What happened to Babylon Health after reaching $1bn+ revenue? And how Ali Parsa is continuing his mission to fix healthcare for all with AI startup Quadrivia.

0:00Hello and welcome to another episode of Riding Unicorns. Today we are delighted to have the return of Ali Parsa. Ali was previously on the show in January 2022 2022 and is now back to tell us more about his new company. Quadrivia. So Ali, thank you so much for joining us. It would be great if you could firstly introduce Quadrivia and then we're going to have some questions about your experience with Babylon and then we'll obviously dive deeper into the new business. James, so good to be back with you, my friend. I've been following the podcast and I can see you guys have been going from strength to strength.

0:38So congratulations, well done and well deserved. If you really look at James, what is the fundamental problem in healthcare is that it is 50 % of the world population has zero access to healthcare. And about another 50%, the rest of us who are lucky enough to live in so-called developed countries, have real problems either with accessibility or affordability or the quality of what we get, right? There is no one, unless you spend a fortune, that can get accessible, affordable, high-quality healthcare at will. So the question for me is why is that? And I think the fundamental root cause of that is a structural imbalance between an elastic demand in healthcare, my kids get sick, I will spend any time, any money, any effort to fix it, and a very constrained supply of clinical services.

1:35And if you have that in basic economic 101 of imbalance between supply and demand, you will end up with inaccessibility and unaffordability. You remember when COVID happened in UK and people went for toilet papers and had something as commodity as toilet paper. We had prices went through the roof. We couldn't find it in our supermarkets. But the problem with healthcare is that this has been chronic and it existed for centuries. So what we are now seeing with the advent of artificial intelligence for the very first time, and what gets me super excited, is that we now have the promise, not yet the reality, but the promise of creating an abundance of clinical supply, clinical skills, right?

2:27Now, I don't think we can do that today. I don't think AI can do a human's job today, but there is about 20 to 30, 40 % of what a clinician does, a nurse or a doctor, there is highly routine, very repetitive tasks and processes that they actually dislike. It's a huge burden on them, but like everything else that is routine, that is repetitive, AI likes. So think about this as you go to surgery. Before you're going to surgery, your hospital will call you once or twice to make sure you're prepared. So when you go in, you're not wasting your time or their time. And that phone call needs to be made by a nurse, sometimes even your doctor, so that any questions you have is answered.

3:11And they take you through a protocol. When you arrive, you do your surgery. If it's a knee surgery, let's say it takes an hour or two. But then when you come out of hospital, especially in the United States, they're responsible for 60 to 90 days of your return. If you return, they have to fix you. So they have a massive incentive in order for you not to go back and to monitor you at your home. So in average, a hospital will make three to five phone calls to check on you while you're at home. And by the way, that should happen everywhere in the world because that's good practice that people should be checked on when they leave the hospital.

3:47Those phone calls, again, will take up to 10 minutes each. Normally, you make five to seven phone calls before you reach a patient. All of that can be done by AI. And we now have a queue that can do all of that extremely well and do a very good job of it. And that's what we built. That's so succinctly put, and I can't wait to dive more into it because that is the focus of today's show. But I just want to, while we've got you, because last time we had you on the show, it was still during your Babylon days. And so I just want to touch on and hear the story of Babylon in your own words. And I mean, it was such a fantastic company.

4:26The idea seemed like it was right. You had amazing people working in the company. The timing was really unfortunate with the market, with the market sort of imploding. funding drying up. But I'd love to hear, and so would our listeners, your perspective on what went wrong. So I think you were telling me that the last time we talked was in January 2022, if I'm correct. And that was just two months after we had gone public, sadly to a SPAC, which is a special purpose vehicle that everybody was pushing at the time. No one's fault, my own stupidity. Our people wanted to see liquidity. So I agreed to doing this SPAC and tried and tested way of going public.

5:15And when we agreed to do that, it was back in February. By the time we got the paperwork done, we were ready to go in June. And then I think SEC came back and asked for some minor changes in our document. They didn't say it's necessary. They say it'll be good to do. And somehow between our lawyers and our team, they decided to make those changes. That meant that instead of going in July, we kind of had to resubmit. They very quickly approved it, but that was September. And you know, in September 2021, the SPAC market imploded. And then we became public, but didn't get the$300 million that the SPAC was supposed to give us.

