AI Product · Fractional CPO Under NDA

Validating the core product before the runway ran out.

As embedded fractional CPO, I set product and design direction and used AI-assisted tooling to ship testable product in days — so the team could learn from real users fast.

Year
2024 – Present
Role
Fractional CPO
Product strategy + design direction
Client
AI fintech · US
Founders + engineering
[ HERO VISUAL ] — anonymized diagram / NDA-safe abstract of the learning loop
The Problem

Limited runway, no design function — and every week building the wrong thing was existential.

They needed to find out what users actually valued — fast — without the time or budget for a traditional design-and-research cycle.

The Solution

A tight operating model built for learning velocity.

Set a sharp product hypothesis, ship a working slice of it in days using AI-assisted tooling, put it in front of real users, and let the signal decide what to build next.

01
Sharp hypotheses
Reframed a broad roadmap into testable bets ranked by risk.
02
AI-assisted shipping
Used AI coding/design tooling to compress build time so we could test ideas in days, not sprints.
03
Instrumented for signal
Built measurement in from the first release so every bet returned data.
04
Double down
Killed what didn’t move, invested in what did.
Background
+ To add · NDA-safe · medium
2–3 sentences of NDA-safe context — the market/problem space, the stage of the company, why speed was the whole game. No names or identifying detail.
Research

Research was continuous and embedded — every shipped slice a live experiment.

With no time for a long discovery phase, I worked directly with founders who had deep domain knowledge, mined existing customer conversations, and treated every release as a live experiment.

+ To add · medium
How user signal was sourced — support tickets, interviews, usage data — without breaking NDA.
+ To add · generic
The user archetype, kept generic — e.g. “small-business owners managing cashflow,” without identifying the client.
How Might We…

…validate the riskiest assumption in the product with real users in weeks, not quarters — without a full design and research team?

Ideation

The bet was that velocity of learning beat polish at this stage.

AI-assisted tooling let one embedded leader do the work of a small team: I could move from idea to a credible, testable artifact fast enough that the company learned something new every week.

+ To add · low
Name the kind of tooling / loop without naming proprietary client details.
Design
+ To add · medium
What you designed — the core flows, the design language you stood up quickly, how you balanced speed against a coherent experience. Even anonymized, describe the design decisions that mattered.
[ IMAGE ] — anonymized flow / design language snapshot
Testing & Iterations
+ To add · anonymized · medium
2–3 anonymized examples of a bet that was tested and what the signal told you — “we shipped X, users did Y, so we changed to Z.” Generalize enough to stay NDA-safe.
Results
+ To add · highest impact · NDA-safe
“Cut time-from-idea-to-test from ~N weeks to ~N days,” “validated/killed N hypotheses in N months,” or “helped the team focus the roadmap around the one feature that retained users.” An anonymized metric is fine and expected.
Final Thoughts
What I learned
+ A genuine lesson about shipping fast with AI in an early-stage, high-stakes context.
Next steps
+ Where the work is heading.
Most proud of
Proving that a single embedded design-and-product leader, armed with AI tooling, can give a cash-constrained startup the learning velocity that used to require a whole team.
Next project →
AI Productivity [NDA]
↗
Like what you see?

Let’s build something together.

felipelebrun@gmail.com →
© 2026 Felipe Restrepo LinkedIn ↗ Behance ↗ Remote — Colombia / France · EN · FR · ES