The Person — Sumit Mahendru, Founder and CEO
900 Facilities. 900 Logins. One Team Trying to Keep Up
Sumit Mahendru, Founder and CEO of Abstrakt Health, has been working with Dexit for close to a year now, with about ten months of that time in active use.
Abstrakt handles revenue cycle management for a range of provider clients, and senior care providers are one segment of that business. These are clinicians who go into assisted living and skilled nursing facilities they don't own to treat patients. That means no direct access to the facility EHRs. Abstrakt's largest client in this segment alone touches roughly 900 different facilities across California. Nine hundred different EHRs. Nine hundred different logins.
A Stack of Paper, Read One by One
Before Dexit, the workflow started with a stack of paper. Facilities scanned face sheets from each patient visit and sent them to Abstrakt Health, where they sat in a document storage system until someone opened them one by one, pulled the demographic and insurance information out manually, and worked out which details were actually correct.
That doesn't scale, not at the volume Abstrakt was running. Sumit points to one client as the clearest example: 7,000 to 12,000 claims a month, each with its own face sheet, each needing that same manual read-and-extract step. An API connection wasn't an option either. And getting it wrong was never a small thing. One incorrect digit in an insurance number can mean a patient gets billed for a service that should have been covered.
If one field or one digit is incorrect in the insurance number, then potentially a patient is billed or denials are happening, which creates such havoc on the back end and so much more cost for both me and my practices, as well as potential headache for patients.
No Time Wasted on Healthcare 101
Sumit shopped around. Other vendors doing document extraction with AI, OCR, or RPA all got a look before Dexit won the business. What separated the conversation wasn't accuracy alone. It was that 314e already knew healthcare and revenue cycle management, so Sumit didn't have to burn two calls just teaching a vendor what a denial even is before any real work could start.
It wasn't like, hey, we're going to spend two calls with me teaching you what healthcare is, and what the importance of getting corrected information is. It was, hey, we've done this before, we've done Epic implementations, and that gave me a comfort that otherwise wasn't there.
That existing domain knowledge saved real time off implementation. There were no month-long scoping calls to explain why accuracy mattered or what the data even meant. The pitch was direct. "Send over some documents," Sumit says. "Let us extract them, and we'll tell you what's missing."
Filling In the Edges, Not Starting From a Blank Canvas
Dexit's AI was already built for healthcare documents, so onboarding never turned into a from-scratch buildout. Most vendor onboarding, in Sumit's experience, starts with a blank canvas. This time the canvas already existed. The work was refining it: dialing in Abstrakt's specific document types and the fields particular clients needed.
I believe onboarding is truly painting a picture. With most companies, it's a blank canvas. With you guys, it was like, hey, we're just filling in the edges and the outlines, and cleaning up what is missing or added.
Sumit puts a number on that difference: 30 to 45 days saved in onboarding. The team also worked around Abstrakt's overseas operation in India, running separate call slots across time zones so both sides could show up without anyone losing sleep over it.
Accuracy High Enough to Retire the Human-in-the-Loop
Extraction accuracy for Abstrakt now sits above 95%, and it keeps climbing. High enough that Sumit and his COO are planning to drop the human-in-the-loop review step entirely before the year is out.
The conversation and the joke I have with my head of ops: hey, if you had an employee at this percent accuracy, what would you do? Typically the answer is, well, I'd probably give that employee a promotion or a raise. So logically it makes sense that we will remove the human in the loop because the accuracy is such high.
A Vendor That Still Treats a Small Business Like a Priority Account
Sumit judges support responsiveness, and he's tested that repeatedly by pushing the team toward building EOB extraction, something he'd chased for years with other vendors and once tried to build himself at an AI company. His description of how 314e responds is consistent: take on the ask, don't deflect it.
Finding the right vendor mattered to Sumit for another specific reason. He didn't want to end up as a small account sitting one contract renewal away from being outgrown. He'd seen it happen before: a prior vendor shifted its focus toward larger health systems, then priced him right out as a result.
I feel like 314e and Dexit have two unique ICPs. You do a lot of work with health systems, but you're working with customers such as myself. I'm a small business, and a fear for me as a company is that vendors will stop servicing a business like mine. I feel like you guys have a dedicated team to focus on small companies, which is very helpful and very comforting as a business owner.
Expanding From Face Sheets Into EOBs, With One Vendor Instead of Ten
I've referred over, probably 10 or 15 of my peers, RCM companies, over to Dexit. I don't make referrals lightly.
Abstrakt's roadmap with Dexit now stretches past face sheets into EOB extraction, with early accuracy numbers that Sumit calls promising. But the bigger shift he's chasing is consolidation. Instead of juggling ten vendors for ten different document types or processes, he wants one platform that handles all of it.
We started on face sheets, we're moving on to processing more and more new document types. That's been the game changer for me because I don't have to go to ten vendors to do all my different document types.