Healthcare M&A due diligence: What Carbon Health's chart data revealed

In this webinar, Charta Health head of marketing Rachel Lee talks with Dr. Caesar Djavaherian, co-founder of Carbon Health and now chief medical officer at Charta, about what he learned as Carbon scaled from 4 to 140 clinics in under three years by building and acquiring sites at the same time. Dr. Djavaherian explains why revenue quality tends to degrade after an acquisition, why five-star reviews can mask poor-quality care, what chart data to analyze before you sign a deal, and how to bring a newly acquired team's charting and coding up to standard.

Key takeaways

• Revenue quality often degrades after an acquisition, even at well-run clinics: when oversight moves from founder-owners to a larger organization, clinicians tend to capture less of the work they do in their codes.
• Patient reviews are not a proxy for quality of care. One Florida multi-clinic group with hundreds, possibly thousands, of five-star reviews treated every cough or sore throat the same way, including antibiotics and steroids that weren't always indicated.
• Before an acquisition, Dr. Djavaherian recommends analyzing the target's last 1,000 or more charts by provider, symptom, diagnosis, and CPT code, then comparing them to national or regional averages to find abnormal spikes.
• Chart analysis cuts both ways in underwriting: it can reveal practices to zero out of the valuation, or documented-but-uncoded work that means revenue should be higher.
• At one point, Carbon's E/M weighted average was 3.1 against a roughly 3.7 benchmark for level 3 and level 4 urgent care visits. Across about a million visits a year, he estimates the gap at roughly $30 million in revenue.
• Random sampling and one-time coding tests produced short-term improvement but longer-term degradation. Pre-billing coding feedback built into the EHR is what made the lasting difference.

Watch the webinar recording: Understanding M&A from a CMO's lens

Introducing the session

Rachel Lee: Hi, everyone, and welcome. Thanks for tuning in to this session, Understanding mergers and acquisitions from a CMO's lens: Lessons from Carbon Health. We'll dig into how Carbon Health scaled from 4 to 140 clinics and what charts revealed along the way. Here's a hint: great reviews can sometimes hide revenue problems that only show up in the chart data. We'll also cover why revenue quality slips after an acquisition, what to look for before you sign, and how to bring a new team up to standard. I'm your host, Rachel Lee, head of marketing here at Charta Health. If you haven't heard of Charta Health, we are an AI chart review platform for top provider organizations like AFC and Family Care Center. Charta uses AI to review 100% of your charts before billing, and that means catching miscoding opportunities, flagging compliance risks, and making sure you're reimbursed accurately for every visit.

Now I'm so excited to introduce our speaker, Dr. Caesar Djavaherian. He's the co-founder of Carbon Health and now the chief medical officer at Charta Health. He helped scale Carbon from 4 to 140 clinics in just under three years, building new sites and acquiring existing ones at the same time. He's seen firsthand what the charts reveal that the reviews, quite frankly, sometimes don't. Caesar, thanks for joining us today.

Caesar Djavaherian: Thanks so much for having me.

From the emergency department to co-founding Carbon Health

Rachel: So let's start at the very beginning. You trained as an ER physician, and then you went on to co-found Carbon Health. Tell us about that incredible journey. What made you want to build something of your own?

Caesar : Well, a few things. I trained in New York City. I have a background in medical informatics, and so in the very early days of the intersection of health care and technology, I wanted to be a part of that movement. At New York Presbyterian, I was fortunate enough to train under Peter Weier, who's one of the gurus in evidence-based medicine. So if you put health care technology together with evidence or data behind it, you can see the direction of creating Carbon.

In the emergency department, we had people who were waiting hours and hours and hours before getting care. Oftentimes, we'd say, "Well, you don't need to be in the emergency department." You could have been seen by your primary care doctor. So access was an issue, and Carbon was a way to address access by creating urgent care centers around the country that were easy to get to. We were also data rich. We wanted to make sure that we controlled our electronic health record and captured data on every one of our patients so that we could improve the health care we were giving them. And so we had this vertically integrated company called Carbon that was doing incredibly great things for society and for patients who were coming in, and keeping them away from improper places of care like the emergency department.

How did Carbon Health scale from 4 to 140 clinics?

Rachel: Very cool. Such an amazing story. I've heard it before, but I'm still in awe every time. Let's talk a little bit about how Carbon actually scaled. As you mentioned, Carbon went from 4 to 140 clinics in under three years, and that's a pace most operators don't get to see. Walk us through what that looked like. You were building new sites and buying existing ones at the same time. How do you even balance the two?