6:03That's not their fault. They didn't have the money. They were getting the money at the same time from others. People didn't give them the money. Nobody's fault. The market just went. They say that if when you're young, you blame everybody else when you're old, you blame yourself. But once you're wise, you blame no one. And this is really one of those circumstances that everybody did their best and nothing could happen. Then we went back. By the time I was talking to you in January 2022, the truth of the matter is by then we were dead. We just didn't know it. Because every single SPAC company that you know that was of the same time that didn't get their money died.

6:42Avalon could not have been an exception because that was a structural situation. You're public. Every hedge fund, shortage fund is shorting you. They're not shorting you as a company. They're shorting the entirety of the sector. And that was really sad because it was happening at the time, Hector, that our company was doing really well. So if you put the metrics of the share price performance away and look at the real metrics of the operating of the company, our revenue was tripling here. We went to do a SPAC when we were doing$80 million. In 2021, we did$320 million. And in 2022, we did$1.1 billion of revenue.

7:28And by the time we went into administration, in the middle of 2023, we were already at$1.5 billion of revenue. Our losses had reduced down to$4 million,$5 million a month, I think,$5 million a month. So we were almost done. We just didn't have the money and we were structurally. the thing. We had shareholders who couldn't give us the money because now we were public. Some had a structural problem. They couldn't invest in a company in which they own more than a certain percentage point if that company is public. And we had other shareholders who just didn't have the money and sadly perhaps didn't behave as well as they should have because they thought that if they wait a little bit, the share price goes even further down and they can take more of the ownership, and they miscalculated.

8:17And as a result, the company didn't get its equity. Then we had a debt holder who, again, from their perspective, did what is supposedly rational, which is to say, wow, the equity holders can't own the company. The company is doing one and a half billion dollars in revenue, soon going to become profitable. It was worth five billion only a few months ago. Why don't we get to own it for free? So they didn't allow us to bring any debt in. Normally, if new debt comes in, wants to go above them, but they didn't let that happen. And that meant we couldn't get debt. We couldn't get equity. We were checkmated.

8:53They thought they can take us and merge us with another company that they had trouble with. Small company never grew unlike us when they invested to 300 million in us. We grew from 50, 60 million when they invested to one and a half. This company stayed at 40 million, right? And wasn't doing anything. The CFO of that company at the time thought that he can mastermind the business of taking a bad company and a good company together and just revaluing both at the same time. But in reality, on the day they were supposed to do the transaction, somebody in there, they were just selling themselves to a Japanese conglomerate as a hedge fund.

9:30And that Japanese conglomerate presumably told them, what are you guys doing? This just makes no sense. So they pulled out. And then when they pulled out, they were supposed to give us some money. They owed us, but they didn't, and they left us with four days of money. So none of these things, sectors, should have happened. They could have all been avoided, but they were an unfortunate series of a perfect store. Now, did we also make mistakes? Absolutely. One of the things I learned now is you will always need to have two years of your cash flow in your pocket, because when things go wrong, if you have a couple of years, things will be fine.

10:07never take debt as an equity company because everything's fine when debt is okay but when things go wrong debt holders can be really really difficult to deal with so on and so forth right but that's life I mean we can delve into it really more but I don't want the lesson of Babylon to be that you see they were too early they were too ambitious they were this they were that it was none of that you're actually executing really well the company was doing exceptionally well But it just, at the moment where it needed another inch, it didn't get that inch. And life is a game of inches, right? I mean, you sometimes get your inch and sometimes you don't.

10:46I remember I was a banker or in the banking team at Goldman Sachs that was serving Amazon in 2001 crash, 2000, 2001 crash. And we just couldn't find Amazon any more money. By complete luck, Morgan Stanley fined them a German bank who gave them$600 million of convertible while Amazon was months away from going bust. And then that saved Amazon, that$600 million, and then another$500 million that came almost a year later from Tiger saved Amazon to become what it became. And I am absolutely sure if we had survived in Babylon at that time, once the market turned back again on the back of AI in 2023, just as we were going under, we would have been a behemoth today.

11:38But that's life. Life sometimes gives and sometimes takes away. It's a really tough story to hear, and it's super unfortunate that it happened. And have you seen anyone building what Babylon could have become? Is anyone rebuilding Babylon in a different way? No, in the same way that no one's building Revolut. I think we were building an exceptional company with a direct-to-consumer, but serving the payer's needs and managing people proactively better. When just a couple of weeks before we went under, we got the first set of results from our cohorts of value-based patients from our insurer. Our deal with them was, quite rightly, that they will wait until the end of the year.

