Tempus AI, Inc. (TEM) Earnings Call Transcript & Summary

All right. Thanks, everyone, for joining. I'm Matt Strom with Morgan Stanley Investment Banking. Really excited to have Tempus with me today. Eric, CEO and Founder, sitting here with us. We'll go ahead and jump right in. Right before I do that, the disclosures are available on the Morgan Stanley research disclosure website, morganstanley.com/researchdisclosure. If you have any questions there, please reach out to your Morgan Stanley sales representative. All right. Great. So with that, maybe we'll just go ahead and jump right in. Appreciate everyone being here today.

Eric, maybe you could just take a second to sort of start out and walk people through a little bit the basics on Tempus, I think maybe most interestingly start, you've got the intersection of these 3 really interesting businesses, and you could talk about how Genomics, Data, Applications sort of fit together and what you guys are building?

Yes. So I mean at a super high level, Tempus is focused on this idea of AI-enabled diagnostics, which basically is rooted in a theory that AI comes to health care through diagnostics. You could imagine AI comes to health care a bunch of ways. We tend to think that it comes through diagnostics because diagnostics sit at the center of almost every major decision people make in health care. So doctors will order a blood test or a CAT scan or an MRI or a genomic test then decide what to do. And so we got focused on this idea of seeing if we couldn't wrap AI around those diagnostics, which basically means kind of connect the diagnostic result with clinical data for those patients. And then kind of build large models that make sense of all that connected data. And we -- and those diagnostics then help physicians make data-driven decisions. So originally, we were just wanted to be a tech company, and we thought, well, we'll build this technology to contextualize diagnostics, starting with genomic sequencing in cancer but we went to the big genomic profilers there at that time, namely people like Foundation Medicine, and they refused to give us any of their sequencing data. So we had to build a lab. And because we were able to contextualize these diagnostics and our tests were smart, our lab began to grow really quickly, and so today, we're the largest sequencer of cancer patient in the United States. So we have a big Genomics business where we sequence patients in bill insurance in a growing space. But off that flows a tremendous amount of connected clinical molecular data that we license pretty broadly to biopharma. 250 clients are licensing our data and at year-end, it was almost $1 billion worth of data licensing agreements. And then downstream of that is our third business, we call apps, which is essentially because in order to connect all this data, we've had to literally build these bidirectional connections to about 3,000 hospitals in the United States, which is about 60% of all hospitals. Over time, as more and more people develop algorithmic models and solutions, we think we're a good distribution channel for that. So we started doing a little bit of that, but it's still quite small in terms of revenue, big in terms of potential.

And maybe you could spend another couple of minutes on the sort of 3,000 connections that you have. It's a significant portion of the hospitals and academic centers in the U.S. I think that's -- as I reflect on the Tempus story, you guys invested in that over a 10-year period. There's not a lot of folks who've done that to date. Maybe you could just spend a little bit more time on how you made that decision? And maybe most importantly, hospitals are notoriously difficult to deal with on this stuff so how did you convince hospitals to have a bidirectional data relationship with you?

I mean so we started as a tech company. So one of the principal differences between us and other people is, our competitors -- and there's a very small number of people that kind of are very big sequencers of patients, us, Natera and Guardant, Foundation Medicine, Caris really, those are the big 5. But they're all principally labs. We really started as principally a tech company. So right from our inception, we began going to people saying, look, we don't want to just sequence patients, we want to actually contextualize these tests, and so we need the data for the patient we sequence. And so we had to convince hospitals and their IT departments and their lawyers to basically share that data on an individual patient level. What's happened over time is as we become such a large sequencer, we now have about 3,000 hospitals that are in that relationship with us and the number of individual patients has piled up to being quite large. And so one of the reasons we have a sizable Data business that other people don't have that also sequence patients is the molecular data in and of itself is not that interesting. But understanding rich molecular composition connected to who are these patients, what drugs are they taking? How are they responding, that's very interesting. But you have to win these hospitals over in units of one and collect patient data in units of one. So during the IPO, as Matt knows we just say to people, it's like mowing 3,000 lawns. You can't mow them at one time and there's nowhere you can go to get them all mowed. You have to literally mow them one at a time. And so anyone who wants to compete with us -- and there are people that are just starting to try, they will have to go do the same thing.

