Ignition by RocketTools

Dan McCoy, MD

Healthcare is getting optimized by AI. But optimized for whom? Ignition by RocketTools breaks down the systems, incentives, and technology reshaping how care gets approved, denied, and paid for — with data, not hype.

  1. Aug 6

    How the Government Cracked Enterprise AI — When 95% of Companies Fail

    Everyone treats enterprise AI as a technology problem — which model, which vendor, GPT vs. Claude vs. Gemini. But an MIT report found that 95% of enterprise AI pilots deliver no measurable return. Most never make it out of the pilot. So why did the U.S. State Department succeed where almost everyone else stalls? They built an internal AI assistant called StateChat and scaled it to more than 62,000 employees across 98% of diplomatic posts worldwide — then published a remarkably candid playbook showing exactly how they did it. In this episode, Dan McCoy breaks down what any leader can steal from it: why they fought the boring security and compliance fight first, why they shipped an embarrassingly small tool before it was "ready," the unglamorous pilot that saved 76,500 hours and $6M — and the single biggest barrier to adoption they uncovered, which had nothing to do with the technology and everything to do with whether people felt they were allowed to use it. If you're trying to get AI adopted inside a real, cautious, regulated organization, this is the closest thing to an answer key you'll find. — LINKS — ▶️ Watch the video version: https://youtu.be/oNiiJS-e2LE?si=SEREnuZoZ9CMu7w8 📄 The State Department Generative AI Playbook (free PDF): https://www.state.gov/wp-content/uploads/2026/07/DOS-Generative-AI-Playbook_July-2026.pdf 📝 Read the companion essay, "Probably Not Allowed": https://danmccoymd.substack.com/p/probably-not-allowed — Subscribe for no-hype breakdowns of AI in healthcare and the enterprise.

  2. Jul 31

    You Didn't Lose Your Job. You Got a Funded Sabbatical.

    What to actually do with the severance, the time, and the expertise that just walked out the door with you. I'm going to make an argument that sounds insane for about ninety seconds, and then I hope stays with you for the rest of your career: getting pushed out of your job might be the single best thing that happens to it this decade. I'm not saying that from the cheap seats. A while back I left the CEO seat of a company through a buyout, so I've sat on both sides of this. I've handed someone the packet, and I've carried my own box to the car. This episode is what almost nobody does with that moment, and what a small number of people are quietly using it to do instead. Press play above. If you'd rather watch it, the video version is right here on YouTube. What this episode is really about The story in your head is that getting let go is a stain, a gap you'll spend two years explaining. That story is about fifteen years out of date, and the numbers say so. Median job tenure in the U.S. is down to 3.9 years, the lowest in two decades. Nearly one in four workers were laid off or fired in the past year. And more than half of this year's layoffs name AI or restructuring as the cause, not your performance. A spreadsheet got reorganized and your name was in the wrong cell. Meanwhile the thing that comes with getting cut has quietly gotten better. Average severance ran about nineteen weeks of pay in 2024, up roughly seventy-five percent from a few years earlier. A lot of packages now include garden leave, a real legal category where the company pays your full salary while barring you from competing for a while. We have a word for being paid not to work while you figure out what's next. It's called a sabbatical. You just got handed one, and you're treating it like an emergency. The 90-day plan we walk through The whole episode is really about what to do with that time instead of panic-applying to the exact job that just let you go. In short: Days 1 to 7: Excavate. The most time-sensitive asset you own isn't the severance check. It's the twenty years of pattern recognition in your head, and it goes soft the moment you stop using it. Before you open a single job board, dictate everything you know. Pattern knowledge, not their property.Weeks 2 to 4: Build the knowledge base. Turn that pile of dictation into a custom AI trained on your own expertise. Building it forces you to actually learn the tools. The excavation is the bootcamp, and you end up owning an asset no one else on earth could build.Weeks 4 to 8: Ship one thing a week, in public. However small. One artifact out into the world every week. Not one more course watched. One thing built.Weeks 8 to 12: Point it at a door. The same body of work feeds three exits at once: consult, build a one-person business, or go back as a far more valuable hire or a fractional one. You pick the door in week ten with a portfolio behind you, not in a panic on day two.And the most expensive mistake to avoid: do not go buy the ten-thousand-dollar AI course. Free courses get finished by five to fifteen percent of people. A course sells you content, and content is the one thing AI just made free and infinite. Spend on a finishing structure instead, not on information. The part that makes this moment different There's a Harvard and BCG study everyone quotes and almost nobody reads correctly. They gave consultants GPT-4. On the tasks the AI was good at, people were about forty percent higher quality and twenty-five percent faster. On tasks that looked similar and weren't, those same people were nineteen percentage points more likely to be wrong. The only thing that told you which side of the line you were on was domain judgment. Experience. The exact thing you just spent two decades building. That's the whole game. AI made good-looking output cheap and infinite. It did not make judgment cheap. Domain expertise plus AI fluency is the arbitrage, and right now you're one of very few people holding both halves. One honest caveat No hype, so I say this in the episode too: this is real and growing, but it's stratified. If you're a contractor, hourly, at a small shop, or you were let go for cause, you may get nothing. Non-competes are still enforceable in a lot of states. Read your paperwork and talk to a real employment lawyer before you do anything clever. This is a genuine perk of salaried white-collar work, not a safety net that catches everyone. Don't do this alone The fastest way a ninety-day sabbatical rots into six months of doomscrolling is doing it by yourself in a room. That's why we're standing up Vibe Lunch, a group of people learning these tools and applying them in their own businesses, trading what works. Join that or join something else. A community is the difference between a plan and a habit. Listen above, and if it lands, forward it to the one person you know who just got walked out. They need to hear it in the first week, while it's all still sharp. I'm Dan. Go build something.