Caesar: It was hard. As you know, that growth happened during the pandemic. We were uniquely positioned because of our technology platform to take care of patients and distribute them appropriately among our clinics. We understood that if we purchased clinics, we could meet the demand faster. But building clinics was somewhat easier, because we had our standard operating procedures and our approach to standardization in clinics and in how we trained our providers. So that was a hard balance to make. We were actually, I think, ranked number two in Inc.'s fastest-growing companies in the country, across all industries. It's rare to have a health care company do that. But again, the technology component was an important part of the strategy.

Why clinical quality is hard to monitor across hundreds of providers

Rachel: Totally. I imagine doing both at once is a lot to manage, and it got even harder on the clinical side. As the chief clinical innovations officer at Carbon, you were responsible for clinical quality across those sites. What kept you up at night?

Caesar: I was responsible for bringing innovation from outside the company into the company in any way I could, particularly as it pertained to clinical care. My colleague, Sujal Mandevia, was actually chief medical officer at that time, so we worked together, along with our national medical director, who's an incredible clinician and manager. But with a lot of the quality issues, as you can imagine, across 13 states and 650 providers, we worried about whether someone could get a very different care experience in one state versus another, or with one provider versus another. How do we keep track of the actual care provided in each clinic? That definitely kept us up at night. We tried to create technologies that allowed us to do that, but it was not easy.

For example, we hired many support specialists who would read charts. They couldn't read all the charts, but they would try. They'd read some charts and provide feedback to clinicians. Our clinical managers would do essentially the same thing: pick maybe five or ten charts a month, read them, provide feedback to the clinicians, and report up to Dr. Roger Wu, our national medical director, to identify any flaws. As you can imagine, that was a bit of a Swiss cheese thing, where occasionally things would fall through the cracks. So you had clinicians who maybe did the exact same thing on every patient with a similar presentation. And as you can imagine, not every patient who comes in with a cough has the exact same diagnosis and procedure and all that done. So that was hard to manage.

The biggest challenge in healthcare M&A due diligence: Knowing the quality of care

Rachel: And that's the part most people outside the clinic never get to see, so thank you for sharing those insights. Let's zoom out a little. When you look back at that period, what were the biggest challenges in M&A?

Caesar: On the acquisition side, we had to make a judgment call. We had to go into clinics, often clinics with multiple sites around the country, and very quickly try to figure out: are those clinicians trained well? For us, we never compromised on the care provided. It had to be evidence-based. It had to be the standard of care. So when we were going to acquire a company, the question was, how do you really know the quality of care that's being delivered?

If you're going in to take over a site that's on—name it—EHR, you generally don't have the data behind you or the bandwidth to read every chart encounter to see whether the clinic you're acquiring is delivering high-quality care, and, on the other side of it, whether they're coding and billing appropriately. Either undercoding or overcoding, it's a problem either way. When you're making an offer, you have to know those details, and it was absolutely a huge challenge to do that in our M&A process.

Why does revenue quality degrade after an acquisition?

Rachel: That brings me to a hot take we uncovered during our prep session. You've said that as you acquire more of these clinics, revenue quality tends to degrade over time, especially right after an acquisition. What does that really mean for this audience, and why does it happen even when you're buying well-run clinics?

Caesar: We were really surprised by that. None of us could have predicted it. What would happen is that we'd find the best-run clinics, as far as all of our diligence uncovered, and then we would underwrite them to a certain revenue and profitability standard. Across the board, we would typically keep the same providers, the same staff, etcetera. We'd put in our electronic health record and our systems, and the EHR would increase productivity relative to the baseline, but the revenue quality would degrade.

That was surprising, because we found that when clinicians were now supervised by a larger organization, they tended to skimp on their coding quality. Maybe they wouldn't go the extra mile for the patient like they would if their boss was standing right next to them. What was more common, actually, is that they would do the work but not capture it in the codes as appropriately as before, because the oversight was a little more removed. The clinics we were taking over would have one or two owners or founders running, let's say, four to eight clinics. Those founders were able to maintain supervision of the clinical care and the coding accuracy better than we could at scale without implementing technology.

Entrepreneur-owned vs. corporate-owned clinics: How the ethos changes

Rachel: So it's not really about good or bad clinics. It's really about what happens after the deal. You brought up something very interesting about the people side of this. What do you think is the real difference in ethos between an entrepreneur-owned clinic and one that's part of a bigger company? You mentioned that it shows up in the numbers, but what is the difference in that ethos?