And so as you think about you alluded to earlier that you have this -- so we have this really important relationships with hospitals. Those have taken a long time to build. They're really unique in the ecosystem. I think on the Genomics side, you're really utilizing that relationship. Obviously, you've got the Data business. On the sort of AI and Application side, you've got some applications there you're building and going to deploy on your own. You've also got the ability to let others get access to your customers to that sort of network. So how are you guys going to think about that as you go forward? And sort of how do you evaluate what you're going to execute on versus where you're going to partner and partner with someone else on?

Yes. I mean I think in the near term, just the way the kind of revenues flow in order to -- if somebody said, I have -- I believe in generative AI, I believe in large language models, I want to build a business around that. You basically would need 2 things. You need proprietary data and proprietary distribution. In our case, proprietary data comes first, right? You can't build the model unless you have data. So we've had to go out and convince all these people to give us their data, and that has just piled up over time. And so I think in the first iteration we're kind of generating lots of proprietary data. We're amassing these massive data sets. We're doing all that model development and pretraining and building a data business. But over time, the back part of that, which as I mentioned, proprietary distribution becomes equally compelling. It's not -- because right now, we're still in like the model -- it's like the model development stage as opposed to like insight distribution. But over time, I think insight distribution becomes a really big deal. And so we would say to people like even though today, 90-plus percent of our revenues are made up of Genomics and Data and Applications is quite small, I would suspect maybe 10 years from now or some point in time, our Applications business could be the majority of our revenues, maybe at some point, the vast majority. Because you will have these cycles where people like us build large amounts of data, all of a sudden, all kinds of people in the life science ecosystem realize, "Oh my God, this is really valuable, they're licensing it. They're producing insights and they're producing models, they're producing products and services of benefit." And then because we're -- we added something like 500 connections last year alone. So over time, if we can get to like 80% or 90% of the U.S. health care system that's connected to us if you go develop something really interesting, we become a distribution channel. So you've seen us already sign some number of those deals we signed one with Artera and people like that. I mean that people are developing models and saying, "Hey, Tempus, I can't reach this many doctors. Can you help me reach that many doctors." And again, I think over time, that probably becomes the vast majority of our revenue. But we don't -- our current businesses are growing quick enough and are exciting enough that we're kind of heads down on this part of the journey.

And just maybe 1 last question on that. On the sort of hospital side, you've got these Data relationships. Like as you start to I guess, sort of sell more down that channel? Like does that change your go-to-market motion there over time? Or how do you think that actually looks? So you sign up Artera as a partner, you're going to go sell that test. You're obviously enabling the distribution, but -- how do you think the sort of selling and go-to-market motion evolves on that?

It's a chicken and an egg. It's like we're connected to a lot of hospitals, in order to build the connection, you have to convince a hospital that whatever it is you're offering is of a high enough value that they should want to do something they don't want to do, which is give you data. They're innately looking to not do that. So you have to convince them that it makes sense. And it's a bit of a chicken and an egg, like these things kind of pile up, the more value we offer at some point, I would envision hospitals are just kind of contributing all their data to us because we have products in oncology and cardiology and immunology and neuropsych and so on and so forth. There's a tipping point where people are like, you know what, I'm just going to let's just -- let's have you become an AI partner of ours because we're in this relationship, and we want you to help us improve clinical decision support. So it's a little bit like many of these models and insights we can develop, but there's far more that we can't develop, right? The world is filled with tens of thousands of researchers. And so I would suspect over time, again, we'll bring more and more of those onto our platform. But one of the challenges with our -- with the Application side of our business -- in the sequencing side of our business, there is well-established reimbursement. So we run a test, we get paid. On the data side of our business, there are very large clients with deep pockets, namely big pharma that wants to pay for that. On the Application side of our business, it's a little more unclear because AI algorithmic diagnostics aren't routinely reimbursed by the U.S. health care system. So we're developing all these algorithms. We're deploying all these algorithms, but we're not getting paid for these things, other than small pockets. And so I think that's just starting to change, and I think that will change pretty dramatically over the next 5 or 10 years. And that's why I suspect that side of our business goes up.

Maybe we can go back to -- just to sort of work ourselves through Genomics for a second. That's the bulk of your revenue today, as you said, the place where there's the best and most clear reimbursement. You guys just announced an acquisition -- or announcing closing acquisition for Ambry, which took you into a different part of that testing paradigm. Maybe you could just talk about the Ambry acquisition sort of how that fit into your Genomics business? And then what your sort of goal or intention is for that business over time? Is it to have a full suite of testing across oncology and other subsectors? Or is it to stay in certain high-value areas? Like how do you think about where you're going to take that business?

