Curate automatically
The old library went stale because every video was uploaded by hand. Now videos are added automatically from internal ratings and client picks, and AI tags them. People correct tags instead of writing them.
← Selected workCase study · 6 min
Vidsy · 2024
In 2024, Vidsy’s creative leads had an inspo library that everyone opened and nobody trusted. It was slow, half-tagged and curated by hand by one very patient person. I rebuilt it around how leads actually search, and handed the tagging to AI so it never goes stale again.
Vidsy
The redesigned inspiration library
01 · The world
Vidsy makes video ads for brands like Pepsi, L'Oréal and Google through a network of independent creators. A creative lead turns each client brief into something creators can shoot.
Reference videos are how everyone pictures the same ad. Show the client three and they relax. Show the creator the same three and they know what "playful but premium" means. So every brief starts with a hunt for references, and that hunt was where time went to die.
02 · What was broken
It's great in principle, but it can be a pain to use.Creative lead, interview
Survey · where leads look first
Rank-ordered, anonymous survey of creative leads. Bars show relative weight. The library was the first choice because it used our frameworks, but slow loading and weak search sent leads out to trawl TikTok and Pinterest for the rest.
03 · The bet
Before designing anything I wanted a number that made the project worth doing. This is the arithmetic that got it funded, laid out the way I presented it.
Time spent hunting for inspiration
~150 hrs/moThat was about a quarter of the time creative leads spent on concepts and writing briefs. It adds up to roughly one full-time lead, and a lead generates around £833k of value a year.
Target reduction
30%A conservative goal, based on what shadowing (watching leads search for references) suggested a working search would save.
Value of the time freed up
~£250kA year, if leads spent the saved time on briefs instead of hunting for references. The exact figure on the spreadsheet was £249,900.
I rounded this projection down on purpose. Section 06 shows what actually happened.
04 · How I found out
Each one answered a specific question. For each, I’ve written down what it told me.
I looked at what people searched for and which filters they used. This told me which of our categories people really used and which only existed on paper.
When the library let them down, leads asked each other for examples in Slack. I coded months of these requests, sorting each one into a category. Those categories became the set of tags every video is labeled with.
I interviewed six creative leads and strategists. People didn't dislike the library. They didn't trust it, which is a different problem to solve.
I watched people look for references as they worked. Most started outside our library, on the TikTok ad library, Twirl or MagicBrief. Those became the competitors to analyze.
I compared platform ad libraries like TikTok’s with build-your-own tools. None combined our frameworks with saving and sharing, and that was the gap.
I ran a workshop on how the library should be curated, and by whom. The answer was automatically, using internal ratings and the videos clients chose. Otherwise it would go stale again.
Google Analytics · what people searched for
Event counts from the old library. Free-text searches (403 of them) asked for things no filter covered: video moments like "reveal" and "countdown", visual styles like green screen and split screen, audiences like Gen Z, and editing techniques like match cuts.
05 · What I decided
I explored a trends view and an immersive player (a full-page video view with boards and comments) along the way. Some of that shipped, some was cut. What follows is what made it in, and why.
Prioritization
The old library went stale because every video was uploaded by hand. Now videos are added automatically from internal ratings and client picks, and AI tags them. People correct tags instead of writing them.
Picking two filters in the same category used to show only videos that matched both, which was usually none. Now it shows videos that match either. Clicking a video's title jumps to the original project.
Leads loved sites that explain why an ad works. A trends tab shows which frameworks are rising, so a lead can start from "what's working" instead of a blank search.
Unglamorous, and the biggest single win: hover to play, one-click download or copy link, and no separate log-in.
The last change before launch was taking color out of the video grid and using it only on the filters. People commented on it more than anything else. Nobody said "it's calmer." They said "I can find things."
06 · What happened
I set three measures of success before designing anything. Did people keep using the library, according to analytics? Did a usability survey score better after launch than before? Did requests for examples in Slack drop? All three moved.
What it was worth
£133,280Leads spent 16% less time hunting for inspiration. Across a year, that's £133,280 of time back for briefs. It fell short of the 30% target, and I said so.
Findability · survey, before → after
63% → 92%satisfied with how easy videos were to find
Overall experience · survey, before → after
74% → 96%satisfied with the library overall, in the same survey
Time spent finding references
−16%measured by time tracking and Slack monitoring
Occasionally the AI tagging is wrong but I'm still able to find something that works, a lot faster than before.Creative lead
I love that I'm able to find content that I've never seen before.Strategist
The survey ran a few weeks after launch, with the same questions as the one before. It had 24 respondents. Leads reported their own time, and Slack requests were counted weekly.
07 · Differently
I'd ship the boring half first. Playback, access and better search drove most of the gain, with no AI needed.
I'd also have set the 30% target lower in public and higher in private. We hit 16%, which is a real result. But the people I reported to heard "half of what she promised" before they heard "£133k."