K is stuck because of timing and stigma, not referral mechanics. Ask only at a few early, low-stigma moments, only about the service experience — inside a free tool people would use anyway.
1 · The finding
0.0835
30-Day K (26Q1)
Target 0.30, stretch 0.42
26.8
Avg days to first referral
~20
Real usable share days
10.22%
Advocate share
~90% never refer
0.1022 × 1.27 ≈ 0.13 vs reported 0.0835 — the gap is advocacy that happens too late to count. Also: just 1.21% share within 1,000 seconds of purchase.
26Q1 published baseline, likely dated; the argument doesn’t depend on exact values.
2 · Why it happens
Timing
The KPI counts a friend who converts within 30 days of purchase. A friend needs ~7–10 days to convert, leaving ~20 usable days to share.
Stigma
People keep this treatment private — especially at visible-results moments, exactly when most programs push hardest.
Evidence
A 2026 Rice University study found GLP-1 users can face more judgment than people who lose no weight; ~18% of US adults report use, many with shame (Georgetown/GWU); a ZipHealth survey found 43% hadn’t told a partner.
Weeks 1–4: low willingness but inside the window · visible results: high but late · plateau: near zero (a retention moment) · maintenance: rare but high-trust.
3 · Regulation
Shareable
Access, speed, care team, privacy, consistency, routine.
Never shareable
Medicine/brand, dose, weight numbers, amount lost, before/after, results.
Stigma and regulation point to the same design. Working model, not legal sign-off.
4 · The four products — and why each exists
Companion (Flagship)
Problem: A tracker built on weight makes people confront the exact thing they feel judged for.
Who: People who find numbers stressful.
Why this form: A tree that grows on consistency — decoupled from body data.
What it proves: A share moment with zero sensitive content.
App
Problem: Some people want direct control and a clear record.
Who: People comfortable with a structured tool.
Why this form: A conventional dashboard, no learning curve.
What it proves: The engine works in the most ordinary surface.
Chatbot
Problem: Opening an app daily is friction.
Who: People who live in WhatsApp — opened many times a day in the UK.
Why this form: A WhatsApp chatbot — one-tap replies that feel like a care team.
What it proves: The engine works inside a platform we don’t own.
In real life: A WhatsApp Business chatbot; started via the welcome-card QR. Simulated here.
Why it matters for growth: Forwarding on WhatsApp is already a habit — sharing is native, one-to-one, and a friend can start in the same app with nothing to download.
Calendar
Problem: New apps get forgotten; the phone’s calendar gets opened.
Who: Busy, schedule-driven people.
Why this form: Check-ins where people already look; sharing = add a guest.
What it proves: The engine lives inside another product’s own features.
In real life: One tap subscribes the calendar already on your phone (Google/Apple/Outlook). Simulated here.
Why it matters for growth: Sharing borrows “add a guest” — one-to-one and private, no new behavior; public-eligible days use a read-only event link.
All four are free. The paid product is the consultation and care service, never the tools.
5 · The engine
| Window | Name | Fires when | Prospect | Pre-advocate | Advocated | Public |
|---|---|---|---|---|---|---|
| Day 0 | First log | First entry | Reflection | Reflection | Reflection | No |
| 4–6 | Access & Relief | Care plan in place | Nudge | Share | Suppressed | Yes |
| 9–11 | Care Responsiveness | ≥2 mood/symptom in 7d | Nudge | Share | Suppressed | No |
| 14–16 | Consistency | ≥10 active days in 14 | PQL | Share | Expansion | Yes |
| 18–20 | Continuity of Care | ≥3 logs/wk in wk 3 | — | Share | Expansion | No |
Day 0–30 · cutoff at Day 20
Shaded = no runway. Guardrails: max one trigger / 3 days; two dismissals suppress shares for the cycle; no share after Day 20.
7 · User-generated content
Personal note · My Story (guided prompts) · Tips (Food/Routine/Mindset) · anonymous feed with reactions only · content guard before public. No rewards for posting.
8 · Product-led growth
Free tracker → Daily habit → PQL signal → Consultation → Expansion.
PQL = ≥10 active days in 14 — the same condition as the Consistency trigger. A loss-leader, not freemium.
9 · The math
K = i × c
≈1.09×
Ceiling at K=0.0835
1/(1−K)
≈1.43×
Ceiling at K=0.30
Stigma pushes i toward zero; better pages and bigger rewards only optimize c. Retention acts as a multiplier (VCT).
10 · The money (illustrative)
~$194
ARPU / month
$350M ÷ ~150k
~2,640
Extra customers / mo
ΔK 0.22 on ~12k new
~$6.3M
Avoided acq. / yr
assumed $200 CAC
~$0.9M
Rewards / yr
+ ~$125K Q1 team
Assumptions flagged — validate before relying on any figure.
11 · The market
Crowded UK market (Voy, Numan, Juniper, pharmacy online doctors), ~£94–359/month, most already have trackers. Numan runs symmetric £100/£100, capped, clinically gated. The difference: competitors ask upfront; this asks only at earned, low-stigma moments.
12 · Push virality
A welcome card in the delivery box: QR to the free check-in on the front, a tear-off friend code on the back.
13 · The experiment
A: Access & Relief (Day 4–6) vs B: Care Responsiveness (Day 9–11)
Primary: friend conversions within 30 days. Secondary: card save rate. 50/50 split, retention guardrail.
14 · The plan
Hires: analytics owner (wk 1), lifecycle designer (wk 2–4), product engineer (mo 2).
| Initiative | RICE |
|---|---|
| Reward redesign | 37.3 |
| Community opt-in | 30 |
| Trigger engine | 28 |
| Welcome card | 24 |
| Chat channel | 20 |
| Community feed | 8.3 |
| Free non-patient tracker | 7.5 |
RICE informs order, not the thesis. 90-day sequence; done at 90 days.
15 · Frameworks
16 · Honest limits
Validate first: the 7–10 day conversion time, current KPI values, illustrative numbers, no patient interviews yet, legal review. This deliberately avoids growth gimmicks and visible referral mechanics.
17 · How this was built
Diagnose → Research → Model the math → Design the system → Write the spec → Build in phases → Test on a phone.
Thesis first, product second — every build decision traced back to timing or stigma.
My calls
The diagnosis, which evidence to trust, the trigger windows and Day 20 cutoff, the content rules, which products to build and cut, every trade-off.
Where AI tools helped
Faster research, drafting and pressure-testing copy, generating code from a detailed spec — strategy with Claude, the prototype built in Emergent.
AI tools let one person go from analysis to a working product. The judgment calls are mine.
Built in 2 days by one person — the prototype is the argument, not decoration.