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

Good afternoon, everyone. This is Rachel Vatnsdal with the Life Science Tools and Diagnostics team here at JPMorgan. Today, we have the Tempus AI presentation. So as we do with these sessions, typically, it's 40 minutes, roughly 20 minutes of a prepared presentation, followed by 20 minutes of Q&A. With that, I will pass it off to Eric.

Thank you. Welcome, everybody. So I think I'll start, and I'm sure I'll end up going way off script, which is the only way these things are exciting. But I think we live at a moment in time when AI is about to have enormous impact on healthcare. There are all these background technologies that have been evolving over time, natural language processing, optical character recognition, certainly cloud and compute and most recently, large language models. And it feels like the moment is now. We've been saying for a long time that will likely show up in diagnostics first, which we will talk about in a second. As a result, Tempus has been focused since inception in this idea of making diagnostics intelligent or essentially personalized. The difference between an intelligent diagnostic per se, non-intelligent diagnostic is simply that the intelligent diagnostic knows who you are. It knows everything about you. It knows what drugs you've taken, how you've responded. And as a result, it's able to reconextualize itself based upon that information, which is the only way you can really imagine diagnostics being precise in any manner. Like if you think about it in the case of Tempus, if we're sequencing a patient and we find that they have a mutation, we don't want to recommend a drug that the patient has taken in a prior line of therapy and failed. If we're recommending a trial, we'd like to recommend a trial that they're actually eligible for. So be no point saying your patient should consider this trial if your patient is a smoker and the first exclusion criteria is they can't be a smoker. The platform we've built to do this is a connected platform. We have to spent time connecting to -- about half of all hospitals in the United States where we built these bidirectional pipes that pull data out of the EHR data warehouse, bring that data into Tempus and then we generate some kind of diagnostic insight, typically often in the form of genomic profiling, whatever it is, and then put the insight back in. It all happens in a matter of days. It happens seamlessly. And it's really the only way to ensure that these workflows can be managed by physicians. You have to basically be part of that daily activity. Otherwise, it's almost impossible. The connections we've established have allowed us to amass a very large data set. We've ingested about 8.5 million records. We have, I think, access to about 11 million or something, maybe 12 million. So it's a fairly large volume of records that we have. There's only about 19 million people with cancer. So you can imagine it's a significant percentage of cancer patients that we're able to bring in. We started in cancer, but have also extended the disease areas. But in cancer alone, we're touching something like 1 of every 2 cancer patients. It spans both academic medical centers and the community. And most importantly, it spans all the different modalities of data. So it's clinical data, it's imaging data, it's molecular data, and it's really rich molecular data at the bottom of the funnel that happens to be connected to clinical data. And that's really the most valuable part of our data set. That north of 250,000 is probably approaching 300,000 now. And it's essentially DNA profiles, RNA profiles, often solid tumor and liquid biopsy, often germline and somatic and often connected to outcome and response. So we know who these patients are, what their molecular profile is, what drugs they're taking and how they're responding. So that's really the first step. The first step, if you want to bring an added diagnostics is you have to build a very broad and connected platform. The second step is you have to build a model that's sustainable, that allows you to essentially generate this data, harmonize it structure, it makes sense of it in a repetitive manner. And so we spent a lot of time trying to figure out how to do this in a sustainable way, which we'll talk about in a few minutes. We're excited that we have finally crossed that chasm. We're generating rich molecular data. And that part of our business behaves like a normal diagnostic business where we generate clinical reports, we bill insurance, we get paid. That data is connected to clinical data, and so we generate these really rich multimodal data sets for discovery and development. That's our data business where we're licensing that data pretty broadly to a few hundred biotech companies in something like 19 of the 20 largest pharma companies. And then we're also -- through those connections we built, we have these -- all these applications that are running in the background, essentially algorithms that route patients to the right therapeutic, namely, for example, doing things like matching patients to clinical trials, closing care gaps, being predictive, all that good stuff. We tend to think that long term, our AI applications business is probably the biggest business, but today, it's still in its infancy. Let's talk genomics for a second. So we have a fairly broad portfolio -- at the heart of -- if someone said I want to basically make diagnostics intelligent or route patients to the right therapy or usher in precision medicine, at the heart of every one of those is molecular data. It's the thing you have to have. You have to understand someone's molecules to basically route them to the right path. So we knew early on, that we had to get our hands on lots of molecular data, transcriptomic data, genomic data, methylomic data, epigenetic data, proteomic data, we need lots of molecular data. Unfortunately, we don't want to give us that data, so we had to open up -- or fortunately, we had to open up a lab and generate that data. The data we want to generate is pretty broad. We want to generate inherited cancer risk data. We want to generate somatic profiling data for solid tumor and liquid biopsy. We want to generate minimal residual disease data and monitoring data, pretty broad data. So as a result, if you look