Caesar: Well, with entrepreneur-owned clinics, if they make mistakes and don't capture appropriate service codes, the money comes out of their own pockets, and they may not be able to make payroll. So they'll go the last mile for patients. They'll do everything they possibly can that's appropriate for patient care, and then they'll capture all the CPT codes that generate the revenue for the clinics. What we found is that they were doing incredible work, but sometimes not capturing proper CPT codes, when we acquired those clinics. And that ethos was the change.

The entrepreneur-owned clinics were doing everything, capturing whatever they could, staying up at night, finishing their charts. When it became a little more corporate, with larger, structured ownership, that ethos would go away. Would you stay an extra fifteen minutes late to fully chart your encounter and fully put in all the CPT codes? What we found is that, humans being humans, the shortcuts came in, and so the quality of both code capture and, sometimes, the services degraded over time.

Rachel: Got it. That totally makes sense. When it's your name on the door, you do care differently.

Caesar: Yeah. I think that's true.

Why patient satisfaction and online reviews aren't a proxy for quality of care

Rachel: Here's the thing that surprised me most in our conversation, and one that might surprise folks in the audience as well. Some of the clinics you were considering acquiring had really great reviews. You mentioned that patients rated them highly because they felt like they were getting really great care. So apart from the difference in ethos between an entrepreneur-owned clinic and one that's part of your company, there was something else under that. That begs the question: why can't patient satisfaction or star ratings be a proxy for quality of care?

Caesar: This is really super interesting, and something I wouldn't have ever predicted. If you had asked me, as I was transitioning from emergency medicine to urgent care—I still practice emergency medicine—and learning how to run clinics, what would be the number one most important indicator of quality, I would have said Yelp or Google reviews. That if all of our patients were happy, then we're doing everything right.

What we found is that there are actually clinics around the country—in fact, we looked at one multi-clinic site in Florida—where they had hundreds, maybe thousands, of five-star reviews across all the clinics, and we were like, this is amazing. We're absolutely going to go and buy this clinic and maybe learn from them about what they're doing. So we flew out to—I won't say exactly what city in Florida—but we flew out to the city, [met] the founders, and all the five-star reviews were compelling. Then we started to do manual chart review. Again, this was somewhat lucky, because we couldn't do it on 100% of charts. But it was so egregious. We found that if you came in with a cough or if you came in with a sore throat, the care was exactly the same across the board.

If you came in with a cough, you would get an inhaler for your cough and some cough medicine, cough syrup. You would also get antibiotics and steroids, which aren't always indicated. If you don't have pneumonia, you shouldn't be getting antibiotics. But the patient feels like they're getting great care because they're getting four prescriptions. From the clinic's side, it's a lot easier, too: Oh, what's your chief complaint? Cough. Here are all the medications that are going to cover anything that could potentially be wrong with you. They would also do X-rays. And so we had to stay away from poor-quality care, where it was easier for the providers. The patients even had the perception of having better care, but from the medical perspective, from the data and evidence perspective, they were actually getting terrible care.

So relying on Yelp or Google reviews wasn't enough. You had to really look at the codes and look for odd spikes in how the codes were done. For example, is every ICD code pneumonia? If you compare it to a national or regional average, it shouldn't be that way. You shouldn't have spikes in that. Or are they diagnosing bronchitis and giving antibiotics, which is a HEDIS quality issue? I can go on and on about all the different ways you can game the system. But for an acquiring clinic, or for someone who's just starting out, you have to keep track of the bottom line, which is: are you providing excellent medical care to your patients? We were just shocked at some of the behavior around the country.

Rachel: Wow. So a happy patient, a very well-medicated patient, and a well-documented visit are not all the same things.

Caesar: Right. And in fact, in these cases, the visits weren't very well documented. It was: three days of cough, lungs are clear, diagnosis pneumonia, four medications. But again, you had to go into the details of reading all the charts. Or: came in with ear pain, exam was normal, otitis media, here are some antibiotics. So there were inconsistencies there. Chart compliance, coding compliance, looking at the data, and really understanding the practice patterns of the providers is critical in any kind of acquisition. Unfortunately, most companies—even ours, with all the technological bandwidth we had at the time—didn't have the ability to do chart review across other EHRs. We could do it for our own providers, but not across other EHRs, and that became something we wish we had, in retrospect.