12:29At the end of the year, they wait another three months to see how the accounts settle, and then we take any savings. That showed that we were getting around 10 to 14 % shed savings on our very first or second year, depending on which cohort of patients they were, among our population. It's huge, right? Considering that we are taking from the insurers what they were already paying for their patients. So that was pure cost, no profit in it. And then we were reducing that by 10 to 15%. and nobody else is doing that today at the scale Babylon was doing it I have no doubt somebody will come and do it again maybe even at some stage we might go and do it again you never say you never know what life throws at you but it was a fantastic business doing really really well I'm super proud of it but it wasn't to me it is what it is you move on no sure thank you oh yeah James yeah I just wanted to add you know we also had Alex Chesterman from Kazoo on here and you only have to look at the Kavanaugh share price to see what could have been because I think even in the last year Kavanaugh is up over 80 percent it's above its August 2021 valuation now and you know they had similar issues I'm not saying it was exactly the same with debt and everything but you know a lot of challenges came in that 2021 2022 era and so it's important that everyone understands that things do go in cycles and macro environments do impact things so i think it's just a really useful lesson to be able to draw out because we're often speaking to so many founders that are in the process of building their business we don't get to have too many conversations on a sort of um diagnostic level of what happened so i think it's really healthy to discuss it and It makes sense.

14:25Alex also made the same mistake I made. Alex went public through a SPAC. All those companies went under. Because it's obvious, you're public, you have a massive gap in your share price. All the hedge funds are rallying to destroy you because they're shorting it, all the short ones. It's a great trade for them because if everybody shorts something, that price will go down by definition. But listen, to me, the message to the founders who are listening here is not to be fearful of what can go wrong, because so many things can go wrong. I was three, four weeks ago, I was having dinner with one of U.S.'s richest people.

15:09He's in his 90s now, a wonderful human being. And he told me a story about his very first deal. and his mother, they were sitting on Shabbat having dinner with his father and his father thought that was a really bad deal. And his mother asked his father to let him do the deal because she said, look, there's no amount of analysis that tells you what your best deal or your worst deal because at the end of the day, your best deal, your worst deal has a lot of luck associated with it. What matters is have you done your best? And I can in all honesty say we did absolutely our best. In Babylon, operationally did extremely well, but we failed in getting the financing, right?

15:55And it is what it is. Sometimes you checkmate it and you checkmate it. Ali, moving on to Quadrivia, you touched on a few specific workflows around patient communications that you guys can automate with AI. Can you expand on those and the grand vision? So how, you know, I have no doubt you are looking to build a company of enormous scale. How do you build a platform once you've built for these initial workflows? So what we're really building, if you want, in the old world would have been a human resources company. We want to build that abundance of clinical skill supply where any clinician can pick up the phone and as if they were hiring an assistant, a junior, a personal assistant, to tell them, go and do the following things for me.

16:54So Hector, imagine you call me and you say, hey Ali, go and call all my patients who are visiting me tomorrow and make sure they're good. Call all the patients that I saw today, make sure they're taking, and I get prescribed medication, make sure I'm giving them medication. All the good things that a good doctor or a good nurse should do, but right now they don't have time to do. But if all of that was available at very low prices at scale, that's what we like to build. I mean, Quadrivia, if we get it right, if we don't get it right, somebody else will make it. and I actually believe many companies will make it, will be a clinical process outsourcing business on par with the process outsourcing businesses that did so well up to now.

17:48The reason they did so well was because they were often doing low-skill things, pay the payroll, I don't know, do some basic back office thing. You could get people in other countries to deliver that. We take care. Partly you have to be local. Partly you have to speak the language. Partly you have to meet the regulation. But partly it's very, very hard to scale a call center of nurses and doctors, which who wants to be a nurse to sit in a call center and make thousands and thousands of phone calls? I mean, it's a soul-destroying job for a nurse, right, or for a doctor. So that's why people could never scale those.