  3. Jun 2

    How One Reused Password Cost Change Healthcare $2.5 Billion (Healthcare Security, Part 1)

    In February 2024, hackers walked into the largest healthcare clearinghouse in America through a Citrix portal that didn't have multi-factor authentication. They used credentials stolen from a previous breach — someone, somewhere, had reused their password. Within hours they had ransomware running. Within days, pharmacies across the country couldn't fill prescriptions. The ransom payment was $22 million in Bitcoin. The total cost to UnitedHealth Group is now over $2.457 billion. The number of Americans whose data was exposed is 192.7 million — roughly 58% of the country. And it all started with one reused password. This is Part 1 of an 8-part Healthcare Security series. In this episode I walk through why your password habits are probably just as dangerous, why "47 logins" understates the reality for healthcare executives, and the four password managers I actually recommend — with the honest tradeoff on each, and no affiliate links. I also explain why a single patient record sells for $250 on the dark web while a credit card goes for $5, and why your AI tool account in 2026 holds more sensitive information than most of your work files. In this episode: The Change Healthcare breach timeline and what Andrew Witty admitted under oath to CongressWhy password patterns ("FirstName2024!" and friends) are now exactly what attackers test firstThe 47-logins-on-average problem for healthcare execs and why the real number is higherThe four password managers I'd recommend: Dashlane, 1Password, Proton Pass, and Bitwarden — pricing, tradeoffs, who each is right forA four-step action plan you can run this week, starting with one email to your IT team📺 Watch on YouTube: https://youtu.be/N62kieISWiI 📝 Read the director's cut companion post on Substack (deeper on Witty's Senate testimony, the dark web pricing texture, and the AI tool risk section I had to cut for time): https://open.substack.com/pub/danmccoymd/p/the-872m-password-mistake-was-actually Next week, Part 2: why CISA and the FBI told Americans to stop using SMS-based MFA, the authenticator app I switched to after leaving Microsoft Authenticator, and the small piece of hardware I added on top. I'm Dan McCoy. Ignition by RocketTools is the podcast for healthcare executives, physicians, and AI builders trying to think clearly about where this is all going.

  4. May 29

    The Target Story Isn't About Coupons. It's About Healthcare AI.

    Twelve years ago, a Target statistician built a model that could predict pregnancy from 25 shopping items. The story usually gets told as a privacy parable. I'm telling it differently — as a preview of how healthcare AI is going to work for the rest of our lives. Your smartwatch can already flag atrial fibrillation days before a cardiologist would. It can detect depression weeks before clinical scoring catches it. It can spot cognitive decline six to twelve months before you notice. The science isn't the question anymore. The question is who gets to see what comes out of the model — and whether we build the governance before the surveillance economy locks in. In this episode I get into: • Why Andrew Pole's 2012 Target model was a dry run for what's coming in clinical AI • What the wearable accuracy numbers (70–95%) actually mean — and where they get softer than the headlines suggest • The function-creep economy that's already running: CGM data sold to ad partners, period-tracking app subpoenas, life insurers bidding on de-identified wearable sets • The 99.98% problem — why "de-identified" data isn't • Habit-disruption windows: the real case for early-detection surveillance in healthcare • Four policy moves that would change the data-broker incentive structure overnight Full written companion with sources and citations: danmccoymd.substack.com Watch on YouTube: https://youtu.be/LbE6TbGIzIY I'm Dan McCoy. Ignition by RocketTools is the podcast for healthcare executives, physicians, and AI builders trying to think clearly about where this is all going. New episode every week.

