Scaled an AI diagnostic from 5 pilot markets to 42
The case in 30 seconds
- Problem
- SkinConsult AI was live in 5 pilot markets and already the group's leading skin analysis service. Competition and new internal launches called for a refresh — and the service could not travel as built. Every new market was a bespoke project, and the journey asked questions that did not change the diagnosis.
- What I did
- Cut the journey to the questions that actually changed the result. Then turned the service into a standardised product, with a fixed set of mandatory inputs and dependencies. Built the market onboarding pack and led the rollout across 7 regions into the remaining 37 markets.
- Impact
- 2 million completed diagnoses by the end of 2020, generating €1.16M in incremental revenue. The service is live in 42 markets.
How it happened
- 2020 — Where it started
The pilot worked. Five markets, the group's leading skin analysis service, and no repeatable way to extend it. Each new market was its own project. The question was not whether the product was good. It was what had to become standard before it could travel. VICHY owned the brand; ModiFace built the generative adversarial network (GAN) engine.
- The architecture decision
Standardise the product, not each deployment. I cut the journey to the questions that changed the diagnosis and fixed the mandatory inputs. I then re-architected data ingestion, so a market plugged into regional cloud infrastructure instead of commissioning a bespoke integration.
- The product
Removing questions that did not change the diagnosis shortened the journey and raised completion. A personalised results page converted into the e-commerce basket far better than the general site. The diagnosis was doing the merchandising.

The guided scan, and the scored profile it produces. - The rollout
Every market received the same onboarding pack: mandatory inputs, dependencies, and milestones stated up front, before anyone committed a date. The rollout ran region by region across 7 regions, so each wave carried the lessons of the one before it. Performance targets were agreed with research and development (R&D) ahead of launch, then held.
- Today42markets for computer-vision rolloutModiFace AI/GAN to production, L'Oréal2Mcompleted diagnoses by end of 2020Skin Consult AI, across 42 markets€1.16Mincremental revenueFrom 2 million completed diagnoses+17%conversion uplift from computer-vision AIModiFace GAN, Skin Consult AI (VICHY)
What I took from it
A pilot proves the product. A fixed input list, a dependency map, and a milestone plan are what let it travel. The rollout is part of the product.