My Framer Site

Starting with the object

Designing the seller's listing experience to support marketplace growth

Starting with the object

Designing the seller's listing experience to support marketplace growth

Starting with the object

Designing the seller's listing experience to support marketplace growth

Employer

Catawiki

Areas

Design, Systems thinking

Platform

Mobile web

Year

2024

Employer

Catawiki

Areas

Design, Systems thinking

Platform

Mobile web

Year

2024

Employer

Catawiki

Areas

Design, Systems thinking

Platform

Mobile web

Year

2024

Background

Catawiki is a curated online auction marketplace for unique objects, from classic cars to rare coins. In 2022, the company set a bold ambition: quadruple the seller base over five years.

Scaling sellers meant scaling inventory, but scaling inventory required better data. At the time, object data was largely unstructured and category-driven. To improve buyer discoverability and future automation, we needed to shift to a structured, object-centered data model.

Background

Catawiki is a curated online auction marketplace for unique objects, from classic cars to rare coins. In 2022, the company set a bold ambition: quadruple the seller base over five years.

Scaling sellers meant scaling inventory, but scaling inventory required better data. At the time, object data was largely unstructured and category-driven. To improve buyer discoverability and future automation, we needed to shift to a structured, object-centered data model.

Background

Catawiki is a curated online auction marketplace for unique objects, from classic cars to rare coins. In 2022, the company set a bold ambition: quadruple the seller base over five years.

Scaling sellers meant scaling inventory, but scaling inventory required better data. At the time, object data was largely unstructured and category-driven. To improve buyer discoverability and future automation, we needed to shift to a structured, object-centered data model.

Who the change affected?

The change wasn’t isolated to one surface. It affected three interconnected systems at Catawiki:

Who the change affected?

The change wasn’t isolated to one surface. It affected three interconnected systems at Catawiki:

Who the change affected?

The change wasn’t isolated to one surface. It affected three interconnected systems at Catawiki:

1

Experts - who curate objects and move them between auctions

2

Commercial teams - who structure marketplace strategy

3

Sellers - who feed the entire system with inventory

1

Experts - who curate objects and move them between auctions

2

Commercial teams - who structure marketplace strategy

3

Sellers - who feed the entire system with inventory

1

Experts - who curate objects and move them between auctions

2

Commercial teams - who structure marketplace strategy

3

Sellers - who feed the entire system with inventory

The challenge

Adapt the seller listing flow to a new structured data system while, maintaining key business metrics, preserving pro seller efficiency and minimizing confusion during the transition.

More than a visual re-deisgn, this was a foundational shift in how data, UX, and marketplace logic worked together.

The challenge

Adapt the seller listing flow to a new structured data system while, maintaining key business metrics, preserving pro seller efficiency and minimizing confusion during the transition.

More than a visual re-deisgn, this was a foundational shift in how data, UX, and marketplace logic worked together.

The challenge

Adapt the seller listing flow to a new structured data system while, maintaining key business metrics, preserving pro seller efficiency and minimizing confusion during the transition.

More than a visual re-deisgn, this was a foundational shift in how data, UX, and marketplace logic worked together.

My role

I led the design across the commercial and expert tools, and ultimately the seller-facing experience as well. This case focuses on the seller submission experience to support that transformation. To read about my work on the commercial and expert tooling, read my case study about Managing a taxonomy tool.

My role

I led the design across the commercial and expert tools, and ultimately the seller-facing experience as well. This case focuses on the seller submission experience to support that transformation. To read about my work on the commercial and expert tooling, read my case study about Managing a taxonomy tool.

My role

I led the design across the commercial and expert tools, and ultimately the seller-facing experience as well. This case focuses on the seller submission experience to support that transformation. To read about my work on the commercial and expert tooling, read my case study about Managing a taxonomy tool.

Pain points

Research on the perception of the existing flow surfaced two problems:

1

Irrelevant data collection

Because the structure was based on broad categories (Antiques, Jewellery, Art), sellers were shown generic forms designed to cover every possible object in that category. Example, a teapot listed under “Antiques” would display a broad, unfocused set of fields instead of relevant ones that improve buyer search. We were collecting data, but not always the right kind.