How to use chart data in pre-acquisition due diligence

Rachel: For folks in the audience who may not be familiar with the process of getting some of these samples of charts to test for this: if you were to do this again today, what would the order of operations be for sampling the charts, to make sure the clinical quality matches up with what it says?

Caesar: Oh, what would I do today?

Rachel: Yeah.

Caesar: Okay. Well, one of the reasons I went to Charta is because the technology Charta built is, I think, critical for monitoring quality of care and standards of care, and it doesn't matter which EHR you're on, which state the person's working in, or what their level of expertise is. You can audit 100% of charts. So what I would do is ask for the last 1,000 charts the company has, which can be downloaded in any format you want—HL7, FHIR, whatever it is. Then I would upload them to a system like Charta, where I can analyze really every encounter: by provider, by symptom, by diagnosis, by CPT code, all the variables you can think of. I would extract that using Charta, and I would look for abnormal patterns.

For example, there's a clinic in California we looked at where their ancillary care codes were just spiked. And we ended up [inaudible] learn from them, and it turned out it was impossible to replicate. The way the entrepreneur owner was running their clinic, it was impossible for us to justify the clinical need. If we had been able to audit 100% of her charts, we could have seen that, for that provider, the services they were offering weren't going to be ones we would continue offering under corporate leadership.

So step one would be: gather the last 1,000 or 5,000 or 10,000 charts. It doesn't matter, because AI is doing the work, so you can do as many charts as you want. Analyze them across provider, specialty if it's a multispecialty group, and then diagnosis, CPTs, etcetera. Then run that against a national or regional average to see if anything is abnormally spiking. From there, you can have a real conversation with the company you're trying to acquire: Hey, you're doing something that's off here, so we're going to underwrite to zero out that value in the acquisition. Or what also actually happens is: Gosh, you're doing all this work. It's documented, but your providers aren't coding properly, so your revenue should actually be a lot higher. We're going to underwrite to that. It's essentially found money.

And that's actually what we found at Charta as well. For most companies that engage with us, the revenue bumps quite dramatically, and appropriately. There's some downcoding from what people are coding in their own encounters, but in general, the found codes are much higher, and you can justify it because you can point into the chart: Where is that coming from? Why is Charta recommending this level of care versus what the clinician thought was appropriate?

Red flags in chart data that should stop a deal

Rachel: Very cool. We love hearing these real stories about the decisions you made not to acquire a pretty successful clinic. Can you share another case when Carbon walked away from an acquisition? What did you see in the charts that made you say no?

Caesar: The two major areas—just general categories—were standard-of-care compliance and appropriate coding. Going back to the California example—and obviously we have ones from across the country, but those are the ones coming to mind now—it's saying no to a clinic or clinic organization where spikes in ancillary care don't make sense, or where you see repeated patterns across different patient types. So the 10-year-old with a sore throat and the 50-year-old with a sore throat are all getting the same diagnoses and treatment plan. And this might be news to even some of the clinical audience, but if you're evidence-based, the Infectious Disease Organization suggests that in adults, you don't have to order strep or treat strep, whereas for 18 and under, there's some evidence of slight benefit to treatment.

Without getting into the weeds of it, those are the things we would really look for and walk away from, if we were lucky enough to find them ahead of time. Part of the Carbon story is that because we didn't have the ability to comprehensively evaluate all the charts, we had to rely on luck to uncover some weird practice patterns. With these newer technologies and AI applied to chart review, there's no excuse these days not to know exactly what you're getting.

In hindsight: How undercoding cost Carbon Health an estimated $30 million

Rachel: Let's end this section with some hindsight. Hindsight is 20/20. Knowing what you know now, if you could go back to that M&A period, what is one thing you would build, do, or check first before an acquisition?

Caesar: We would have built an analytical platform that really ingested all the charts and was able to evaluate those charts based on provider, place, code, diagnosis, all the components you can come up with, and then compare them to expected averages for those patient populations, regions, national averages, etcetera. One of the things we obviously, in hindsight, wish we had figured out on our own sooner was that we were really underbilling as an organization with 650 providers.

If you're in the urgent care business, you might know that your E/M weighted average should be roughly 3.7. That means if you're just doing level 3 and level 4 codes, 70% of your charts should be level 4 and 30% should be level 3. That's assuming you only do threes and fours; obviously, some have fives. In fact, we saw spikes in some clinics where 30% of their charts were level fives, and that was a red flag. Luckily, we caught that. But in the world of, let's say, making it simple with threes and fours, your weighted average should be 3.7. Carbon, at one point when I did the analysis, was at 3.1. So we were mostly underbilling.