at our portfolio, it spans that entire continuum. We do hereditary screening. We do solid tumor profiling, liquid biopsy and now minimal disease detection and monitoring. We have a broad portfolio. We are at scale in the market across all these areas. And we think this is the necessary continuum you have to have to generate the kind of insights that you would use to usher in precision medicine. Historically, the biggest part of our portfolio has been in this treatment selection or comprehensive genomic profiling area, which is kind of right in the middle. But as we expand into MRD and expand earlier through the acquisition of Ambry that I want to talk about in a second, we're kind of playing on both ends of the barbell. We're excited about our MRD portfolio in large part, because we've spent a long time working on a tumor-naive assay that we brought to market after ASCO this year, focused on colorectal cancer and the performance of that assay keeps getting better and better. And so it's a super exciting platform we built, which will take other disease areas. But we're also excited because we partnered with Personalis to bring their tumor-informed assay to market. And so we have both in market today. We are both a tumor-informed MRD supplier and a tumor-naive supplier. We tend to think the market long term will want both, and so we feel well positioned. We announced a few months ago that we were acquiring Ambry. Some of the folks are here right now, if you want to grill them afterwards, they're smack dab in the middle. And we're excited about that acquisition. As I said historically, they were our supplier in this category, we had a test called XG, which was our germ line inherited risk assay. They were our primary supplier in that category. So the vertical integration is fantastic. It also puts us into new disease areas. They're not just in cancer, but also rare disorders and reproductive health, gives us all kinds of advantages in our data business long term because they tend to touch patients much earlier than we do and obviously gives us geographic expansion. They're located in California. We've got labs in North Carolina, Chicago and Atlanta. So we now cover the country. And the combination of both of us together, which you'll see in a few minutes, produces a scale of molecular data that's really -- largely unparalleled. All that data fuels our data business. We operate a data business that, again, is fairly large at this point, represents about 1/3 of our revenues, give or take, maybe a bit more. And we're licensing data pretty broadly for discovery and development. People are using our data, because they can essentially recreate large elements of the trials are looking to run, and see what patients look like in the real world as opposed to just a very small trial. And so this data is used to help them redesign Phase IIs, figure out should they interrogate a program, should they shelf a program, it's invaluable for the work they're doing. The data business, as I mentioned, operates at scale. We look at a few metrics. I mean, obviously, the First metric is growth. In Q3, our data and services -- Q4, our data and services business grew about 45%. So it's a high-growth business for us. We measure two elements of that business. One is this concept of total contract value or what's the total amount of deals we've signed that we would likely bill in the future. That's $940 million at the end of Q4. It went up from Q3 to Q4. It's a robust number. All it really means is that we've got kind of lots of unsigned -- I mean, lots of agreements that we've signed, that we've yet to deliver. So it gives us good visibility in the next couple of years, certainly of our data business. The second concept we measure is called data licensing retention or net revenue retention. That was 140% at the end of 2024, was 125% the year before, so it went up. 140% is probably not sustainable. It means it's like -- it would be like going to Walmart and saying, my same-store sales went up 40% year-over-year. It's a huge validation point that our existing clients when they're licensing our data in 2023 on a same like-for-like basis, on average, licensed 40% more data than the year before. So it means that they're finding value in the data, and it's a great place to be. Our genomics business and data business, which together basically are keeping all these connections intact across thousands of hospitals allow us to basically deploy applications at some scale. We actually run these applications or algorithms across hundreds of providers touching tens of thousands of patients in multiple disease areas in oncology, deploying algorithms to close care gaps and route patients to trials and neuropsych, looking at pharmacogenomic profiling and cardiology, we've got a whole portfolio around electrocardiograms and being predictive to look at undiagnosed AFib or things like that. While these things run at real scale, they generate very little revenue because at the present moment, at least in the U.S., we haven't yet really figured out how to pay for AI in health care. We believe at some point, we will figure that out, and these things will generate meaningful revenue. We see some early signs of that, which are exciting in our ECG portfolio. We have an FDA-approved algorithm to predict undiagnosed AFib, and that now has a code for hospitals to be paid about $128 or something like that. And so it's exciting to see people starting to pay for AI, which we believe is really the only way to rationalize the irrationality of the current spend. The whole point of these algorithms, when you have a connected ecosystem and you can see what's happening to patients every day. You know what drugs they're taking, you know how they're responding, you can essentially deploy these algorithms to kind of try to make sure people don't fall through the cracks. So did somebody not get tested when they were supposed to get tested? Is somebody not on a clinical trial they should be on? Is somebody on the wrong dose? Is somebody failing to note what could be a critical adverse event? Like you just want to make sure people are -- that you're using data in a way that, that data can be presented to a physician to make high-quality decisions at the point of care. Here's a view of one of those products called Tempus [Next]. [Presentation]

