Catawiki

A CASE STUDY

An object-first selling experience

Navigating a structural change for sellers, to set the stage for better product discovery at Catawiki.

OBS_cover1

PROJECT DURATION

YEAR

10 months

2022 - 2023

ABOUT CATAWIKI

Catawiki is an online auction marketplace for special objects and collector’s items ranging from classic cars to ancient coins. Being a curated C2C marketplace, these are the three main actors in the journey connecting supply to demand.

Frame-52
Frame-53
Frame-54

CONTEXT

In 2022, Catawiki’s focus was to set up and optimise the product for scalability. The 5 year ambition for the company was to grow their seller base four times. More sellers means more objects on our platform which means more inventory for sale. To present the large volume of objects to buyers and facilitate personalisation and easy discovery, the first step was to move from unstructured to structured object data.

Simply put, if a seller uploads a vase for sale, all of its attributes (eg; material, condition, size) are stored systematically making it easy for a buyer to search and sort the data while browsing. Here’s an example of a vase with structured (left) vs unstructured data (right).

Vase

WHY THE CHANGE

For buyers, this change would improve discoverability of objects, helping them find the object they’re looking for.

For sellers, it would enable object listing with more accuracy. It was crucial to not disrupt their current workflow while doing so.

For Catawiki, it meant that experts would be able to compile and publish new auctions with more flexibility. Having structured data would build grounds for automating some of their process. While I also led the design for the expert experience, this case study focuses on the seller experience.

THE CHALLENGE

To adapt the object listing experience for sellers to the new data structure, while maintaining key user and business metrics.

MY ROLE

I joined the team leading this project as their UX designer during the project's final 9 months and saw it through till the end.  I conceptualised and led the design for the object first listing experience, collaborating closely with the previous designer, a UX researcher, a writer and two project managers.

PROBLEMS WITH THE EXISTING DATA STRUCTURE

We were requesting irrelevant data from sellers

The data structure at Catawiki was based on merchandising categories. As a result, when uploading their object, sellers would choose the category appropriate for their object eg, Antiques, Jewellery, Art etc. Once the category was determined, the seller would be asked to fill out a specific set of details about their object relevant to that category. Imagine selling a teapot in the antiques category. To cater to all the possible objects that could fall under antiques, the set of details requested was quite broad and generic. This didn’t allow us to gather very specific data about the teapot that could eventually help buyers during search.


Sellers knew their object, not how it should be categorised

We knew from user research that the object listing flow didn’t match the mental model of sellers who thought of the object before the category their object best fits in. Figuring out the category was secondary and sellers were often confused about what to choose. Determining the object category was important to Catawiki for assigning the object to the right expert but not always relevant for the seller.

FROM CATEGORIES TO OBJECTS

To enable structured object data, we needed our data to be centred around objects rather than it's categories. The year before, we had updated our product taxonomy tool to have objects instead of categories. This allowed us to collect and store data for every object.

Frame-58

THE OBJECT FIRST SELLING EXPERIENCE

Tell us what you're selling and we do the rest.

Instead of a seller having to figure out which category to merchandise their object in, Catawiki would work their magic ensuring just the relevant information is collected from the seller.

OBS_idnetification1

Identifying an object

Sellers could identify the broader object first and even narrow down to a more specific type of object if needed. This reduced cognitive load on sellers and enabled them to move quickly by selecting the closest appropriate object. The second step to narrow down their selection was optional, for those looking to be really specific. This way we catered to the range of seller types we have as users and kept in line with our design principle about flexibility.

OBS_category

Implicit category selection

Research showed that new sellers seldom cared about choosing a category, whereas pro sellers were particular about where on the platform their object was displayed. Our second design principle - intelligence, was about letting the platform do the heavy lifting. We automatically assigned a category to the object a seller was submitting with the flexibility to change it if needed.

Cascading

A smarter form to gather exact information

We started with a broad details like era and period to classify objects (eg as antiques, modern, prehistoric etc) and determine the details that could be relevant in each case.

HOW DID WE GET THERE?

