Where the hours come from.
The assembly is what caps a caseload. Cut it and the ceiling lifts.
34 min saved per patient × ~135 patients across three Nudge Friends × 26 cycles a year = ~1,990 hours freed.
Illustrative, using discovery figures and a 38→4 minute assembly estimate. We calibrate every number against your real caseload and channel volumes before anything is committed.
The same team, holding more.
Drag to your real headcount. Capacity assumes about 45 patients held well by hand versus about 120 with assembly offloaded: the same care depth.
The return doesn't sit still. It grows.
The gap between what your team can hold by hand and what they can hold with the system widens over time. That's the curve that makes the insurer play possible.
New hires start amplified
Every Nudge Friend who joins begins at the higher baseline of about 120 patients from week one, up from the manual 45. Nobody starts from the old ceiling again.
Per-patient time keeps falling
As the context layer deepens and drafts need less editing, the four minutes shrinks. The system gets cheaper to run per patient the longer it runs.
The insurer play becomes viable
Serving 10,000 on a high-touch model by hand is impossible economics. This is the unit-cost curve that makes B2B2C real.
Old cost versus new cost: everything in.
Cost is hidden.
- About 1,990 hours a year of your scarcest, most expensive people spent stitching the picture together.
- A hard ceiling on patients per counsellor, so you hire to grow.
- The read lives in someone's head. Quality varies, and it walks out the door when they do.
Cost is small and flat.
- Running cost is agentic compute plus hosting: a few hundred dollars a month at your scale, and it falls per patient as you grow.
- No per-seat licensing that balloons the way a make.com bill does when you add users.
- You own it. Spec-driven and re-specced in days when your model shifts, on your own roadmap.
Even if the running cost were higher than a shelf tool (it isn't), the trade is your rarest resource, counsellor time, for cheap and flat compute. That maths only improves with scale.
Where we'd start.
A focused first build of the Tier-1 wedge, the Nudge Friend view, on a slice of your real caseload, proving the 34 minutes and the assembled read against your own patients before anything scales. Every use case you listed is sequenced on its own tab.