So it's kind of -- again, these things -- when I say these things operate at scale, they're really operating at enormous scale, but they don't generate lots of revenue. But we are watching patients in real time, figuring out that somebody is falling through a care gap, correcting that. And I think at some point, the system will find a way to pay for this efficiency and these could become very large businesses. We are fortunate that we've kind of built the necessary platform, the infrastructure, the pipes, again, to pull data out of the EHR, structure and harmonize that data, make sense of it, generate some kind of diagnostic insight, typically tied to molecular data in some way, shape or form, put the insight back into the hands of that doctor. And that's ultimately, we think, the platform you need to make all this real. In terms of just some quick updates and then we can take questions or go through your questions. We did quite a bit in 2024. We have a lot going on. With Ambry, we're going to approach getting close to 4,000 people. So we're not -- this is not a small effort, and that includes about 1,000 folks that are technical 300 PhDs, 500 or 600 software engineers, 100 AI engineers. So it's a very large effort focused on delivering across all these businesses, genomics data and applications. We expanded into MRD. We obviously announced the acquisition to acquire Ambry, made all kinds of progress in terms of ASP across payers, expanded a bunch of data collaborations, did a fantastic job on the R&D side and brought a bunch of new algorithm test to market. One of them, which I'll talk about is our IPS, our immune profile score. This gives you an idea of what is coming that I think is really profound. IPS, unlike a typical test doesn't look at one molecular signature to make a prediction. So it doesn't look at one thing and say, "Oh, this is predictive, like MSI or PD-L1." It looks at a bundle of molecular signatures that as a bundle give rise to certain responses. And one of the inherent problems we've had with precision medicine is that, it has been largely point-based, singular point based. So we sequence a patient, they're EGFR mutated. We give them EGFR inhibitor, then we wonder why half the population doesn't respond. And even those that do respond, have a wide variety of responses. Some people respond for a year, 2 years, 3 years, 4 years. And the answer with these new technologies, especially generative AI, the ability to build very large models, you can essentially look at all kinds of characteristics you didn't even know were associated and build much better predictors. Our IPS score, which we think will be a very big deal is far more precise than, for example, tumor mutational burden. So over time, we think -- first of all, it opens up the market and it refines the market. And we tend to believe these kind of bundled molecular signatures, which we call fuses are going to be the future of the space. In terms of 2024, we delivered $693 million of revenue, about 99% of our forecast, which we're proud of, about 30% growth in the overall business, 35% growth, I think, in Q4, 45% growth of our data business, and we delivered 100% of our EBITDA target, which for those people that do the math, means we were getting very close to being EBITDA breakeven in Q4, just EBITDA breakeven in Q4, which is another milestone for us. With the acquisition of Ambry, we have that scheduled to close on February 1. We had originally thought that would close January 1, but the state of California has decided that it wants to also decide whether or not people can buy companies. So it has its own [indiscernible] thing. And so now you have to clear the federal government and the state of California. So we cleared the federal government, but we still have to clear California, which we think will happen in the next week or 2, whatever. And so we just scheduled the closing for 2/1. It does mean, however, we lost 1 month of the consolidated business. So we reforecasted guidance at $1.23 billion of revenue, which would actually be north of the high end of our historic range. And then we've told the world we expect to be adjusted EBITDA and free cash flow positive in 2025. So -- which is a super exciting milestone for us. We'll turn 10 this year and to build a business that's growing at our core business in 2025 is expected to grow about 30% Ambry, a bit less, but it's really nice to see a business that's growing 25%, 30%, making money, just a great place to be, especially given all of the exciting investments we have that could drive growth in 2026 and beyond. On that note, I'm 16 seconds late.