Yes. I mean, well, the first thing, if you're trying to generate like really rich insights and you want really rich molecular data to power those insights, you need the right kinds of molecular data. So the question is like, let's start in cancer in the U.S., what is the right kind of molecular data? And basically, in cancer in the U.S., there's 3 big buckets of molecular data. There's hereditary risk, am I at risk of disease? There's, I have disease, what drug should I take? And there's, I took a drug and how should I be monitored. Is my disease coming back? So basically inherited risk, the treatment and therapy selection, which also called compositional profiling and then what's called monitoring and MRD or minimal residual disease detection. So you're like, those are your 3 big buckets. So we -- 1 of the largest players in that space want to be in all 3 of those buckets. We want to have a very big hereditary screening business, therapy selection business and monitoring business. And so we've made investments in all those areas to make sure we've got all the right kinds of molecular data flowing into our platform. And then from there, again, we're going to turn that into data products that we can monetize and applications we can deploy. And some of those applications might be like, hey, you're at risk, so you should do this. Some of them might be -- you should have taken this drug or this is the standard of care and somebody by accident deviated and made a mistake? Or here's an opportunity that you should consider. And so the ability to kind of which we can do and we do, do that, the ability to go into the EHR in real time, flag a non-small cell lung cancer patient that wasn't profiled for eGFR by accident, deviated from the standard of care, someone made a mistake, and we can catch that and correct it, which we do every day at scale is pretty wild. And so that eventually will be paid for.

As you think about -- or as I think about the trends in -- that are sort of driving some of the Genomics movement now, I think it's you've got more and more drugs that are characterized by a mutation. So a certain patient as a mutation, they should get a certain drug or therapy regimen -- therapeutic regimen. I think you've also got this phenomenon where the cost of a genome continues to come down, AGBT last week. There's more news on that front. Obviously, the sort of changes in compute -- cost of compute. Like how do you sort of think about the next phase for that business as you see the cost of a genome continue to come down greatly. And you've got more and more therapies that are personalized or meant to be in certain patient populations.

I think what -- so it's kind of complicated for tech investors because the health care is scary, it doesn't work the same way. But in general, if you think about it, first of all, the U.S. health care system is about $5 trillion of spend, something like that. So it's roughly the equivalent of the next 19 largest countries. So it's like 330 million people in the United States and 6.7 billion people. So it's kind of crazy, so it doesn't work like Amazon, where you have like rush to the U.K. You don't have to rush anywhere. This is -- the U.S. is like literally most of the population here. Within that, oncology is a significant piece of the spend. So we're -- even though we're only in oncology in the U.S., that's like a $1 trillion space. So it's not small. These kind of AI-enabled technologies and tools and all of the corresponding products around precision medicine are really only at some small amount of maturity in oncology in the U.S. So this is the -- the very tip of the very tip of the iceberg. So likely, what happens is that precision medicine, which is just starting in the U.S. in oncology becomes pervasive in the U.S. in oncology and really is routing every patient the optimal therapeutic. And then those benefits will then people be like, "Oh, I want that in cardiology. I want that in immunology, I want that rare disorders" and you'll start to see that proliferate. But it has taken longer than most people would have thought when the human genome was sequenced, for those benefits to show up because the system is like literally ridiculously broken. So what should take 2 years takes like 14, because the system is designed for nothing to ever happen.

Maybe we can go to the data business for a second. And I think maybe a good place to start there would be -- we've talked a little bit about the data business. You've talked about your pharma customers. But could you just maybe quickly for this audience like explain what are those customers actually using your data for? And why is it not something that they've had for some long period of time for many, many different partners?