2

Sellers knew their object, not it's category

The listing flow didn’t match how new sellers think. Sellers knew the object they were selling, but were forced to choose from niche categories on a specialist platform. A teapot could reasonably sit in Antiques or Culinary yet that distinction mattered for experts and search algorithms, much less for the seller. This was especially a problem for new sellers unfamiliar with the platform.

Pain points

Research on the perception of the existing flow surfaced two problems:

1

Irrelevant data collection

Because the structure was based on broad categories (Antiques, Jewellery, Art), sellers were shown generic forms designed to cover every possible object in that category. Example, a teapot listed under “Antiques” would display a broad, unfocused set of fields instead of relevant ones that improve buyer search. We were collecting data, but not always the right kind.

2

Sellers knew their object, not it's category

The listing flow didn’t match how new sellers think. Sellers knew the object they were selling, but were forced to choose from niche categories on a specialist platform. A teapot could reasonably sit in Antiques or Culinary yet that distinction mattered for experts and search algorithms, much less for the seller. This was especially a problem for new sellers unfamiliar with the platform.

Pain points

Research on the perception of the existing flow surfaced two problems:

1

Irrelevant data collection

Because the structure was based on broad categories (Antiques, Jewellery, Art), sellers were shown generic forms designed to cover every possible object in that category. Example, a teapot listed under “Antiques” would display a broad, unfocused set of fields instead of relevant ones that improve buyer search. We were collecting data, but not always the right kind.

2

Sellers knew their object, not it's category

The listing flow didn’t match how new sellers think. Sellers knew the object they were selling, but were forced to choose from niche categories on a specialist platform. A teapot could reasonably sit in Antiques or Culinary yet that distinction mattered for experts and search algorithms, much less for the seller. This was especially a problem for new sellers unfamiliar with the platform.

Reframing the problem

How might we simplify object listing for sellers while preserving the structural logic required for experts, merchandising, and search?

Reframing the problem

How might we simplify object listing for sellers while preserving the structural logic required for experts, merchandising, and search?

Reframing the problem

How might we simplify object listing for sellers while preserving the structural logic required for experts, merchandising, and search?

Designing the object-first experience

Pulling together insights from seller research, expert workflows, and search design, we designed the object-first submission experience guided by a simple principle: “Tell us what you’re selling, and we’ll do the rest.”

Designing the object-first experience

Pulling together insights from seller research, expert workflows, and search design, we designed the object-first submission experience guided by a simple principle: “Tell us what you’re selling, and we’ll do the rest.”

Designing the object-first experience

Pulling together insights from seller research, expert workflows, and search design, we designed the object-first submission experience guided by a simple principle: “Tell us what you’re selling, and we’ll do the rest.”

Flexibility for new and advanced sellers

Sellers first identified what they were selling. They could stay broad (“ring”) or narrow down (“engagement ring”). This improved precision while accommodating different levels of expertise.

Flexibility for new and advanced sellers

Sellers first identified what they were selling. They could stay broad (“ring”) or narrow down (“engagement ring”). This improved precision while accommodating different levels of expertise.

Flexibility for new and advanced sellers

Sellers first identified what they were selling. They could stay broad (“ring”) or narrow down (“engagement ring”). This improved precision while accommodating different levels of expertise.

Intelligent category assignment

I designed the listing to preselect the most probable category based on a ranking system. In collaboration with the commercial team and a data scientist, we mapped each object to a ranked list of probable categories. The highest-ranked option would be pre-selected while surfacing the the top five options. Sellers had agency without forcing heavy decision-making.

Intelligent category assignment

I designed the listing to preselect the most probable category based on a ranking system. In collaboration with the commercial team and a data scientist, we mapped each object to a ranked list of probable categories. The highest-ranked option would be pre-selected while surfacing the the top five options. Sellers had agency without forcing heavy decision-making.

