Carhartt WIP, 2026

Replatforming search, browse and recommendations

I own Carhartt WIP's move to a new discovery platform and its scale-up: the requirements, the vendor choice, the roadmap, the UX, and the scrum team of engineers delivering it on web and app.

Role: Product lead and product owner of the scrum team
Timeline: Requirements from February 2026, MVP in October, full integration by year end, then the remaining markets
Team: Web and app engineers, an implementation partner, the vendor and merchandising
Scale: 3.5M+ searches a month, 25 markets, 39 locales, web and app

The problem

Search and recommendations ran, but nobody could steer, test or trust them.

Ranking nobody could steer

Merchandisers couldn't see why a product ranked where it did, so every change became an engineering ticket.

No way to test

There was no A/B testing in search, merchandising or recommendations, so every change was a judgment call.

A catalog out of sync

Product data reached the old engine through a separate feed and drifted from the source.

What I did

Defined the platform before picking a vendor

I wrote 87 requirements across search, merchandising, recommendations and the platform itself, and split them into MVP and later phases. Every demo and every vendor score used them, which kept the decision about our problems rather than the best demo.

Ran delivery as a scrum team

Six two-week sprints from contract to MVP. I own the board, the priorities, the measurement strategy and the merchandising rules, and I set the go-live plan across web, app, the implementation partner and the vendor.

Fixed the data first

Attribute names now match the source catalog, new-arrival dates no longer reset overnight, and revenue tracking is in place before launch, so the new engine starts clean and results can be measured.

Designed the browse experience

Curated filters per category, six to nine per page. Merchandising rules sit on top of the ranking model, so campaigns survive model updates, and sold-out products stay visible but rank low.

Category page

Boost: new arrivalsBury: sold out

Rebuilt recommendations around business goals

I audited every existing placement and proposed a new strategy. Each placement gets one goal, a primary strategy, a fallback and guardrails. Search keeps sold-out items ranked low, while recommendations leave them out.

Recommendation pods, proposed setup

Planned the scale-up beyond the MVP

The MVP covers Germany. I own the plan after it: recommendations and a new search results page by the end of the year, then the remaining markets, and the testing setup the old platform never had.

How we'll know it worked

Vendor dashboards
Quick to read, but each engine reports differently, and the old one saw different traffic.
What we did
GA4 before and after, against a locked baseline
One source and one baseline, captured before launch. Less flattering, and much harder to dispute.

Where it stands

The MVP goes live in Germany in October. Full integration follows by year end, then the scale-up across markets.

3.5M+searches a month

87requirements across four pillars

25markets in scope

~12 weeksfrom contract to MVP

What's working

One set of requirements kept the vendor choice about our problems, not the best demo.
Fixing catalog data before the switch, so the new engine doesn't inherit old errors.

Risks I'm managing

Revenue wasn't tracked in the proof phase. Tracking is rebuilt before launch so results can be compared at all.
The success target sits close to what can be reliably detected, so it's read against a locked baseline over 16 weeks.

What I'd do differently

I'd agree the measurement design before signing, not during the build. Two tracking gaps only surfaced in implementation, and both would have shaped how we defined success.