Second-hand luxury fashion marketplace

I worked 4+ years on Vestiaire Collective, a marketplace with 20M active users.

Collaborated with 2 product squads (Engagement & CRM, Seller) to improve the current user experience within the scopes, do user interviews and testing, explore and design new feature opportunities, help with A/B test analysis.

Vestiaire Collective wordmark

Designing a photo-first, AI-assisted listing experience

Helping sellers create listings faster by using their photos to pre-fill the listing, while keeping them in control of the final information.

Role
Product Designer
Team
Big squad of 1 PM, 1 EM, 1 Data, 2 devs per platform, 4 backends
Timeline
2-3 months to launch, followed by 3 months of iteration
Platform
iOS, Android and Web
Scope
Discovery, UX strategy, interaction design, prototyping, testing, launch and post-launch improvements
Full case study

The public version leaves out the discovery numbers, the competitor benchmark, the decision rationale and the exact post-launch results.

Context

Context

Creating a listing required sellers to move through a long, multi-step experience and manually enter important information about their item. The process was important for marketplace quality and discoverability, but it was also repetitive and difficult to correct when something was wrong.

At the same time, sellers were already preparing their items through photos before entering the listing flow. This created an opportunity to explore whether photos could become the starting point for a faster, more assisted experience.

What if sellers could start with their photos and let Vestiaire do more of the repetitive work?

The opportunity

Discovery

The discovery combined existing user research, seller feedback, quantitative listing data, a UX audit of the current flow and a benchmark of competitor listing experiences.

The research highlighted recurring friction around the length of the listing process, finding the right category or model, managing photos, writing descriptions, understanding pricing and correcting information after submission.

  • The process felt long and complex, especially when sellers had to complete many fields manually.

  • Taxonomy and model selection did not always reflect how sellers described their items.

  • Sellers needed more flexibility to review and correct their information.

  • Incomplete descriptions with limited support and missing details could reduce buyer confidence.

  • Rigid flow, with too many clicks and an outdated UX.

Out of scope

The discovery also surfaced several pain points that were important to sellers but fell outside the scope of this revamp. These were kept in the roadmap or assigned to the relevant teams.

Editing after submission

Sellers still couldn't edit a listing once submitted. This remained on the roadmap, with the added consideration of the recropping and revalidation costs triggered by changes.

Price recommendations

Some sellers felt recommended prices were too low. This was linked to the existing pricing algorithm and therefore remained outside the scope of the listing experience.

Curation and rejection

Sellers reported long validation times and limited clarity around rejection reasons. These issues were primarily related to curation operations and were therefore addressed separately by the OPS team.

Taxonomy revamp

The current taxonomy remains unchanged for now due to its complexity, but it remains an area for future improvement.

How might we use AI to reduce the effort of creating a listing without taking control away from the seller?

The challenge was not simply to add AI to the existing form. It was to create an experience that balanced speed, accuracy, clarity and seller control.

Defining the experience

The new experience was structured around a photo-first flow.

  1. 1Photo selectionPhotos became the starting point of the experience instead of the final step of a form.
  2. 2AI analysisThe system prepares information such as category, brand, model, colour, material, size, condition and description.
  3. 3Recap and editThe seller reviews everything in one place and edits anything the AI did not get right.
  4. 4PublishThe listing goes live once the seller confirms the information.

The AI could help prepare information such as category, brand, model, colour, material, size, condition, description. The seller would still review and edit the information before publishing.

Design principles

  • Start with photos

    Photos became the starting point of the experience instead of the final step of a form.

  • Let AI do the repetitive work

    The system could analyse the photos and prepare information that sellers previously had to enter manually.

  • Keep the seller in control

    AI-generated values were presented as information to review and edit, not as irreversible decisions.

  • Design for uncertainty

    When the AI could not infer mandatory information, the seller could complete the missing information manually.

  • Keep the MVP focused

    The first release focused on photo analysis, prefilled information and seller review rather than redesigning the entire listing ecosystem.

Prototyping and validation

I created an interactive prototype in Figma Make to test the experience as a complete flow rather than as a collection of individual screens.

We explored whether sellers understood what Smart Listing was doing, whether the generated information felt credible and whether they knew where they could make changes. Testing helped us refine the structure and hierarchy of the recap screen, which became the central place for reviewing the generated listing.

Before and after

A view per platform

Before Manual listing Step after step, with the seller typing or selecting every attribute.
After Smart Listing Photos, a short analysis, then the recap with everything already in place.

Timeline

  1. 2-3 months Discovery to launch

    The project moved from discovery to launch in approximately 2-3 months.

  2. Progressive Rollout

    The MVP was designed for progressive rollout, first with internal and beta testers group of users, allowing the team to test the experience, gradually expose it to sellers and learn from real usage.

  3. 3 months After launch

    Launch was the beginning of the next learning cycle. For the following three months, we continued to monitor the experience, collect feedback and identify opportunities for improvement.

A short contextual feedback module at the end of the flow helped us gather feedback while the experience was still fresh for sellers. The post-launch work focused on fixing issues found in real usage, improving the generated description, improving accuracy of prefilled information and handling edge cases.

Impact

Impact

The project was followed through four product metrics. We decreased Time to List and increased Completion Rate and Listings per Seller, which shows the listing form became easier.

  • Lower Median Time to List

    Whether the new flow reduced the time required to create a listing.

  • Higher Completion Rate

    Whether sellers were able to complete the new experience.

  • Higher Listings per Seller

    Whether the experience supported additional marketplace supply.

Measured against the manual listing flow as baseline. The exact figures, and the fourth metric, are in the full version.