Yes. I mean, again, so up until us you didn't have -- we started from this idea of let's contextualize diagnostics to do that, let's connect rich amounts of molecular data, comprehensive molecular data to clinical data. So what's the molecular profile of these patients? What's their DNA, what's their RNA, like what's going on? And then what drugs are they taking? How are they responding? So -- and let's build that from the clinical flow. So meaning let's build that data set from patients that were treated a week ago, a month ago, 2 months ago. So we started building up this data to make our tests smarter, not to build the data business. We didn't launch Tempus thinking we'd have a data business. We launched Tempus thinking with a really smart diagnostics. In order to do that, we get to retain a copy of all that data and de-identify and use it any way we want to. And then pharma companies started coming to us being like, I need that data. I'm designing a Phase II, and I'm wondering if I should have a biomarker. I'm designing a Phase II and should I have a companion diagnostic. I'm designing a Phase II and I'm worried I'm in gastric cancer, and I should be in bladder cancer, like basic questions that they couldn't answer from cell lines and mouse models and blah, blah, blah, that our data can answer. So if I would have said to you 5 or 6 years ago -- and you've been in the space a while, I would have said to you we're going to build a $400 million data business, you would have been like -- I left the room you have been like 0 chance, 0. So people back then thought you'll never get people to pay you this kind of -- and the answer is like it will probably one day be a $4 billion business. I mean, it's just it's -- pharma budgets are going to migrate from 100% biology and chemistry to some percentage data. I don't know what that is, 10%, 20%, 30%, some meaningful number, and that will kind of create this new data market, not just in oncology but again, another disease areas as well.

That's actually -- your point about pharma budget is a good one. There's been obviously a lot of noise and pressure on the pharma budget side broadly. The tool space has been under pressure. I think even a lot of smaller data peers of yours, or if you could call them peers, have been too. You guys have a very large backlog, $900-plus million in your Data business. You had an NRR of 140% last year. Like, I guess, maybe you've explained it, but maybe put a bow on just sort of what allowed you guys to continue to grow nicely, while pharma budgets have been under such pressure? Like how you sort of lived in that critical zone, I guess, for pharmaceutical companies?

Yes. I think, first of all, we're still -- again, we're in a part of the journey where people are just realizing this data can have enormous value. And the value it has is it essentially illuminates the inefficiency of their current programs. So like with all the pressure that is going on from Washington, okay? This is like 2% pressure, okay? In a world where there's something like 60% nonsense. So 2% pressure, 60% nonsense and then some amount of real than profit. Profit, real, I don't even know what. So like, let's call it 20% profit, 20% real, 58%, I don't know what, 2% pressure. So like that's the world we live in. So like it's not like so much to feel bad for Pfizer or BMS or Merck, like they will just take all the 58% that no 1 really knows what's going on and just start to shrink that down. But I suspect the bottom line, regardless what happens to the top line, the bottom line will end up being roughly the same. Everything else in between is just whatever. And that's just the way the process is. I mean it's an unbelievably inefficient process where companies that are incredibly sophisticated like Roche can spend $2 billion or $3 billion on a TIGIT trial and have it fail at the very end. And so the idea of saying, well, I'm going to license $10 million or $20 million or $30 million of data to avoid a $2 billion failure is kind of both core and fundamental and small and rising in a way that we are just unaffected by the larger pressure of competitive pricing dynamics. And so I think -- I suspect that will be the case for some period of time. I don't see -- I mean, we've been growing rapidly. Our Data business grew -- I don't even know what, 60%, 70% last year in the midst of a tremendous amount of pressure. The far bigger pressure is not from federal pricing pressure, the far bigger pressure is that biotechs were just literally annihilated. The capital markets shut down. And so coming out of COVID, biotechs were flushed with a ton of cash. But then when that reversed itself with high interest rates, they literally ran out of money. So we had dozens of clients that wanted our data and just simply had no cash, maybe more than a dozen. So I think that's been the bigger pressure. And I think as the capital markets eventually start to kind of come out of that, interest rates come down, people are looking for risk, they can't make money other places, some of that money will go into biotech, and then I think it will be some tailwind for us. But we're pretty immune for the most of the stuff.

I guess talk about -- as you -- I know it's hard to probably have an average, but if you thought of your average pharma customer, where are they sort of at? Like what inning are they in, in sort of adopting this more data drive mindset to drug development, drug discovery?

I would say like kind of first batter is on -- we're second batter -- or second batter, first inning.

And so how is that -- maybe talk a little bit about how the cell works to those clients? Are these sort of larger strategic initiatives? Are they discovery area by discovery area? Or how does that sort of work...

I think the good news is that like the at all these big pharma and biotechs, people, I think, have a genuine appreciation -- people that make drugs understand the value of the data we have. So what I'm trying to say to people like you've never heard this kind of data, they know what a BAM file is, they know what BCF is, they know what a DICOM file is and the digitized H&E. These are bioinformaticians, they're computational biologists. These are people that use Python and R, they're sophisticated scientists. So they understand the value of the data, they just have never been able to kind of get their hands on it. And so it's hard to figure out how to incorporate it into their practice. So I think that's why we're so early in the journey. But fortunately, for us, we don't have to convince people the data has value. And also, we don't have a ton of competitors. It's not like people don't license our data and go somewhere else. I said it's like 99% percent of the time or something like that. If they don't license our data, they just don't license data. We just have this unique data set right now that if they want this kind of data, they're licensing it from us or they may not have the budget for it.