  5. May 28

    The 9-Person Insurance Company and the Real Line in AI-First Healthcare

    Y Combinator has a name for it: burn tokens, not headcount. A health insurance company called Decent runs with nine people total. Twofold does revenue cycle management with three. Deep Cura Health handles patient scheduling, prior authorization, and insurance verification with two humans and seven AI agents. The AI-first model is real. It's working. And in healthcare, every one of these companies has made the same quiet choice: they're attacking admin, not clinical. This episode unpacks why — and why the conventional wisdom ("admin is safe to automate, clinical isn't") is the wrong frame. The real line isn't admin versus clinical. It's decision support versus decision making. IBM burned $4 billion learning the difference with Watson Health. The next generation of healthcare AI companies will either learn from that, or rebuild the same trap with better UX. What's covered: How AI-first companies are quietly rewriting healthcare staffingWhat IBM Watson Health actually got wrong — and why "the AI was wrong" misses the lessonWhy most "decision support" products today are decision-making in a trench coatThe payment-model problem nobody is pricing: a 5,000-patient panel breaks fee-for-serviceThe companies positioning themselves on the right side of the line🎥 Watch on YouTube: https://youtu.be/9fHKQqm15qo 📝 Companion essay (with the receipts I had to cut for length): https://danmccoymd.substack.com/p/the-part-of-ai-first-healthcare-that

  6. May 21

    Are the Blues AI-Ready? Blue Cross vs. the Optum Platform Race

    In March 2026, the Blue Cross Blue Shield Association published research blaming hospitals' AI billing tools for $2.3 billion in added healthcare costs. It was a grievance — not a strategy. And it stands in sharp contrast to 1981, when the same Association faced a national-platform problem and built something: BlueCard, the shared claims-routing layer that turned 36 independent regional plans into the reason one in three Americans carry a Blue card today. This episode asks the contrarian question: in an AI world, is the Blues' patchwork of 36 plans a fatal weakness — or the exact architecture the future rewards? What we get into: Why Elevance and Highmark are racing in opposite directions (multi-vendor horizontal vs. Epic single-stack vertical) — and what the other 34 plans aren't doing The plan-level wins that already shipped (BCBS Minnesota, Illinois, Arkansas) — and why none of them are federated The federated-learning research that says the Blues' structure is the ideal AI architecture — including a peer-reviewed BCBS Louisiana study where regional models beat national algorithms The Optum problem: what happens if UnitedHealth builds the AI equivalent of BlueCard before the Blues do Three concrete signals to watch by the end of 2027 Full companion essay on Substack with sources and the three-signal checklist: https://open.substack.com/pub/danmccoymd/p/blue-cross-built-the-last-healthcare Watch the video version: https://youtu.be/LCrRFqTTtuo Connect with me at RocketTools.io for AI Strategy Consulting and podcast or speaking engagements.

  7. May 20

    The Hospital Cost Crisis: How Washington Banned Cheaper Care

    You've been told American healthcare is expensive because of greedy insurers, pharma profits, or the cost of innovation. That story is incomplete to the point of being misleading. The largest single driver of US healthcare spending isn't drug companies — it's hospitals. And hospital prices haven't merely risen; they've grown roughly 3x faster than overall inflation since 2000. No sector does that for two decades by accident. In this episode, Dan McCoy MD breaks down the three federal policy choices that designed America's hospital pricing crisis: • ACA Section 6001 — the 2010 ban on new physician-owned hospitals, the one competitor proven to be roughly a third cheaper. $2.2B in planned development killed; 75 hospitals never built. • Certificate of Need laws — still active in 41 states, letting incumbent hospitals veto their own competition. • Site-specific Medicare payment — paying hospitals 2–3x what it pays a physician office for the identical service. Add a starved FTC (~13 challenges out of ~561 hospital mergers from 2010–2015) and you get the result: ~97% of metro areas with highly concentrated inpatient markets, and prices that rise 15–30% higher than competitive ones. It isn't a mystery. It's a mechanism. If you run a health plan, here's the takeaway: your hospital costs are set by market structure, not market forces — and the policy landscape (site-neutral reform, Certificate of Need repeal) is finally starting to shift. 📺 Watch the video version: https://youtu.be/aWtOw8PxHTU 📝 Full research sources, all 12 cited studies, and a bonus analysis of the political economy of hospital lobbying — on the Substack Subscribe so you don't miss the next episode. This episode is for educational and informational purposes and is not medical, legal, or financial advice.

About

Healthcare is getting optimized by AI. But optimized for whom? Ignition by RocketTools breaks down the systems, incentives, and technology reshaping how care gets approved, denied, and paid for — with data, not hype.