, talking about theory. So I declined. I don't want to be disrespectful, but there's a mission that goes beyond that.

A hundred percent. Probably the biggest mistake is to try to say things that no one dislikes. That's the perfect way towards a boring life, a mediocre career, a mediocre project. By definition, the only way to do things that are interesting, impactful and charismatic is to be disliked by many others. To always go 50/50 in life. The only way to have 200 hospitals that trust Savana is to have another 200 that hate us — because we're stealing their data, they think. By definition, the only way to have a tribe, as I like to call it — a tribe of doctors who believe in AI — is to have another tribe that, if they could, would make us disappear, because they think AI is crazy, harmful, naive, whatever. It took me some time to understand this. It's probably the best tip I could give: to convey messages that are original, without fear of being judged. Because you're not interested in the big group that's like everyone else; you're interested in the small group that follows you.

Yes, I love Seth Godin. He's probably the most influential person in my life.

That's very interesting. There's something similar I read from Nassim Nicholas Taleb, whom we all love — who, by the way, is the kind of person not everybody likes, and he doesn't try to be. But he's a genius. He says something like: the best way to spot a bullshitter is that they tell you the 10 rules to do this, the five ways to do that. True people normally tell you the don'ts. If you want to be a fit mother, or if you want to go to Cairo by yourself, the true ones will tell you the five things you shouldn't do — because that's a subtle texture of empiricism, as he says. Whereas the five do's are much more like the general messages everyone could say, not from experience, but from having read it somewhere.

There are a few ideas. But you're talking about money, about a wealthy, successful company — and maybe it's too late, maybe there are already players there compared to when I had the idea a couple of years ago. But I always thought that a media agency using AI to spot fake news in healthcare is going to be incredibly useful. It's going to be worth millions, because we're heading directly towards a world where you don't know what's true and what's not. It's like a battle of AI — so having your own army of anti-bad-AI in healthcare is probably very promising.

Yeah. I never watched television. For younger people that's obvious, but it's super simple, and it's the truth. When people ask why I succeeded, I really think the real reason is that I never watched TV. I never lost one second of my life watching TV — not even as a teenager. Never. I don't know anything about the popular TV shows. Here's an anecdote: my wife is the main character of the most popular show in the history of Spanish television. She's a very famous actress. And I never watched one episode. It connects me with the idea that all the time I spent doing other things ultimately paid off.

I like to be very divergent. I like to throw the ideas out — don't keep them; you can always channel them later. People tend to think divergence is a problem, a lack of focus. But from my workshops in creativity, I learned that the worst thing you can do with an idea is to judge it before expressing it. First you express it, you throw it out there, you put it on a blackboard, then you judge it. But people usually do it the other way around. So probably what took me here is that I never kept a crazy idea inside — because from the simplification of those crazy ideas came the good ideas. It's like in the fashion world: you see them with incredibly fancy dresses, obviously not going to the street, but it's the simplification of those which ends up out there as prêt-à-porter. With innovative ideas it's the same: you throw them, you judge them, you simplify them, and then you bring them out to the world.

Yeah. I'm surprised there's now a lot of cognitive research around this — that when we're bored is when we have the best ideas. Even the Greeks said so. When you get bored, you analyse your personal story, you do a kind of self-assessment, and so many cool things happen. That would be one reason. The other reason — I was reflecting on this recently, I've never been asked this question, so it's quite new for me — I think the real reason isn't even that I spend more time reading or working. The real reason not watching TV helps me a lot is that it made me think in an original way. What characterises watching television is that you think like everybody else thinks, because you're getting your information from the same source. By not watching TV, I wasn't exposed to that culture, so I kept my original way of seeing life intact — not in everything, but on many occasions. So that's probably the reason.

No, I don't think so. Those are the kind of people we don't appreciate — the normal ones. You're never going to say, "this guy, wow, he's so normal, I love him." You talk about the weird one, the strange one, the one who knows things nobody else knows, the one who did an incredible trip, the one with a rare hobby. Connection is about being unique, not about talking about what everybody talks about.