Founder Stories: Eric Lefkofsky, Founder and CEO of Tempus Labs

So we're lucky to have Eric Lefkofsky here with us today. He's a serial entrepreneur and an investor, a tech investor. He's basically done a half dozen or so companies. I tried to count. I think it's around 6 or maybe 7. That's right. A couple of— a trio of public ones, Innerworkings, Echo Global Logistics. In 2007, and this will start to sound familiar if you already don't know who he is, Eric put a million bucks into a kind of a, sort of a very idealistic activist website called thepoint.com, which eventually became Groupon, which then turned into like the fastest growing startup ever, I think, by some counts. He's now running a company called Tempus, which is a, it's like a health data company. It's super ambitious, they're growing really fast. So let's just get into, to Tempus quickly. I mean, you know, you had done a bunch of stuff, you know, business services companies, media companies, Groupon. What caused you to sort of want to get into the healthcare space, want to do what you're doing now?

I mean, so I've spent my entire career basically structuring unstructured, messy data. If you look at the common thing of all those companies, it's that. And whether it's manufacturing or logistics or printing, we're trying to bring technology to industries that have not had a lot of technology. Technology. I never thought I would get into healthcare, but about 4 years ago when I was dealing with somebody who was fighting cancer, I was just perplexed at how little data had permeated that process. And I became convinced that we had to bring the same kind of technology to healthcare that we were bringing to other industries. And so about 3 years ago we started Tempus, trying to figure out if we could just fix the underlying data infrastructure in cancer care and then eventually all healthcare. And the challenge in fixing the underlying data infrastructure was really about getting the kind of data you would need, which is clean, structured clinical data connected to molecular data in one place. So it's all about figuring out, you know, who are these patients, what drugs are they taking, how are they responding to those drugs, and then by sequencing those patients, which we do at Tempus, you can begin to understand why, why they're having that response. So, it was really the personal experience of seeing this firsthand and dealing with it that made me say, you know, I want to spend my life focused on it.

[Speaker:Joshua Elkington] Got it. So, and just so everyone, for people who aren't experts, including myself, I mean, when you say sequencing, this is different, I guess, from 23andMe or ancestry.com. I mean, you know, we've seen on the kind of consumer side these kind of consumer genetic tests become very popular, or kind of popular. I mean, what are the sequencing that you're talking about, it's much more sophisticated?

—Yeah, so we operate, I think by sample count we might be the second largest, probably the second largest clinical sequencer of cancer patients in the U.S. And, you know, growing quite rapidly. The kind of sequencing we do is governed by what's called CLIA CAP, which is a laboratory act that basically governs companies like ours who do clinical sequencing that patients would get when they're being treated. Which is very different than the kind of sequencing that other folks might do, what's often referred to as Mendelian sequencing or, you know, the kind of germline sequencing that you would get from people who kind of run those tests, where it's often about looking at characteristics more on the prevention side. We're on the managing disease side. And as a result, because these are patients who are sick, who are being treated, and who are likely going to get, in our case, targeted oncotherapies, which could be chemotherapies, radiotherapies, or targeted drugs, which are by their very nature toxic and have adverse events associated with them, the sequencing we do has to be in great depth. So I often tell people who don't understand sequencing, think of it like a magnifying glass. You know, we're sequencing what's— our average depth of coverage could be, you know, 500 to 1,000x depending on the panels we run. And often the kind of sequencing that some of these other folks do is 25x or 30x or 50x or 75x, and so it's a much lower depth of coverage.

Right, and so this, I mean, very quickly I imagine becomes a data problem. And, you know, we were talking, I think, just before, and this, the kind of sequencing we're talking about, you know, used to cost in the millions of dollars. Now it costs, what, in the thousands of dollars? And that's by virtue of, you know, what, Moore's Law and cloud computing and, you know, the—

It's actually, I mean, what's interesting in the case of cancer care, and I think healthcare more broadly, is you actually have multiple massive technology paradigm shifts hitting at one time. So not only can we store data at costs that are unimaginable, and we have all these tools like machine learning tools such as natural language processing and optical character recognition, which are pretty amazing. But you also have a genomic revolution that is really unparalleled. So it's something like a million-fold reduction in the cost of generating this data over the past 12 years or so. And if you look at, and there's all these kind of famous graphs looking at the cost of generating molecular data, and it's actually been decreasing at about 2 times the pace of Moore's Law.

So obviously servers are getting cheaper, but what's the other part of that? What else is driving the cost?

Well, it's a series of, I mean, there's a series of companies, Illumina being the most well-known. Who have developed technologies over the past decade that, no different than Intel did this with processors, that have allowed this data to be generated at super low cost compared to in the past. And so we now, for the first time ever, have the ability at a sustainably low cost to collect and structure clinical data and combine that with molecular data and then use machine learning and basic artificial intelligence to try to figure out what patterns exist so actually we can help clinicians, in our case oncologists most often, make real-time data-driven decisions by figuring out of all the different drugs that they might think about for a patient, which is likely to have the best response.

