Problem
Problem
Solution
The process
Problem #1: Sellers used the title for marketing, not for product identification
The first step of the funnel asked sellers to write a title for their listing. We then used that title to predict which product they were trying to sell.
However, sellers often wrote titles aimed at attracting buyers rather than accurately describing the product. As a result, the information was too vague or misleading for us to reliably match it to a product in the catalog.
Problem #2: Manual category selection created unnecessary friction and errors
After the seller entered the title, we used it to predict possible product categories. But instead of auto-selecting the most likely match, we always asked the seller to manually choose a category.
This added an extra step to the flow—even when our prediction was correct—and introduced room for error. If the seller chose the wrong category, we couldn’t match the listing to the correct product in the catalog.
Problem #3: Sellers frequently entered incorrect attribute data
After choosing a category, sellers had to manually complete product attributes. This led to frequent mistakes — for example:
Confusing RAM with storage capacity
Entering generic colours like "Black" instead of the official name, like "Space Grey"
These errors often prevented us from correctly matching the product with our catalog.
Enter listing title
Manually select category
Manually fill in product attributes
System tries to match listing to catalog
Search product in catalog
Select matching product from results
Confirm or edit suggested attributes
Listing created with correct catalog match
List vs. grid
We chose a list view for better readability when many results appear.
With or without filters?
To keep the flow fast and simple, we removed filters at this stage.
⚠️ Problem: This approach wouldn’t scale, as not all categories use colour as the main identifier.
⚠️ Problem: I had hidden the search box to save space, but this wasn’t a good decision, as users might need to refine their search.
⚠️ Problem: These two actions were visually different, even though they served a similar purpose—opting out of the suggested results.
⚠️ Problem: Viewing full details made the task more complex, so I needed a simpler way to provide more information.
Working closely with the development team from the start helped us understand technical constraints and adapt the design to how the search logic actually worked. This early collaboration made the implementation smoother and avoided major rework later.
Designing for the ideal flow is easy, but real value came from supporting less straightforward scenarios—like when no product is found or multiple categories are suggested. Making sure the experience stayed clear and useful in all seven use cases was essential for the success of the solution.
User testing helped uncover key points of confusion, like the unclear exit options when no product matched. Observing how users behaved in the flow allowed us to simplify the experience and make better design decisions.
Back to top