Maybe we can -- so we have about 10 minutes left, maybe we can switch to applications and a few other questions. So as you think about the Applications business, so you guys have sort of AI infused across your business, but within Applications, as you mentioned, you've got certain algos there that you're starting to monetize and build, et cetera. Maybe you could just, again, sort of double-click a little bit on what you're building there today and what you think -- it sounds like reimbursement is 1 of the things, what you think needs to unlock there for that business to really start to grow and contribute revenue growth over time?

Yes. I think it's back to my point, it's a bit of a chicken and an egg. You need lots of -- so if you look at, for example, we developed an algorithm. So we run about 300 million ECGs in this country, electrocardiograms. Roughly 3% of patients that have been ECG that are told they're fine, have a heart attack or stroke within a year. So clearly, that's super bad. So we developed an algorithm over with a lot of time and money that predicts 1 of the largest contributors to that, which is a condition called atrial fibrillation or AFib. We've built that model. It took -- cost a fortune, time, data, all of it, took it to the FDA, years, got approval. Once we had approval, though, we had -- there was no reimbursement mechanism because like literally, it doesn't get paid. So you have this algorithm that's high enough quality that the FDA approved it as a medical device, a pending -- a 40-year-old medical device, the ECG equipment, but it doesn't get paid for. So you need lots of those. And then what's also just starting to happen like -- so about a month ago or 2 months ago, CMS or Medicare said they would pay $126 for that algorithm, so which caught us off guard. We didn't think that was coming.

In a good way.

In a good way. But now like we were flat-footed. I mean, we were in the roadshow together. I'm like people are like, when is that coming? I'm like I don't know 3 years, 5 years, not 4 months. So that showed up. So now we're going to hospital saying there's now a new payment mechanism for this algo and we've just begun deploying this solution in our first few hospitals to try to figure out how to make it work because these people don't have a mechanism to bring that -- they're running hundreds of thousands of ECGs all day long. And so their primary care doctors, they put you on the table. They run the ECG says, "You're fine, you go home", like just the change management of saying, you're fine, maybe but probably, but I might call you tomorrow and say you're not fine. Like it just takes time for the system to absorb these changes. But we run 300 million of these things. And so it's not impossible that we run tens of millions of these algos. And if anything, tens of millions times $126 is a lot of money. So at some point, like you could imagine the app side of our business gets much larger than the other side.

And I guess on that payment side, is that the AFib algorithm, is that a model you think that's able to be replicated across other indications or...

Yes, we have dozens of these things in some stage of that, some in front of the FDA, some don't require FDA approval, but we have dozens. And we have 1 -- we launched an algorithm, we call it IPS or immune profile score. In this country, if you are a likely candidate for a checkpoint inhibitor, you're 1 of the main metrics of whether you get the drug is a test called tumor mutational burden which is essentially calculated from next-generation sequencing like we do. And TMB is the conventional marker for whether or not you'd get that drug. Well, TMB is a very rudimentary marker. And so we built an algorithm that looks at not just TMB but all kinds of other stuff and predicts whether or not you're going to respond to immunotherapy much better. And so that's another example. Like we have just dozens of those that -- at some point, I would suspect we'll get reimbursed.

Maybe just spend 1 second. We've talked today about, so far, a lot of things that have -- are focused towards the enterprise, either hospital or pharma company maybe it impacts the patient, of course, but maybe doesn't directly interface with them. You guys announced more consumer-facing application, olivia. Maybe you could just spend a second on how that sort of fits into your offerings and how you think about that going forward, too.

Yes. So if we think about again, like how do you bring [indiscernible] to health care, you need proprietary data, proprietary distribution, so everything we do is designed around how to get more proprietary data to train these models. And then once we can generate insights in the models, how to basically distribute them back to the people that need them, which namely would be either physicians or patients. We've spent a few billion dollars building technologies that make sense of all this multimodal health care data. Radiology scans and mammographies and electrocardiograms and genomic files. So we took all that technology and realized that it could be we could kind of leverage it into a consumer application which we built called, olivia, and we're still in a beta of rolling it out. And olivia basically is like a personal health care locker, you sign up for olivia and we can connect to your health care system and bring in all your health care data. So not just the stuff that would be like a lab results but like literally scans, like if you had a scan, we can pull that in. We can connect with every major pack system, we can bring in your digitized H&Es we can bring in your molecular data. So not all your health care data is on your phone. And so you can also leverage our -- we built this agentic platform to basically query that data and ask questions but also you can send it to other people for a second opinion or carry it with you as you move from 1 provider to another. So that's what the application does.

