Why it mattered
The company was working on the trial experience, and templates were supposed to be the fastest way in. A new user who starts from a good template gets to a working page quickly. A new user who starts from a blank canvas has to make every decision from scratch, in a tool they're still learning.
Nine out of ten pages were starting from blank.
A wall, not a library
I went through the new-user flow myself. It went the same way every time: open the templates page, see around 80 thumbnails with no visible order, scroll, scroll again, find nothing that fits, click "blank page".
The templates themselves were fine. The system around them didn't help anyone choose:
- No way to narrow down. You couldn't ask for "e-commerce pages in a dark style". You had to scan until something looked right.
- No grouping people would recognise. Templates were organised by internal names, not by industry or goal.
- Too many look-alikes. Years of additions had produced a lot of similar layouts, and some looked dated. The weaker ones made people doubt the good ones.
People chose the blank canvas because it was the only option they could make sense of.
Structure before visuals
I started with how templates are organised and found, and left thumbnails and polish for later. With the PM and engineering lead, I defined five filters people could combine:
Filters are where catalogs quietly fail, so I asked the team to write down two rules:
Within one filter, show any match
Across filters, show only what matches all
At the same time, I worked with the Professional Services team, who build pages for customers and know what they actually use. We cut the old catalog down, then added new templates that looked clearly different from each other. Good filters on top of a stale catalog would have fixed only half the problem.
What I pushed back on
- Search in the first release. The infrastructure wasn't ready, and I also wanted the filters to work on their own first. If people found templates through search, we would never know whether the filters did their job.
- Sorting by popularity. It would keep pushing the same few templates and bury the rest.
- Deeper categories right away. More levels meant more upkeep before we knew the basic structure worked.
- Keeping every old template. A big catalog of uneven quality makes people trust all of it less.
One risk we accepted: with 294 templates, combining several filters could sometimes return nothing. Growing the catalog was planned as the next step.
The library, rebuilt
Five filters
Multi-select within each filter, in a panel beside the grid.
Filters you can see
Selected filters show as chips you can remove one by one, plus Clear all. They stay applied when you come back.
Preview before committing
A full preview of the template before it opens in the editor.
Behind the UI, the whole catalog was tagged so the content team could reorder and update it without engineering.
Templates became the way in
The recording sample is small, but it points the same way as the analytics.
What I'd do differently
- Tag the catalog before cutting it. We did both at once, which made deciding what to remove harder than it needed to be.
- Test the five filters before building them. A quick card sort and prototype with five users would have given me more confidence before engineering started.
- Make the empty state useful. When filters return nothing, the page should suggest which filter to remove instead of just saying "no results".