No, that was perfect. We round down to 20 minutes. Perfect. Thank you. So Eric, congratulations, first time for Tempus joining us at the JPMorgan Healthcare Conference post IPO, so as a public company. I just wanted to kind of ask you, given a lot of us in the room are more focused on this diagnostic side of the portfolio, can you walk us through in your own words, why should we see the value in the data side of the portfolio? And kind of reflecting on the last 6 to 9 months as a public company, what would you say you're most proud of as well?

Yes. I mean it's tricky. One of the challenges we have is that we have -- today, we have two large businesses, and often investors are very stuped in one or the other. So like one of our largest investors, [indiscernible], is a tech investor. So they like understand the data business intimately. But we also come across other folks that only want to talk about our ASP and what's happening with our blah blah, and they're very focused on that. And it has been a bit tricky for us, because you have to walk in the room and kind of reorient yourself. Are these folks that really are deep in diagnostics? Are they deep in data? Very rare somebody is deep in both. You have to be deep in both. Like it's a bit like a car analyst. And then 10 years ago, you could say, do you understand EVs and they're like, no. But today, you have to understand EVs. And 10 years from now, you'll have to understand autonomous driving and EVs or you can't be a car analyst. And it's the same thing here. Diagnostics today are stupid. They will 100% be smart. That's 100% AI. So 100% of diagnostics are going to be wrapped around AI. So you just can't -- every -- like you're going to have to kind of figure out how these two things interweave, not just for Tempus, but for LabCorp and Quest and everybody else because those -- that's just going to be a big driver. And that's what we're super excited about, I think, is that we saw that world coming. We got very lucky -- we didn't realize LLMs would be here this quick, but we certainly saw this world coming. We were preparing for it for a long time. And we just think we're in a really good spot in light of the platform we built and the amount of data we amassed.

Perfect. And then maybe digging into some of the announcements that you alluded to from earlier today. You unveiled the announcement of the development of a whole genome sequencing assay, which you're referring to as XH. So can you elaborate more on the strategic rationale for adding this test to the menu? And how do you view it competitively as well? And what sort of timeline should we see associated with that launch?

Yes. I mean -- so XH is a whole genome assay. It's the first of our whole genome products, come out later this year. It will roll out over time. Long term, and again, I don't know what long term is, but 2, 3, 4, 5 years, pick some number, the whole platform will migrate to whole genome. There -- we embrace that world. We're excited to have a whole genome be the backbone of all of our assays. There won't -- as costs keep coming down, there just won't be a reason to run targeted panels. Now you have to phase it in appropriately, because reimbursement and studies and all kinds of stuff, you don't want to be disruptive to your business. But again, over a period of time, you'll see companies like ours, I think, just replace their historic targeted chassis with whole genome, especially with [Novax Plus] and advancements coming. And we have always said that we're not interested. We're fortunate that we operate at real scale. We make money, so we don't have to be trying to maximize every little dollar of margin in our sequencing business. So as we -- as our ASPs improve, as our costs come down, we don't just harvest the gross margin. We actually invest in bigger panels. Our XH is a whole genome assay is the first of those, but we're now in active move to migrate the rest.

Perfect. Maybe sticking with some of the announcements that you had today. Earlier today, you also announced preliminary 4Q revenues. And you noted that genomics growth was around 30% and the data and services business grew around 45% in the quarter. So I was wondering, could you unpack that for a little bit? What color and trends did you see across those business lines? You called out some of the weakness in the CRO side of things. But also if we look at numbers, genomics did fall a little bit short of Street estimates. So walk us through really what trends did you see in the quarter?

