ОТНОСТНО ТОЗИ ЕПИЗОД
Anshul Rathi joined Oscar Health as one of its earliest employees and inherited a single spreadsheet. Every network was rented. He built the first system himself, one spreadsheet into fifteen spreadsheets talking to each other, and then it broke. He walked into the executive team and asked for engineering resources. What came back was a five-year roadmap and a $50 million price tag to build it in house.
So he looked at every vendor on the market and found the same thing everywhere: point solutions that didn't talk to each other, each one fixing a single symptom. Six years later Oscar had gone from two states to twenty, more than 40,000 providers credentialed, over 100 delegated hospitals. Somewhere in there he picked up the nickname the one man credentialing army, and the conclusion that the industry had misdiagnosed itself.
His reframe: provider data isn't a data problem. It's a trust problem wearing a data costume. Every health plan, every hospital, every state has its own version of "trust me," none of them talk to each other, and all of them still want a screenshot.
Nick Helfrich and Mitchell Gorodokin break their own rule and spend an episode on CertifyOS: the diary he filled with twelve pages about how the system should work while he had Covid in India, the battle that went all the way to a national plan's chief legal officer over whether a screenshot was ever legally required, why he walked away from the payer enrollment business and the revenue that came with it, and why he thinks a company like his shouldn't need to exist in the end state.
In this episode:
• One spreadsheet to twenty states, and the moment the buy-versus-build math stopped working
• Why the Federated States of America means fifty different healthcare ecosystems
• The accidental entrepreneur: a diary, a wife who told him to brush up his coding, and fifteen days
• Everyone says trust me, and everyone wants screenshots. How they got one plan to accept JSON instead
• Dorks vs. posers: swap "provider data" for "supply chain" and does the pitch still hold?
• What actually separates agentic AI from a bulk trigger pipeline with better copywriting
• Two models pointed at each other, and a hallucination rate that drops from 3-4% to 0.09%
• Why credentialing and payer enrollment keep getting conflated, and what breaks when they do
• Why the point solution era is ending, and what disintegrates a system of record
• The airport with no TSA, and what shared credentialing is actually trying to be
• Why 70% accuracy is the industry's gold standard, and why nobody should be celebrating that
• The dork litmus test: remove the word AI from your pitch deck. Does the plan still buy it?
Subscribe to Provider Data Dorks for the people fixing healthcare's hardest data problem. New episodes weekly on YouTube, Spotify, and Apple Podcasts.
Timestamps:
00:00 Intro
03:48 It's a trust problem wearing a data costume
05:16 The accidental entrepreneur: a diary, Covid, and fifteen days
08:56 Everyone says trust me, and everyone wants a screenshot
12:15 Dorks vs. posers: who actually has the scars
16:08 Agentic AI, or a bulk pipeline with better copywriting
19:24 Why the point solution era is ending
38:41 An airport with no TSA
40:57 Why he says CertifyOS shouldn't exist
55:03 Remove the word AI from your pitch deck