So today, I think, or maybe late yesterday, the Wall Street Journal reported that Amazon, which is sort of scary to all sorts of people in all sorts of industries, is coming for you, to some extent anyway. They announced that they're going to sell software that's going to mine patient medical records. Of course, a lot of Apple, Google, they're all kind of playing in this space a little bit. I'm kind of curious what you make of the Amazon announcement, but maybe more broadly, big tech in general getting into this world.

Yeah, I think with Amazon, it's really far too early to tell. I think the larger theme I think is quite positive, which is there's only a few large sectors of the US economy that haven't been completely upended by technology. Maybe the government being one, healthcare being another. And so I think the fact that companies like Google and Microsoft and Amazon and Apple have all announced initiatives in this space is really positive. And I think it's likely to be the start of a very prolonged trend. Because again, if you think about these underlying technology currents, the ability to collect and structure data at historically low cost, to analyze that data with all kinds of imaging technologies to generate molecular data, what this means is that we should be able to use technology to keep people alive and have them live better lives. And given the size of the healthcare market, 3+ trillion, this is a huge space that I think all of these big technology companies are gonna wanna find a way Sure.

So, we've heard versions of this in the past. Watson, IBM has been telling, I would say, a related story. I don't know if you think that's a fair comparison about using artificial intelligence and their natural language processing to help doctors make sense of unstructured data. It feels like a lot of these things haven't gone that far or maybe haven't gone as far as than kind of the marketing promises have suggested. And I'm kind of curious what you make of that. I mean—

I think it's completely fair. I mean, I think when we started Tempus, we wanted to take a holistic approach to this problem because we felt like every single approach, every siloed approach, even though they're quite good, was only solving part of the problem. And so at the end of the day, we talk a lot in our world about combining clinical molecular data. We're really combining 3 kinds of data. Phenotypic data, morphologic data, and molecular data. So text, images, and molecules. And it's only when you combine all three of these that you really can get a picture of as to what's happening to this patient. There's great imaging companies that look at anatomic pathology slides and radiology scans and do cool work, but they're often missing the context of what happened to the patient that's locked inside physician notes or pathology reports. And likewise, there's great companies who sequence or great companies that structure data, If you wanna solve the problem, we think you have to do it all. You have to have it all under one roof so you actually can see what's happening to these patients and begin to look at the patterns that might describe why. And so I think we hear all the time that some of these historic efforts didn't work. And I don't think it means a whole lot other than often early-stage siloed efforts in the midst of technology paradigm shifts don't produce what you ultimately will see through time. A great example would be early search engines like Lycos and AltaVista didn't have the same impact on our life that Google had. And so I think the question for us is, do the underlying technologies exist that would allow a company like Tempus to gain scale? And the answer is yes. And I guess the evidence of that would be, I think we now have something like 250 hospital systems working with Tempus, including about 75% of all NCI cancer centers., and roughly 25% of all cancer patients in the US come through Tempus. And so this kind of scale was unimaginable, I think, a few years ago until these historic trends were in place.

I want to talk about the kind of implications, but before I get there, you know, you started this company, you know, after whatever, a bunch of interesting experiences. You know, obviously Groupon, you definitely came into contact with Amazon, which bought a competitor. I'm kind of curious, are there lessons from the Groupon experience that you actually, that have been useful in this business, or is it just so wildly different that it's like one chapter of your life closes, another one opens? What do you take away from that crazy experience?

I mean, well, I think that I've been fortunate that I've been part of starting 6 companies And they all have had really interesting pasts, and so you learn something from each one. 3 went public. We sold one called Media Ocean. And so those were all— and 2 we operate today. One is a company called Uptake that my partner Brad Kewall operates, and I run Tempest. But you take something away from each of these, and you kind of build on it. And I always say to people, it's pretty simple. In each experience, I try to repeat whatever worked well. And avoid whatever didn't. And there's not much more to it than that. And so at Groupon, we did a bunch of things amazingly well, which led the company to grow from, I think, $30 million in sales our first year to $4 billion in sales our third year, which was kind of crazy in terms of growth. But there's a lot of things we didn't do well that I wouldn't repeat.

And I think— What's the most important thing in your mind that didn't go well that you wouldn't repeat?

Well, I think for a lot of companies, we were the end of an era where companies went public very soon. First of all, the rules changed. The SEC actually changed its basic guidelines of how many shareholders you could have to allow for companies to stay private longer. And I think what you've seen in this recent crop of big technology companies like Airbnb and Uber and the like is a willingness to stay private much longer and ensure that their business model is really fully developed before they enter the public domain. And I actually think it's a much healthier pattern for people. Some of these businesses that are really high growth there's just natural volatility that comes with that. So, and especially when you think about a company like Tempus, we're not just in cancer today, we also have expanded into diabetes and depression, and next year we'll move into other diseases like cardiovascular disease. So you're talking about these huge industries, huge technology paradigm shifts, and so you really want to make sure, in terms of your model, that things are in place before you think about the public.