Intelligent category assignment

I designed the listing to preselect the most probable category based on a ranking system. In collaboration with the commercial team and a data scientist, we mapped each object to a ranked list of probable categories. The highest-ranked option would be pre-selected while surfacing the the top five options. Sellers had agency without forcing heavy decision-making.

Smarter fields

Fields were designed to progressively reveal or hide based on the object’s period. The 'Era' field became a pivotal input, linking the object to the most relevant category while dynamically surfacing the right level of detail. This allowed us to collect precise, context-aware information without overwhelming sellers

Smarter fields

Fields were designed to progressively reveal or hide based on the object’s period. The 'Era' field became a pivotal input, linking the object to the most relevant category while dynamically surfacing the right level of detail. This allowed us to collect precise, context-aware information without overwhelming sellers

Smarter fields

Fields were designed to progressively reveal or hide based on the object’s period. The 'Era' field became a pivotal input, linking the object to the most relevant category while dynamically surfacing the right level of detail. This allowed us to collect precise, context-aware information without overwhelming sellers

Conditional logic for titles

For certain categories where buyers browse for highly specific items, like Archaeology, we adjusted how titles were generated. Instead of starting with the object type, titles pulled more distinctive attributes such as the artist, manufacturer, or title of the work.

Conditional logic for titles

For certain categories where buyers browse for highly specific items, like Archaeology, we adjusted how titles were generated. Instead of starting with the object type, titles pulled more distinctive attributes such as the artist, manufacturer, or title of the work.

Conditional logic for titles

For certain categories where buyers browse for highly specific items, like Archaeology, we adjusted how titles were generated. Instead of starting with the object type, titles pulled more distinctive attributes such as the artist, manufacturer, or title of the work.

Process

Mapping systems

To understand how object submission really worked, I mapped the systems involved in collaboration with Product and Commercial teams. The existing flow was tightly coupled with internal tooling, which limited flexibility. By visualizing how these systems interacted, we were able to identify opportunities to decouple parts of the process and redesign the submission experience more intentionally.

Mapping systems

To understand how object submission really worked, I mapped the systems involved in collaboration with Product and Commercial teams. The existing flow was tightly coupled with internal tooling, which limited flexibility. By visualizing how these systems interacted, we were able to identify opportunities to decouple parts of the process and redesign the submission experience more intentionally.

Mapping systems

To understand how object submission really worked, I mapped the systems involved in collaboration with Product and Commercial teams. The existing flow was tightly coupled with internal tooling, which limited flexibility. By visualizing how these systems interacted, we were able to identify opportunities to decouple parts of the process and redesign the submission experience more intentionally.

A simplified systems map of entities and tools involved

A simplified systems map of entities and tools involved

A simplified systems map of entities and tools involved

Exploring directions & learning from failure

We began by testing an object search that still exposed category information. We believed this would maintain transparency for pro sellers while simplifying discovery. We A/B tested it against the existing category-first flow and it performed worse. Over 20% of searches returned zero results and CES dropped by 3 points

Through deeper analysis, we uncovered that object-to-category mapping was far more complex than anticipated. A single object, like a painting, could belong to 17 categories. The result list became overwhelming. Sellers either scrolled through long lists or selected the closest approximation. Exposing category logic added cognitive load rather than reducing it. That failure was pivotal. It clarified that category complexity needed to be handled behind the scenes.

Exploring directions & learning from failure

We began by testing an object search that still exposed category information. We believed this would maintain transparency for pro sellers while simplifying discovery. We A/B tested it against the existing category-first flow and it performed worse. Over 20% of searches returned zero results and CES dropped by 3 points

Through deeper analysis, we uncovered that object-to-category mapping was far more complex than anticipated. A single object, like a painting, could belong to 17 categories. The result list became overwhelming. Sellers either scrolled through long lists or selected the closest approximation. Exposing category logic added cognitive load rather than reducing it. That failure was pivotal. It clarified that category complexity needed to be handled behind the scenes.

