PRODUCT FEATUREPRODUCT · CONSUMER FITNESSPLACEHOLDER CASE · RIGHTS PENDING
AICoach —
Customer-facing AI features.
+18% activation.
A model-powered onboarding flow inside AICoach's product, with full eval suite + observability for the customer team. Lifted activation by 18 points in the first eight weeks post-launch.
+18 pt
Week-one activation lift
94%
Prompt eval pass rate
<500 ms
TTFB on streamed first token
8 wk
Kickoff → live in production
01 · THE PROBLEM
Where they were stuck.
AICoach had a 47% week-one drop-off problem. Users signed up, looked at the empty dashboard, and never came back. The static onboarding form felt clinical and the team's manually-written coaching prompts didn't adapt to the user's specific goal or activity level. Activation was stuck below the rate the business model required.
02 · OUR APPROACH
How we built it.
- 01Adaptive onboarding — coaching prompts generated from goal + experience + constraints (time, equipment, injuries)
- 02Streaming LLM responses for perceived speed, with skeleton states the user reads while the response composes
- 03Edge-deployed via Cloudflare Workers AI — sub-500ms TTFB worldwide
- 04Behavior eval suite: prompt-pass-rate across 40 user archetypes, gated in CI
- 05Observability dashboard the AICoach team uses to spot bad outputs before users do
“We'd tried two off-the-shelf onboarding tools. Neither moved the needle on activation. The Kensink team scoped the simplest version that could possibly work and shipped it — and then handed us the eval suite so we own quality going forward.”
Product Lead
AICoach
[TECH STACK]
- TypeScript
- pgvector
- Cloudflare Workers AI
- Sentry
[ENGAGEMENT]
Duration8 weeks
ShapePRODUCT FEATURE
HandoffFull ownership
Warranty90 days