Imagining object submission as a real world conversation

In the early stages of the project, I imagined the current experience as a conversation between a seller and Catawiki to understand it’s pain points and potential improvement opportunities. I then sketched out a conversational future state which helped us define an experience goal. Object submission should feel assisted and relevant to a seller's profile. Here’s a snippet of the exercise.

CALLS-Object-centered-flow-3

Looking forward and working backwards

From 2020 to 2023, the object listing experience had evolved to be as described in this case. Through all the iterations, intelligence and automation were a common thread. We designed the future of the object listing experience to be smart, in which Catawiki assists the seller with least effort from their end. We ensured that the object first designs were a step in that direction.

Frame-59
Hack

The vision - to use image recognition and prefill all the details

Designing a first iteration of the object first experience and learning from it’s failure

We knew the object was key but categories weren’t unimportant. Pro Catawiki sellers needed to see where their object was going to be displayed and wanted agency on the decision. Additionally, the category determined some parts of the experience such as the number of photos a seller needed to upload. The first iteration of object search included the category name in the search.

Sketches

This was A/B tested with the old category based search and fared poorly. More than 20% searches produced zero results. For Germany this number was even higher, around 37%. Hot jar analysis confirmed the experience was even more confusing and time consuming than before. The CES score dropped by 7 percentage points.

Diving deeper we found

The object to category mapping wasn’t straight forward. A single object e.g. a painting could belong to about 17 different categories. The permutations and combinations were never ending and the search result list was massive. This led to one of two, either a more time consuming experience scrolling through the list, or a very narrow range of results if we chose to show only the top 5-7. Hot jar analysis showed that sellers tried different entries to find what they’re looking for, eventually choosing the closest alternative. Automatically linking an object to a category forced sellers to consider if the pairing suits what they are submitting and added more cognitive load.

Collaborating with Catawiki experts and a data scientist to design a second iteration

I revisited one of the explorations and developed it further to be a two step approach. First selecting the object and then an optional category selection. I worked with in-house experts and a data scientist to map each object to a ranked list of categories, highest rank being the most probable category for that object. We showed the top 5 categories probable for each object with the highest ranked pre-selected. Sellers could change this selection or expand the list to view the lower ranked categories.

Final-d

Beta testing to uncover critical issues

We opened the new flow to 7 large volume sellers for a week and interviewed them after. The most critical issue we found was about title generation of object based submissions. Previously titles were generated per category an object was sold in. The same object sold in different categories emphasised different details in the title. Eg, a figure sold as a decorative object (left) vs an archeological one (right) would highlight the artist vs the place of origin, culture or dynasty.

Figure_title-1

With the new flow, each object had a single title irrespective of the category. This was problematic for buyers because of the need to differentiate between objects. Sellers who beta tested the flow were rightly so, very concerned about this impacting discoverability. To solve this we utilised the mapping created during the second iteration to create conditional titles for the most problematic categories so that titles don’t appear all the same.

Helping sellers be more specific during object selection

During beta testing we observed that sellers would manipulate the titles to be more detailed about the object being submitted. To avoid this we added an additional optional step to help the seller narrow down their selection and be more specific incase they didn’t do so during search. Eg selecting an engagement ring instead of a ring.

ROLL OUT

An incremental release

After fixing the most critical issues, we started rolling out to all sellers in an incremental AB test. The first 20% of traffic was AB tested for 3 weeks. As the results were stable, we increased the traffic by 20% every week closely monitoring till all objects being listed were going through the new flow.

IMPACT

Stable business metrics and improved customer effort score (CES) for new sellers

This was a massive structural change and the goal was to keep the main business metrics and the CES for the listing flow stable through it. 

Objects in auction, Objects in auction and Objects sold with a reserve price were stable for both variants.
The CES for new sellers improved by 2 points during the test.

REFLECTING BACK

My top takeaway
Understanding the domino effect of an interconnected project such as this and anticipating ahead of time. Mapping a system of teams involved to understand how the smallest change affects the entire chain was valuable for moving in sync and pulling in the right people at appropriate moments.

One thing I would do differently
Looking back, we focused heavily on making the new seller experience smoother as a result negatively impacting existing sellers who were set in their ways. While we made efforts to inform existing sellers ahead of time via email I would consider educating them in more interactive manner in the object listing form itself.