COVERAGE / NM-01 · FLAGSHIP
Hope's Assistant
A New York mental-health practice was paying twice for every patient visit — once in care, once in re-keyed admin. NM originated the AI-assisted product that closed the loop.
1.0 The situation
The practice ran on two platforms that never spoke: patient visits lived in the records system, billing and scheduling in another. Every day, a staff member re-keyed every visit by hand — find the patient, read which doctor saw them, work out the right supervising doctor, attach the correct billing codes, book it again, verify. Slow, error-prone, and invisible: the kind of cost that never makes it into a meeting, because no one is hired to look at it. That entire job is what Hope's Assistant deleted.
2.0 What NM did
He identified the loop, quantified the waste, conceived the product, and defined its behavior end to end: read each visit → map the diagnosing doctor to the correct supervising doctor and billing codes → book the matching appointment automatically. Two decisions in the spec are distinctly his:
Trust is a feature. Clinic staff don't trust invisible software with billing. So the app performs every action with a visible, narrated cursor on screen — the operator watches it find, type, and click. Adoption was designed in, not hoped for.
The AI has a hard ceiling. The assistant's AI may organize information — turning free-text doctor notes into a clean doctor→codes roster — but it may never invent a billing code. Every AI output is re-validated deterministically before it touches the schedule. In a healthcare workflow, that guardrail is the product.
He then partnered with a software engineer who built the system to that spec, while NM owned the workflow design, the mapping logic, and the client relationship.
3.0 How it works
name · date · diagnosing doctor
supervisor + codes · AI re-validated
under the correct doctor ✓
The assistant runs fully unattended on a schedule: at the set time it opens the records system itself, scrapes every visit, resolves each doctor and billing code, and books the day end to end — no staff action at any point. Prefer to drive? Manual mode is one click: pick the day, press run, and watch it work. A dry-run option can also preview a full day's plan before committing — the same instinct as checking a trade ticket before sending it.
4.0 Why this case leads the book
Because it's the analyst's job in miniature: find the mispriced thing (hours of skilled labor spent on re-keying), define the trade (a product spec with the right guardrails), and see it through to execution (shipped, adopted, still running). The domain happened to be healthcare operations. The skill is transferable to any desk that pays for clear thinking about where value leaks.
A report nobody acts on is noise with formatting. A product a clinic actually runs every day — that's a conclusion someone acted on.