Exploring directions & learning from failure

We began by testing an object search that still exposed category information. We believed this would maintain transparency for pro sellers while simplifying discovery. We A/B tested it against the existing category-first flow and it performed worse. Over 20% of searches returned zero results and CES dropped by 3 points

Through deeper analysis, we uncovered that object-to-category mapping was far more complex than anticipated. A single object, like a painting, could belong to 17 categories. The result list became overwhelming. Sellers either scrolled through long lists or selected the closest approximation. Exposing category logic added cognitive load rather than reducing it. That failure was pivotal. It clarified that category complexity needed to be handled behind the scenes.

Earlier iterations of object selection

Earlier iterations of object selection

Earlier iterations of object selection

Through another round of iterations I arrived at the smart category selection. This work required deep partnership across functions. Catawiki experts helped validate category logic. Data science supported ranking accuracy. Commercial teams ensured merchandising needs were preserved.

Finding patterns across key categories

I mapped object details (specs) across three high-volume categories (Fashion, Furniture, and Art), focusing on objects that were hardest to structure. This revealed recurring patterns that helped define consistent object fields across categories. I also grouped fields into anchor details and general details, allowing us to design clearer hierarchy and display logic.

Finding patterns across key categories

I mapped object details (specs) across three high-volume categories (Fashion, Furniture, and Art), focusing on objects that were hardest to structure. This revealed recurring patterns that helped define consistent object fields across categories. I also grouped fields into anchor details and general details, allowing us to design clearer hierarchy and display logic.

Finding patterns across key categories

I mapped object details (specs) across three high-volume categories (Fashion, Furniture, and Art), focusing on objects that were hardest to structure. This revealed recurring patterns that helped define consistent object fields across categories. I also grouped fields into anchor details and general details, allowing us to design clearer hierarchy and display logic.

Object field mapping and heirarchy

Object field mapping and heirarchy

Object field mapping and heirarchy

Beta testing to surface the real risks

Once we had a working second iteration, we opened the new flow to seven high-volume sellers for one week and interviewed them after to understand real-world friction before scale. The most critical issue we uncovered wasn’t the form or the object search, it was title generation.

In the old world, titles were generated based on the category an object was listed in. The same object could be described differently depending on where it lived. For example, a figure sold as a decorative object would emphasize the maker, while the same figure sold as an archaeological object would highlight the historic period. That nuance helped buyers scan and differentiate listings.

In the new object-first flow, each object initially had one title regardless of category, which made listings feel overly similar. Pro sellers immediately flagged this as a discoverability risk and they were right.

Beta testing to surface the real risks

Once we had a working second iteration, we opened the new flow to seven high-volume sellers for one week and interviewed them after to understand real-world friction before scale. The most critical issue we uncovered wasn’t the form or the object search, it was title generation.

In the old world, titles were generated based on the category an object was listed in. The same object could be described differently depending on where it lived. For example, a figure sold as a decorative object would emphasize the maker, while the same figure sold as an archaeological object would highlight the historic period. That nuance helped buyers scan and differentiate listings.

In the new object-first flow, each object initially had one title regardless of category, which made listings feel overly similar. Pro sellers immediately flagged this as a discoverability risk and they were right.

Beta testing to surface the real risks

Once we had a working second iteration, we opened the new flow to seven high-volume sellers for one week and interviewed them after to understand real-world friction before scale. The most critical issue we uncovered wasn’t the form or the object search, it was title generation.

In the old world, titles were generated based on the category an object was listed in. The same object could be described differently depending on where it lived. For example, a figure sold as a decorative object would emphasize the maker, while the same figure sold as an archaeological object would highlight the historic period. That nuance helped buyers scan and differentiate listings.

In the new object-first flow, each object initially had one title regardless of category, which made listings feel overly similar. Pro sellers immediately flagged this as a discoverability risk and they were right.

Adding specificity without adding complexity

First, we used the object-to-category mapping we built in iteration two to introduce conditional title logic for the categories where nuance mattered most. That allowed the system to generate differentiated titles again without pushing complexity back onto sellers

Second, during the beta we noticed sellers were manually editing titles to add missing detail. That was a strong signal: the system wasn’t capturing enough specificity early on. To address this, we added an optional “narrow it down” step during object selection so instead of “ring,” sellers could choose “engagement ring,” and the rest of the flow could adapt accordingly. This kept the experience light for casual sellers while giving power sellers a path to be precise.

