AI that's tackling healthcare’s biggest problem

How Abdel Mahmoud is disrupting healthcare’s trillion-dollar administrative costs 

Healthcare is one of the most complex industries in the world, with myriad stakeholders and systems, vast amounts of data, a maze of regulation. That’s also what makes it one of the most potentially disruptable industries in the global economy. But to date, many efforts have created one-sided wins that benefit either the physicians or their patients; patients or insurers. 

We believe that true success will need a three-sided win: for the clinician, the patient, and the insurer. That’s where Anterior, an AI co-pilot for healthcare administration founded in New York in 2023, comes in. Large language model–powered and built to manage all the work that comes before clinical decision-making, Anterior is taking on embedded administrative overhead in healthcare. In the U.S. alone, this represents a cost that potentially reaches hundreds of billions of dollars per year.

NEA’s lead partner on the investment, Mohamad Makhzoumi, recently spoke with Anterior  co-founder and CEO Abdel Mahmoud about the company, its vision, and why the opportunities for technology in healthcare are endless. 

[Mohamad] Your journey to Anterior goes back to medical school. How did it all begin?

[Abdel] I went to medical school wanting to make a difference. But I quickly realized that how good you are as a clinician is less about your individual skill set, and more about the ecosystem and tools you have, and the environments you’re in. I started to see what was broken — and where technology could help. That spurred my interest back into computer science and tech. I got a master’s degree in computer science, and went to work on product for Facebook, then Google. 

I kept thinking about the opportunities for technology in healthcare. They’re almost endless. You walk into a healthcare system and you’re thrown back 15, 20 years. Tech advancements and product-centric, user-centric design hasn’t filtered through yet. A big part of that is the unique constraints of healthcare.

[Mohamad] The user experience in healthcare is analog as a patient accessing the system, but even more so for clinicians and administrators. The No. 1 purchaser of fax machines in the world is the National Health Services in the U.K. And the No. 2 is the Centers for Medicaid & Medicare Services in the U.S. We have iPhones, laptops, autonomous vehicles, but the bar in healthcare is fax machines. 

[Abdel] And pagers.

[Mohamad] How big is this opportunity?

[Abdel] Of the $4.5 trillion that’s the U.S. healthcare system, nearly a trillion dollars is administration. It’s pure information flow and bureaucracy. And it’s not just that the user experience is terrible. The systems require a lot of manual work. That’s the bigger problem. You have very highly trained, expensive resources — clinicians and other staff — spending hours on the kind of stuff that should be invisible: all the backend billing, coding, collecting, calling that is purely manual. 


[Mohamad] Every minute a clinician has to spend on non-diagnostic care is a waste. Every nurse practitioner who sits in a call center reviewing charts could be treating patients.

[Abdel] And that is the Anterior journey. Doctors, nurses, and prescribers are drowning in administration and burnt out. We asked ourselves, what could we do? And we realized something interesting.

When a health insurance company makes decisions, the underlying data is often clinical data: previous medical records and claims, and details about the member population. Because that data is hugely unstructured, you need expensive jargon-understanders — medical professionals — to go in and look at the claims. What if we could free up that resource by creating an intelligence system?

[Mohamad] Authorizing a medically necessary procedure, and paying for that procedure, are such manual processes. So time intensive, so expensive — and so critical. How much time does it take on average for a human to go through all the different electronic medical records, faxes, and prescriptions, and pool them to say whether this surgery is necessary?

[Abdel] The short answer is four hours for one single case. Hundreds of millions of these authorizations happen every year. At the other end is a patient waiting for care. It’s an interoperability problem. A clinician files a request. The insurer needs clinical evidence, so it prints out a 400-page electronic medical record. That gets faxed, printed out, scanned, and then text recognition is applied so that it can get put back onto a computer. It was electronic when it started, it’s electronic at the end, but there’s this gulf in the middle. 

We realized this is the best place to start: uplifting clinicians, freeing up that resource from all the administrative back-end work. This can also help solve the labor crisis. Medical professionals are retiring faster than they’re graduating.

[Mohamad] And you’re doing it by combining aspects of LLMs with other technology. That’s different from how most people think of AI today.

[Abdel] If you look at any company that’s building with AI and doing well, you’ll find it’s not just an LLM, — or even multiple LLMs. It’s also a lot of conventional computer science techniques, and computer science paradigms too.

Take a plane as an analogy. What powers it are jet engines. But the engine maker isn’t making the plane. LLMs power what’s possible. But they’re not enough. You need scalability, robustness, databases, reading and writing data, good platform design, good UX, good front-end experiences, reduced latency. All of these things come together. 

That’s our approach: LLMs playing in sequence with all the other tools.

[Mohamad] That’s important because healthcare is very different from any other vertical where AI is aiming to enter.

[Abdel] The stakes are really, really high. Almost any AI system when it first deploys will be 60%, 70%, even 80% good enough.

[Mohamad] That’s not good enough in healthcare.

[Abdel] Absolutely not. That’s why we’re starting with administration. It’s about a financial outcome, and you are deploying inside a sandbox. You’re still getting volume and training, and improving. The outcome is: do we pay for this, or do we recommend that? 

[Mohamad] So much of the zeitgeist around technology and healthcare is whether doctors and nurses will be willing to adopt the technology. My perspective and part of our investment thesis around Anterior is that it’s saving the clinician time, but it’s all in that backend. It’s working through that billing, coding, collecting part of healthcare that the physician and patient never touch. But it’s the single biggest bottleneck to improved care, quality, and access.

[Abdel] Say a clinician gets 100 pages of facts, which they have to compare to some sort of criteria. What if a system could look through those pages and match the criteria, find all the evidence and lay it out for you? Now you can review 30 cases because each takes a minute. It’s speeding up the process, but keeping the clinician as the main decision maker. When the specialist charges thousands of dollars an hour, you’ve gone from 10 hours to 30 minutes. Immense value.

[Mohamad] All of this is what got us so excited about Anterior. We’ve looked at every flavor of AI for healthcare. We’ve never seen a product this robust, this impactful.

[Abdel] NEA is really the best partner for why we’re here. We’re here to figure things out, together.