18:26and we think with AI we can scale this to enormous scale. If we are again lucky, if the inches go our way, if not, somebody else will do it, but we will give it our best. Yeah. And how are you viewing trust in AI within healthcare? Do you come across naysayers who think, you know, if something's going to go wrong, we need a human to be responsible for that. We don't want an AI running wild on our systems with access to patient data, with access to healthcare and clinical data. How do you navigate that? So, look, most fraud happen by humans. Most bad actors are humans. And this idea that I don't want AI or I don't want software or it used to be digital health, you'd have any conversation with anybody and they would say, I don't want a digital health company, a digital health to kill me.

19:25And look, that just never usually happens, right? It was very, very rare. And when it happened, it was global news, right? In the same way that a Tesla car crash is incredibly rare and humans crash their car all the time. But when one happened in the middle of Minnesota, everybody hears about it. I mean, why is it news that a Tesla car just did something wrong? Who knows, right? But that's what it is. So trust matters in healthcare. We spend a humongous amount of time making sure our AI is trustworthy. But what it does is like, this is like a series of processes that are very clear. So what we build is we build our main agent that is talking to you.

20:15At the very same time that she's talking to you, there is 30, 40 other safeguarding agents that are listening to the conversation. And they are listening to make sure that if something goes wrong, they intervene or interject or stop the AI from talking. That's why when I play you our AI, you often see that there is a little bit of a lag because in that two seconds of lag, there is a huge amount of work that is going in the background to check that AI. One of the reasons we haven't brought it into market at scale yet, and one of the reasons we don't raise a lot of money is because we don't want to be rushed to bring it to market at scale, is because think about what we're doing, Hector.

21:03Imagine you're going into a world that you're not having a co-pilot that a human can check and fix all the time, but you're having a real-time conversation with no human supervision with a patient directly. That is a huge ask. Autonomous, real-time, on a clinical subject with a real patient. So that requires a massive amount of customizability, controllability, and those are the things that we do. And to be honest with you, we focus initially on things that can not go wrong that much. Like we have an agreement with one of our clients to make 15 ,000 calls a month to remind people of an appointment they have, right?

21:55That reminder is really important because every miss that somebody doesn't show up costs them a huge amount because it's a procedure that is costly. But it's clinically is not a very big deal that happens in that stage. Have you ever heard one of Quadrivia's conversations? Let me just show you one of the conversations we were having with a mock patient. This is not a real patient, so I'm not betraying anything. But this is one of the mock patients. but this patient is acting like this is a real phone call taped to a person. Hello. Hi, is this Alex Smith? Yes, it is. Hi, Alex Smith. I'm calling on behalf of Dr.

22:38Rice. I'm Kiyu, a clinical AI assistant, and I wanted to follow up after your hospital discharge. Is now a good time? Now's a great time. Great. How are you feeling after your knee surgery? I'm doing okay. My knee feels pretty stiff still. and I've noticed a bit of swelling. Some stiffness and mild swelling are normal during recovery. On a scale of 0 to 10, where 0 is no pain and 10 is the worst pain imaginable, how would you rate your pain? Pain is around a 4. The swelling seems to be getting a bit worse, especially around my calf. I see. Have you noticed any redness, warmth, or tenderness in your calf area?

23:17Actually, yeah, a little bit. My calf feels a bit tight and sore when I press on it. Thank you for sharing that, Alex. Are you experiencing any shortness of breath or difficulty breathing? My breathing feels pretty normal. But you get it, right? So the point is, none of this was scripted. She was only asked, and I can demo that, we can set it up and just say, hey Q, go ask James about his knee surgery and give them your phone number and she will call you now and have the same conversation, right? But she reacts to what she hears, right? And she asks supplementary questions. She figures it out.

23:58But when it comes to judgment, she doesn't diagnose. She doesn't say, hey, here is what is wrong. She basically has figured out the human needs to be in the loop now. Let me just connect to the human. And probably for every 10 calls, one or two, a human needs to be in the loop. So we've taken the burden of eight of the calls out from the people. This is phenomenal, right? I mean, think about this. It really is, because the more you can do around that, the data and the documentation of the post-surgery experience in this example, it's a huge time save, absolutely. And we want the humans to be adding the real value that they need to provide, which is when someone's got a problem.

24:40So it's so interesting. And Ali, what springs to mind, following Hector's previous question and then hearing the demo, is actually customers might be even more willing to share personal information with an AI because they don't feel judged. And, you know, I think everyone will have experienced how they work with ChatGPT and maybe tell it things they might not tell friends and family. So do you think that's a big part of what this can offer is actually more insightful information where maybe users might be embarrassed to tell a human. 100%. And not just embarrassed. The last time you went to a doctor, she probably was rushed.