Adding specificity without adding complexity

First, we used the object-to-category mapping we built in iteration two to introduce conditional title logic for the categories where nuance mattered most. That allowed the system to generate differentiated titles again without pushing complexity back onto sellers

Second, during the beta we noticed sellers were manually editing titles to add missing detail. That was a strong signal: the system wasn’t capturing enough specificity early on. To address this, we added an optional “narrow it down” step during object selection so instead of “ring,” sellers could choose “engagement ring,” and the rest of the flow could adapt accordingly. This kept the experience light for casual sellers while giving power sellers a path to be precise.

Adding specificity without adding complexity

First, we used the object-to-category mapping we built in iteration two to introduce conditional title logic for the categories where nuance mattered most. That allowed the system to generate differentiated titles again without pushing complexity back onto sellers

Second, during the beta we noticed sellers were manually editing titles to add missing detail. That was a strong signal: the system wasn’t capturing enough specificity early on. To address this, we added an optional “narrow it down” step during object selection so instead of “ring,” sellers could choose “engagement ring,” and the rest of the flow could adapt accordingly. This kept the experience light for casual sellers while giving power sellers a path to be precise.

Rollout strategy

After addressing the critical issues from beta, we moved into a controlled rollout. We increased exposure by 20% each week until all seller submissions were going through the new flow. This approach let us monitor marketplace health, catch edge cases early, and scale safely without destabilizing key metrics.

Rollout strategy

After addressing the critical issues from beta, we moved into a controlled rollout. We increased exposure by 20% each week until all seller submissions were going through the new flow. This approach let us monitor marketplace health, catch edge cases early, and scale safely without destabilizing key metrics.

Rollout strategy

After addressing the critical issues from beta, we moved into a controlled rollout. We increased exposure by 20% each week until all seller submissions were going through the new flow. This approach let us monitor marketplace health, catch edge cases early, and scale safely without destabilizing key metrics.

Impact

+5 pt

Customer effort score

Stable

Object submission rate

Stable

Objects in auction rate

+5 pt

Customer effort score

Stable

Object submission rate

Stable

Objects in auction rate

+5 pt

Customer effort score

Stable

Object submission rate

Stable

Objects in auction rate

Reflection

Top learning

Looking back, the strongest part of this project was how tightly we aligned product, data, and expert teams. This wasn’t just a UX problem, it was a structural shift that touched category strategy, internal tooling, and marketplace logic. Continuously looping in those teams early helped us uncover edge cases long before they became technical blockers.

Top learning

Looking back, the strongest part of this project was how tightly we aligned product, data, and expert teams. This wasn’t just a UX problem, it was a structural shift that touched category strategy, internal tooling, and marketplace logic. Continuously looping in those teams early helped us uncover edge cases long before they became technical blockers.

Top learning

Looking back, the strongest part of this project was how tightly we aligned product, data, and expert teams. This wasn’t just a UX problem, it was a structural shift that touched category strategy, internal tooling, and marketplace logic. Continuously looping in those teams early helped us uncover edge cases long before they became technical blockers.

What I'd do differently

I would introduce lightweight prototypes for the trickiest parts (like object-to-category mapping and title generation) get qualitative signals much earlier in the process. These were the components with the largest failure potential, testing earlier than the beta roll out could have saved iteration cycles.

What I'd do differently

I would introduce lightweight prototypes for the trickiest parts (like object-to-category mapping and title generation) get qualitative signals much earlier in the process. These were the components with the largest failure potential, testing earlier than the beta roll out could have saved iteration cycles.

What I'd do differently

I would introduce lightweight prototypes for the trickiest parts (like object-to-category mapping and title generation) get qualitative signals much earlier in the process. These were the components with the largest failure potential, testing earlier than the beta roll out could have saved iteration cycles.