At a $120M ARR SaaS company, I rebuilt how product, design and engineering work, around AI. In six months, releases went from 20 a quarter to 81, ideas reached customers in weeks instead of months, and retention rose five points. Headcount did not change.
This is the process, what changed for each role, and the practices I bring to every engagement.
Every step has a clear split: what the AI does, and what a named person decides. That split is what makes it fast and safe at the same time.
AI: Reviews every customer call, ticket and CRM note.
People: You see every request, not a sample.
AI: Groups requests and shows the revenue behind each.
People: You and your team pick what to build.
AI: Drafts the spec, the design and the code.
People: Your people review and own every piece.
AI: Tracks each release with the customers who asked.
People: You see what moved retention.
The time AI saved went into better decisions, quality and the next bets.
| Before | After | |
|---|---|---|
| Product managers | Wrote tickets and chased status | Own the decision and the customer evidence behind it |
| Designers | Drew every screen by hand | Own the design system and choose among AI-drafted options |
| Engineers | Wrote every first draft | Design the architecture, review AI code and own quality |
| Leaders | Argued over priorities | Decide from ranked evidence and measure the result |
Retention measured on customer cohorts before and after the most-requested fixes shipped. The company is not named; details can be shared privately.
Case study (PDF) →| Before | After | |
|---|---|---|
| Net revenue retention | Baseline | +5 points in one quarter (cohort analysis) |
| Idea to production | 4–6 months | 2–4 weeks |
| Features released per quarter | 20 | 81 (2× by month 3, 4× by month 6) |
| Customer conversations analyzed | A sample | 16,000+ a year, refreshed weekly |
| Product, design and engineering headcount | Baseline | No change |
Review every call, ticket and CRM note. Samples hide the requests that cost you renewals.
Ten requests from small accounts can matter less than one from the customer about to churn.
A PM states what we are building, for whom and why. Only then does AI draft the spec.
AI drafts. Named owners review and sign off. Nothing ships unreviewed.
A PM, a designer and two or three engineers own one problem end to end, from customer request to release.
Small releases are easier to review, easier to roll back and faster to learn from.
Tell the customers who asked when it ships, then check whether they use it.
Redesign the work first. Cutting people before the process changes is how output collapses.
Roadmap Sprint → Run the process once, in 30 days, on your top customer request.
Fractional CPO → I set it up across your team and run it with you.
Book a free diagnostic: 45 minutes with me, no preparation needed. You leave with three observations on where value is leaking and an honest view of whether I can help.