AB Tasty
Product Manager · 2023 – 2025
+15%
Activation rate
Context
AB Tasty, an Experience Optimization SaaS platform (A/B Testing, Personalization). PM in charge of "Product eXperience", within a team of 3 developers, 1 tech lead, 1 QA and 1 Product Designer.
Challenge
A 15% churn rate and CSM support overloaded with requests, while the product offered no in-app onboarding help.
My hypothesis
An activation problem rather than a product-fit one. Checked before investing, by comparing our numbers to the Userpilot benchmark: clearly below par.
What I ruled out
Tackling churn as a whole (ICP to revisit, tool accessibility...): I went for the gap right at the start of the funnel instead, where the potential gain was highest.
A quick onboarding fix: it would have treated the symptom and left nothing behind for what came next.
Action
I ran interviews with freshly onboarded users: fear of getting started despite CSM support, and a setup that took too long, came up again and again. Digging further, I noticed our users covered very different profiles (developers, marketers, analysts, product people), with different skills and expectations.
So I designed a qualification engine at sign-up (role, A/B testing level, technical skills, main goal), framed as a service to the user rather than a plain form to maximize response rates — serving a double purpose:
- Ship a personalized onboarding: a different journey per profile, rather than one guided tour for everyone.
- Feed user segmentation for other product needs.
Result
On top of the +15% activation already shown above: -20% Time-to-Value (the delay before a user gets first real value from the product).
Lesson
The qualification engine was an obvious move in hindsight. What I'd do earlier next time: map out available cohorts/segmentations from day one (and spot the gaps) — the sooner that breakdown exists, the easier continuous product improvement becomes afterward.
Everysens
Product Manager · 2022 – 2023
-50%
Entry time
Context
Everysens, a SaaS platform for managing and tracking rail freight transport, aimed at shifting freight from road to rail and decarbonizing transport. PM in charge of the "Execution" module (transport entry, waybill validation, real-time tracking), within a team of 4 developers and 1 QA, later joined by 1 Product Designer.
Challenge
The legacy, aging "Execution" module didn't talk to the newer "Planning" module either, even though it already held part of the information needed (route, wagons, goods).
On the entry side, filling in the full detail of a train (number, goods, weight, seals for every wagon) to produce waybills and enable tracking was a nightmare from 30 wagons up, on a UI clearly not up to the task:
- Only 6 wagons shown at a time.
- No quick navigation between fields.
- No spreadsheet import.
My hypothesis
Entry time wasn't only about the interface: part of the information being asked for already lived in the "Planning" module. Make the two talk to each other and a good share of the form could fill itself in.
What I ruled out
Patching the existing interface as we went, with no migration plan: faster to start, impossible to sustain, and far less reassuring for a leadership team that had to commit to an architecture rework.
Action
I built and pitched a migration plan for the Execution module toward a new architecture to leadership, delivered incrementally with a first MVP shipped in under 2 months. This new architecture let the module talk to "Planning" to auto-fill the form from information already entered elsewhere (route, wagon count, goods).
In parallel, I personally led user research and UI/UX design (Hotjar, Heap) — before a Product Designer joined — to redesign a form this time built for trains of 30+ wagons.
Result
On top of the -50% entry time already shown above: +20% user satisfaction measured after the overhaul.
Lesson
A well-structured migration plan reassures as much as it organizes. The incremental approach let us migrate clients feature by feature, based on what was already ready — tracking that on the fly would have been far harder to keep up with, and far less reassuring for leadership.
SNCF Connect & Tech
Product Manager · 2021 – 2022
+100%
Conversion rate, ticket retrieval
Context
SNCF Connect & Tech, the Prêt à Voyager health-pass platform (pretavoyager.sncf.com, now decommissioned) — up to 40,000 visitors a day in the middle of the Covid-19 crisis, letting travelers self-check their health pass and ticket. PM within a team of 4 developers and 1 Engineering Manager, Kanban methodology, on a platform initially built externally.
Challenge
On this platform, I was working with two hard constraints:
- No product data available at all — the platform had been built externally, so all we had were visitor counts and the health-pass validation rate.
- Classic user interviews were off the table: in a station, in the middle of an epidemic, they'd have created an avoidable crowding risk.
My hypothesis
The blocker was the first step itself: the tool required scanning the ticket's QR code, while a large share of travelers only had their ticket on mobile — nothing to scan.
What I ruled out
Classic user interviews: in a station, in the middle of an epidemic, they would have created a health risk. The signal had to be contact-free — hence the tweets.
Action
I set up a daily automated extraction of tweets mentioning the project, Covid, and SNCF, to get a user signal without physical contact.
It confirmed the hypothesis. I added a second retrieval method, by name and booking reference, as an alternative to the QR code.
The conversion rate of that first step doubled.
Lesson
Look at the whole chain, not just your own product scope: only 10% of travelers ever made it to the tool, a far bigger bottleneck than anything achievable on the product side. Acquisition was outside our control here (legal constraints between carrier and distributor) — but it's a reminder that the earliest steps of a funnel weigh the most on its overall outcome.
AB Tasty
Senior Growth Product Manager · 2025 – present
From Sales-Led to a hybrid model
Shipped to production behind a feature flag, ready to launch — put on hold after AB Tasty's merger with VWO.
Context
AB Tasty, a SaaS Experience Optimization platform, historically sold 100% Sales-Led. Senior Growth Product Manager within the Growth pillar, with 4 developers, 1 tech lead, 1 QA and 1 Product Designer.
Challenge
The enterprise market is starting to saturate. What growth is left sits with small companies — the ones the offer cannot serve while staying profitable.
In a sales-led motion, a customer is expensive to acquire, and it takes over a year to earn that cost back. It sets a price floor, and the floor stays too high for a small company.
- The few small companies that did sign could not afford the support that makes the product valuable — and churned all the more.
- So the ICP ruled that segment out on its own, without anyone having to decide it.
My hypothesis
To open that segment, the cost to serve had to come down before the list price. With no sales and no CSM on those customers, an affordable subscription becomes profitable again.
What I ruled out
Cutting prices within the existing model: the floor came from the cost of acquisition, not from the price list. Going below it meant selling at a loss.
Pushing harder on enterprise: that is exactly the segment that is saturating.
Action
I built the business case and roadmap for the hybrid model and carried them to leadership: keep the sales-led motion for enterprise accounts, which have a procurement department and review the subscription as a spend anyway, and open a self-serve path for everyone else.
I designed and built that funnel end to end: public sign-up, limited freemium, and the integration of the Hyperline billing platform, which I scoped (payments, subscriptions, feature access by plan). I led 2 other Product Managers on the workstream: splitting the plan, scoping trade-offs, overall consistency.
Two bridges were planned from the start:
- From self-serve to sales-led, when usage volume or the need for advanced features justifies it.
- The free month doubles as a demo account for enterprise prospects — something the historical model made cumbersome to set up.
Lesson
The price floor came from the cost to serve, not from the price list: as long as a customer required a sales cycle and hands-on support, no subscription could come down to a small company's level. The business case convinced leadership by showing what that cost becomes once sales and CSMs are taken out of a segment — not by restating the size of the market, which they already knew.