With a million visits per year, imagine roughly a $30 to $40 swing in that move from 3.1 to 3.7. For us, that meant $30,000,000 of revenue. I don't have a finance background to say that for sure, but that's roughly how the math pans out. So if you're a clinic owner and you're really struggling—you're providing amazing medical care, and you're trying to figure out why that care isn't translating to appropriate revenue to pay your clinicians and the overhead—the codes are probably where you're losing the revenue. That was true for Carbon. If we had caught it a year or two earlier, I think the trajectory of our company would have been different. Instead of having 100 clinics today, we'd probably have 1,000 clinics today. In those early days, catching these inconsistencies and errors makes a huge difference to the future version of your company.

How to bring an acquired clinic's charting and coding up to standard

Rachel: Thank you for sharing that. We have one question from the audience that I'd love for you to answer: Once you've acquired a clinic, how do you get the new teams charting and coding up to standard without slowing down the business, and how did you do it at Carbon? I feel like it's a great segue.

Caesar: Well, first, at Carbon, we had an incredible team of chart reviewers on the RCM team, and they created a 10-chart coding test for our clinicians. As we acquired clinicians, they would come in, and we would say, look, let's look at 10 of your charts and walk you through how to code them. Then we'd randomly sample five to 10 additional charts of theirs and make sure they were in compliance. What we found was that they might do a fine job in that random sampling—need to get four out of five charts right. But even four out of five means 80% of the charts were accurate. And what we found is that over time, they didn't adhere to those standards. When they knew they weren't being watched, some degradation occurred in their coding, and it's not their fault. Doctors, providers, APPs: we're trained to take care of patients. We're not trained to do coding. So that random sampling and spot teaching ended up not working. There would be a short-term change in behavior and then longer-term degradation in the quality of coding and charting.

So we implemented Charta-type technology within the Carbon EHR, and that really made a huge difference. In pre-billing, we would prompt the provider to choose the appropriate type of code, based on their documentation. And even after their documentation, we could provide that feedback. It would be very specific feedback. It would say: Hey, here's your chart. Here are the things you wrote, but here's the code you chose, and that code is likely inappropriate. We need your input. It's likely inappropriate given there's evidence that you prescribed medications; instead of a three, it should be a four. And it's a case that could have had a bad complication, like pneumonia.

So that Charta-type technology we implemented at Carbon is what started to make a huge difference in our bottom line and quality of care. And it's really why, when I met Justin and Scott, I thought: every single outpatient provider in this country needs this technology. How do you run your clinic without being able to sample 100% of your charts and have a dashboard that shows how all of your providers are functioning, the type of care they're providing, and the types of diagnoses they're making? We've all been blind on the administrative side and the management side in health care delivery, but now there's no excuse. Now you've got these technologies that can really uncover what's going on when you're not hovering over your clinicians day to day.

Rachel: Very cool. So from operationalizing random sampling to automating and building the workflow, that's what brought you to Charta, and I love that.

Caesar: Yes.

What's next for healthcare in 2027: AI, automation, and the end of random sampling

Rachel: We're almost out of time, so I'll leave you with one final question. We like to broaden it out for the group. Looking ahead, what do you see as the biggest shifts on the horizon for health care leading into 2027?

Caesar: I think it's such an incredible time to be in health care right now. The AI solutions being developed, and those already in market, are really going to change the economics of running clinics, but a lot of that's going to come from automation, workflow improvement, standardization, and standard processes. The most important takeaway for me is that the days of ad hoc random sampling of your providers, of what's happening in your clinics, are over. Now there's no excuse not to have detailed oversight of everything that's happening, leveraging technology.

I think AI is going to automate much of what's happening in the clinics. Workflows will be truncated in a good way, so you can see more patients and do more comprehensive care. But now you also have to be able to watch in detail and have that oversight over your clinicians, to make sure your patients are well cared for and that your finance team is getting paid for what the clinicians are doing, so that you can continue to pay your providers appropriately and grow your business, so that people can [stay away from] the emergency department, where they see me.

Rachel: Caesar, thank you so much. This was a real hard look at what scaling actually takes and what the charts tell you that the reviews sometimes don't. Thank you again to everybody for joining. If you want to see how Charta helps providers review 100% of their charts, we're happy to walk you through it.

Caesar: Thank you so much.