25:23You probably had a lot of questions to ask or things you wanted to talk about that you didn't just want to do because you thought that I don't want to take the time. You're of a generation that is far more demanding, far more knowing what you want than the generation previous year that was very reverent towards the medical community. So that is just a reality. Well, this conversation that I just cut off could have gone on for 40, 50 minutes. We would not stop it. We could ask any source of question. We can go on as long as you want for as many subjects as we want. Obviously, this is a clinical AI, so we cut it not to have non-clinical conversation.

26:05But this sits on the same large language models that you could use to inquire about your legal situations or your, you know, where do I find my pharmacy or whatever it is that you want to ask it. Right. So the capability are enormous because it sits on the large language models that billions and billions of dollars are being spent on. It is highly customizable because every doctor, every nurse changes what they wanted to do. I could just pick this up and say, Q, call James, ask him how he's doing about his knee surgery. Meanwhile, check out that he also can come and see me and do this and do that.

26:47And he will just follow that instruction and get that done. And eventually it is highly controllable because what we can do, one hospital may be okay for Q to talk about subjects that are not related to your knee surgery recovery. And another may not want it to. We want to say, as soon as you say, yeah, my knee's okay, but I got a cold. One may say that, look, I don't want you to talk about the cold at all. Refer to the GP. She can follow those insights. And what you can do in Europe is sadly very different than what you can do in the United States nowadays. And so that controllability come into it.

27:22And you mentioned LLMs. I mean, because it needs to be pretty accurate and it needs to have clinical knowledge, does it require a lot of rag and compute? Like, how are you thinking about managing margins and how does that impact your pricing model and things like that? Just to get inside the entrepreneur's brain on it. Yes. So, look, normally the way we try to work is to figure out our pricing model is actually really interesting. Because I think there are so many companies right now that are selling AI agents as if that was a piece of software. $10, they pretend it's$10 an hour, but in practice when they add like setup fees and management fees and time that you pre-buy but you don't use, that will count.

28:15the average to a customer cost was as high as$19,$20 an hour, right? I mean, at that rate, the nurse is$40 an hour. You would better give an employment to somebody, right? The way we are looking at this is that we are telling people, we will do a job for you. Give us all your post-surgery calls. We will get those done. And we may do them in a hybrid because the AI may fail at some times. You just heard the AI say, let me get somebody to call you. We will be happy to hire those nurses and make those calls for you ourselves and deliver you the end job. So what we try to do is say that, look, if we do it with pure AI or with a hybrid model, we can give you anything from 30 % to 40 % savings, sorry, 30 % to 40 % of the cost.

29:10So 60 % to 70 % saving on what it costs you today to deliver. And we try to be at 50 % gross margins or so, because as you know, there is massive operating costs, research and development costs, everything else that adds to it. But I think this has to be a true way of reducing costs and increasing accessibility in healthcare. Otherwise, it's a cheat. And Ali, how are you building the organization this time around? You've had a ton of learnings from, you know, you've been an entrepreneur many times over and built huge companies. This one is truly in a new era of AI. I mean, Babylon was an AI company, but the speed of development these days with tools like Cursor and Codex, you have to build differently and hire different profiles.

30:01So what's changed in the way that you're building Quadrivia from an organizational perspective and a hiring perspective? So I'm a big believer that you need very small organizations right now. this is the time in which you have to move very fast and you need small organizations to give you the structural advantage of being able to move fast you know the old adage that if you want to go fast go alone if you want to go far go together i think going fast is now a matter of its strategic importance at the time that everything is moving so fast and it is incredibly hard to be able to move fast if you have a huge number of people in your organization.

30:43So we try really hard to optimize the number of people we take in our organization to the point that when a skill set we need is done, we let that individual go and hire somebody for the next skill set we need. So we've been brutal at staying at around 20 people, right, in our organization for now. So, so that's one. Two is, look, we are all learning how to do this. And anybody who tells you otherwise, they really are BSing you. And this is a world in which BS is a structure, because people want to raise money. So they want to sound like what VCs want them to sound with. The VCs are in this world that say, unless you're growing hundreds and hundreds of percentage a year, you're not growing, right?