So, okay, so you said Tempus is doing 25% of cancer patients sequencing.

No, no, 25% of patients come through Tempus. Okay, so, and what—

Sequencing some smaller percentage. So what percentage of cancer patients are getting sequenced today, roughly?

So we touch cancer patient data, some clinical, some molecular, for about a quarter of the country. Got it. Which also happens to be roughly the percent of cancer patients that get sequenced throughout the entire country.

So, and is this the kind of thing, I mean, how long before everybody who gets a cancer diagnosis is going to get genetic, you know, get sequenced?

You know, I mean, I would think, and again it's speculation, but I would— today about 25% of cancer patients will get sequenced. There's 1.7 million people diagnosed with cancer a year, about 25% will get sequenced. I would think in about 5 years that's 100% of patients. Okay. And I would think anybody who has metastatic disease or late-stage disease or high-risk disease will be sequenced multiple times, both their solid tumor and through what's called cell-free DNA analysis or do some kind of liquid biopsy. So, and I think that same thing, that same pattern will start to emerge in, for sure, neurological disorders, cardiovascular disease, infectious disease, immunology, and the like, because we're starting to see pharmacogenomic and other patterns where sequencing patients can produce really interesting data to tell doctors that one class of drugs work better than another. So, you know, I think, I would assume in 5 or 10 years We're all being sequenced a lot.

[Speaker] Got it. So that's a huge opportunity. I mean, that's like, potentially thousands of dollars. I mean, I guess prices will probably come down, so maybe it's not quite as good as it sounds. But what does that do to the overall cost of healthcare? I mean, is it gonna make healthcare cheaper, or is it just gonna be like another kind of expensive thing that is good for patient outcomes, but is also causing, patients, insurance companies, the whole system to be further strained?

Yeah, I think, so purely the sequencing side of this, I think some, and I'm not a, I can't forecast industries, but I think there's a few people that do, and typically worldwide it looks like about $100 billion in 2025, so it's a seriously large space. But I would think the global healthcare market obviously is 30 times that in the United States alone. And so I would assume what will happen is all of these data efforts that cost money, whether you're generating clinical data or molecular data, will be sizable. But I would think there's probably something like a 2-for-1 or 3-for-1 or 5-for-1 or 10-for-1 reduction in the total cost of healthcare relative to the collection of this data. In large part because the, the whole point of collecting the data is to navigate people to the right therapeutic.. And so much of our healthcare, so many of our healthcare dollars go toward the last 90 days of care and go toward people being hospitalized when the care goes wrong. And in cancer patients in particular, where you have a relatively short duration of life if you have late-stage or high-risk disease, you can't afford to have things go wrong. And so, you know, I'm bullish on a kind of massive reduction in the cost of total healthcare. And the material prolonging of life by virtue of collecting this data. But we don't, I mean, that's what we're gonna see, right?

[Speaker] Right. This space is really interesting just 'cause there's so many ethical issues. And we did a story, Businessweek did a story about these murder cases, these kind of amateur and professional detectives using public genetic data. Are there ethical issues here if we're all getting sequenced? Multiple times in our lives or once a year or like, do we have to be thinking, all of us have to be thinking about where does this data get stored and who's gonna have custody over it and like those kind of difficult ethical issues?

[Speaker] Yeah, I think this is actually one of the few areas where from its onset, or at least over the last 10 years from its onset, the laws and the federal government have actually done a good job at thinking about this data. There are all kinds of laws, HIPAA and the like, that govern health information data and how it can be used and what makes data identifiable and not identifiable and what rules exist. Patients are appropriately consented. They are made aware of what happens with this data in ways that don't exist on the internet today. So a lot of the things you hear about with data people using that data to target ads or to do other things just don't exist in our world.

So you're not gonna sell my genetic information to Facebook?

Yeah, I mean, I'm not sure what they'd even do with it, but at the end of the day, the reality is we're completely managed and governed by a series of regulations that I think have largely helped patients. Now, those laws have a fine balance because On the one hand, you want to protect patients' information. On the other hand, you want to make sure that that data is in the hands of researchers broadly enough that cures can be found. What we find, which is super interesting, is in cancer, patients who are asked to consent to sharing their data, it's something like 99% of patients agree to have their data broadly shared, identifiable or not identifiable. Now, we don't share identifiable, only de-identified data, but If you're sick and you have a disease, you're less focused on what's going to happen with my data, and you're far more focused on can somebody help find a cure that's actually going to make me better. Yeah.