Yes. Well, the genomics business, I think the challenge is the two parts of our business that we called out last quarter that we're experiencing some pressure are once -- there are two companies we acquired years ago. We acquired a geno-typing business that literally does SNP genotyping, and we acquired a very small CRO. And the landscape in this part of the world has been under pressure. Capital has not been flowing into biotechs. The markets have sucked, so pressure. So the parts of our business that feel that pressure are those parts. They're also not parts of the business we have any interest in like trying to fix or like focus on. They're very small parts of our business. The main parts of our business, data and genomics, we're fortunate are actually right on plan. But the genotyping that genotyping business, very low-margin genotyping business rolls into genomics. So it's a very small part of a larger number. But if that business is off by $3 million or $4 million, when you're our size, it shows up. Same with CRO. So we had about $5 million, $6 million, $7 million of softness in those two businesses, which we paid no attention to. We told people when we went public, we would try to treat public -- our public investors like they were in our Board. So if you were in our Board, I would be like -- that was a perfect year. And these two things that were slightly off, I'm not going to spend any time on fixing. And you can see that in the EBITDA number, right? Any time you're slightly off on revenue, but it shows up in EBITDA, it basically means that the core of the business is doing exactly what you want. And we don't just chase empty calories. We don't just chase numbers. We just try to stay focused on the bigger picture.

Perfect. Yes. Maybe digging into that, you did achieve your adjusted EBITDA target of growing sequentially in 4Q. So any other incremental drivers you want to call out on that EBITDA line for 4Q?

I think when we -- there's a lot of ifs when we went public, like would the data business continue to grow? Would it grow at 30%, 40%, 50%? Could you generate leverage? I mean, we posted kind of 35% growth in Q4 at a pretty big number in an almost EBITDA-neutral manner, whereas our -- if you look at a lot of our competitors that are of similar size, they're still losing hundreds of millions of dollars. So we're kind of super fortunate that we're able to execute, grow and not burn a ton of cash. And I think that is a byproduct of the fact that we've been investing for a long time in building very high-margin technology components to our business that are starting to pay off.

Perfect. Maybe shifting to 2025. In the presentation, you noted that you expect Tempus and Ambry combined to grow roughly 25% in 2025. So I was wondering, could you unpack that for us a little bit? How should we think about Ambry given a little bit later close and contribution from that assay versus contribution from clinical genomics and the data business as well?

Yes. I mean -- so we're a bit cryptic -- first, we're a bit cryptic for a few reasons. One, we haven't closed yet. So like -- and so it just feels like inappropriate for us to take a private company and go like nuts with disclosure for a transaction that hasn't closed. That said, we've tried to at least give people some color that when things grow really quick and Ambry had a fantastic year in 2023. Just like our data business, if our data business grows at 60%, it's not going to grow 60% again. Like if we tell the world we're going to grow 30%, things that grow super fast when you lap them, you see that deceleration. So we never look at like one quarter growth rates. We tend to look at like 1 year, 2 year, and they have a very strong solid business, but ultimately, one that will likely grow closer to 20% over the long term as opposed to 30%. Now again, they may benefit from certain new business lines like if rare starts to really take off, that may grow faster. But we don't -- we try not to ever count on what we can't see today. So just as Tempus, our genomics business could start to really accelerate if we unlock the pent-up demand that there is for our MRD assays, which is pretty extraordinary. We don't forecast that in. We assume on our $1.23 billion that it's going to be a very inconsequential contribution of MRD and the same thing on the Ambry side. We're not counting a bunch of new volume. But they have a really healthy business at around 20%, and we have a really healthy business that's around 30%, and you can kind of do the math. I get asked occasionally like about hereditary profiling to begin with. And people say to me like, why did you buy a company that is in a space that seems kind of saturated? And I'm like, I don't know what that means. I mean, I'm pretty sure there will come a day when every single one of us will have germline profiling, like all of us, like the whole earth. So how many tests are running today? And let's add up all the Myriads and Ambryes and GeneDxs and Invitaes and pick a number, what number of tests does it get to and then take the earth. That's probably your lack of saturation. So I wouldn't be surprised if there comes a time in the future when that side of the business grows way faster than the other side of the business.