Is that going to be -- does that feed into your broader data business over time as well on the back end or...

Every patient who -- right now, when you -- we have a significant -- if you look at how we get data today, a significant percentage of it comes from the consenting we do with patients. A significant percent comes from the BAAs and other agreements we have with hospital systems. But yes, we're very interested in making sure that we're not overly dependent on 1 source of data. And so having a direct relationship with millions of patients eventually is good for both ends. It allows us to make sure we continue to flow proprietary data, including there's a whole aspect of patient-reported data that's really interesting that we want to bring in but also the ability to kind of deploy insights back to patients is powerful. So both.

Yes. Right. Maybe we spent a lot of time or almost all of our time talking about sort of the U.S.-facing business, you made the point that, that's the vast -- a very large majority of health care spend certainly per capita. You guys announced your relationship and partnership in Japan with SoftBank, -- like maybe you could talk a little bit about how you think about the international markets and how you plan to play in those.

Yes, it's been hard. We spend a lot of time trying to figure out how to enter international markets. And they're tricky both because many of those markets have centralized health care systems and also because they are much slower to innovate and adapt to new technologies. We made a decision to go to Japan because we were in partnership with SoftBank, and we felt like that was a good way to enter a market. And so that kicked off about 6 months ago, and it's going well. But the -- it is, I think, just like the health care system, it's just kind of so stoic and archaic in so many ways that like once something really is proven to have value and providers start jumping on, they tend to jump on and scale. Like early on, it was very hard to convince people to give us their data. But once we had enough people giving us their data, it was incredibly easy. Now it's like -- it takes 0 effort. But in the early days, it was like 20 meetings for people to give us data. The same thing with AI, like once providers realize the benefit of AI at scale, they'll all jump on. And once the U.S. is able to demonstrate that it's really bringing personalized medicine benefits at scale to a population, I think you'll see the rest of the world be like way, we need that, too. But right now, they're kind of sitting back being like -- the U.S. is spending all this money, and we're all -- and we're not doing so bad. So they're not looking to kind of be more innovative or open up those investments. So I think it's going to just take some time for those markets to fall.

Maybe just as we wrap, I want to give a quick opportunity for anyone in the audience if they've got a question, do you want to hit? All right, maybe just last one for me -- sorry, go ahead.

[indiscernible] uniform in nature or do you have [indiscernible].

They're not uniform. My God, whatever the opposite of uniform would be. I mean when we're connecting to Epic or Cerner, those might be uniform or OncoEMR, whatever, but in general, also going to get data warehouses. And then even within the data warehouse, the connections we built for radiology scan is different than a DICOM file. I mean different than a molecular file. So you have to -- each one of these data modalities requires their own -- both their own connection, and then once the data lands, what do you do with it.

[indiscernible] going forward. How much of it is cancer, and how much of that is [indiscernible].

I mean we're in cancer right now pretty heavily. We also do quite a bit in cardiology a little bit of neuropsych and a little bit in rare disease. So we're in kind of multiple areas today, and I'll call cancer, pathology and radiology as well, buckle that all together. But I -- again, just like I would -- if I come back here 10 years from now, like our Applications business could be bigger than Genomics and Data and I would say I can also imagine that the noncancer sequencing way outpaces the cancer sequencing. I mean I would suspect there comes a time when everybody in this country least is sequenced for risk for inherited risk. So if people like Ambry is a leader in that space, we bought them, it's 400,000 tests. Like it would not surprise me if 10 years from now, it's 40 million tests. So I don't know, like someone is going to run a lot of tests.

No reason oncology is the only place, for sure, for...

Every single disease area where a physician makes a mistake, which is everywhere. Everywhere there's multiple drugs, which is almost everywhere other than like things like Alzheimer's. But anywhere, there's multiple drugs and trial and error. So mistake, multiple drugs trial and error. I think you'll see AI and precision medicine show up. So like almost everything.

All right. Eric, thanks a lot for being here. Really appreciate it.

Thank you. Thank you.