+20%
More data collectors activated
70%
Of new users answered the sign-up questions, up from 50%
3 weeks
To ship Secure Studies, against an estimate of months
Industry
AI and SaaS
Role
Lead Product Designer
Team
3 squads, 4 product managers
A three-month contract kept being renewed, and I ended up covering design leadership too
Prolific is an AI infrastructure company. Data collectors run research on it, and participants take part in that research. I joined the self-serve growth team on a three-month contract, to get new data collectors to their first study faster. Prolific kept renewing it, then brought me back for a second contract, and I worked on 24 projects across 3 squads, with 20 of them shipped.
While Prolific had a gap in its Head of Design role, I covered design leadership. I mentored 4 designers, about one session a week, ran weekly critique, and set up working agreements with Product and Engineering to cut late changes. Filsan Adan, one of those designers, said “I learned a lot working with him. Great, great mentor.” Jaini Shah, another, said “He’s really helped actually strengthen me as a designer.” My official title was Senior, because I came in as a contractor, but I was asked to do a lead’s responsibilities.
These are three of those projects, plus the work that never shipped.
UI samples from across the projects I did at Prolific
I kept every sign-up question, moved the ones that didn’t need to come first, and 20% more data collectors activated
Half of new users skipped the questions Prolific’s teams needed, and some of the wrong people got in
Sign-up was two screens, and the verification email arrived after the first. About 50% of people clicked it and skipped the questions about who they were. Then a popup locked the whole app until they’d verified their email.
Data collectors and participants used the same sign-up too. Some participants picked Run research by mistake and couldn’t undo it, and some did it on purpose to skip the participant waitlist.
Marketing, sales and product all needed those answers, so marketing were nervous about changing sign-up. But the old flow was already losing half of them.
One question per page, easy ones first, and nothing locked
I was the only designer, working with the PM, engineers and a data analyst. I designed the flows, the sign-up concepts, the checklist and the error states, and I started by putting every idea on one board with the psychology behind it.
The new sign-up opens with one question, whether you’re here to take studies or run research. Then it asks one question per page, easy ones first so people keep going. It’s more clicks, but each page asks for one thing. The verification email now waits until people are logged in, so it can’t pull them out halfway.
Anything you don’t need to get in moved to a Getting started checklist. You can explore everything straight away, and the only thing an unverified email stops is publishing a study. I worried a forced checklist would put people off, so it nudges and never blocks, and unfinished steps stay visible so people come back.
Five bigger ideas didn’t make it, each for its own reason
A flow for each persona meant far more paths to build and test, so one first question does the sorting.
Popups over a blurred app still lock people out.
Magic links were too much engineering for the first release.
Auto-filling from your work email felt creepy to pilot users and only matched about 40% of the time. We tested it later as its own A/B experiment, and it didn’t work out.
A Getting started page that changed at each stage was messy to build and our data couldn’t make the business case, so we saved it for later.
When the email doesn’t arrive, the checklist shows you why and lets you fix it
The checklist shows which address we used and suggests checking junk, and you can resend it or fix the address from there. Sign-up already turned away personal addresses like Gmail to stop abuse, so that rule isn’t mine. Neither is the whole of the Welcome Study, the pre-built survey on the explore cards, which I joined late to finish off.
20% more activated, and what’s live is almost the same as the design
We shipped it in steps so we could A/B test each part, and the data analyst on our team measured the results. Over 8 weeks in Metabase, the changes to onboarding got 20% more data collectors to activate. Moving the verification email took answers from 50% to 70%.
Today it’s almost the same as the design, with brand refreshes on top, and Prolific later reused the patterns for participant sign-up. The next step was a new Home that changed once Getting started was done, and that was dropped.
Google and OpenAI needed their studies locked down, and I cut months of new screens down to one toggle
Engineers estimated months, for a problem that needed fixing fast
Prolific passed participant IDs to the researcher’s study tool as plain URL parameters. Some of its biggest customers, including Google and OpenAI, needed that locked down fast, with few changes to the integration API. The plan had new screens for researchers, admins and participants.
I scoped it down to what we could test first
The integrations team didn’t have a designer, so I joined them, working with their PM and engineers. I reframed the work as test first, then invest, so we’d use what we already had and only build more once people used it. There was pushback, because the team had spent time on the bigger plan. It won them round, and it left engineers free to work on the security.
My first idea added signed URLs as a fourth option and locked the other three, and it needed extra explaining. Signed URLs send the same IDs, so they belonged where researchers already set those up. The version we shipped is one toggle inside the URL parameters setting, off by default, so nothing changes for researchers who don’t need it. Organisations can enforce it, and then the toggle stays on and says why. It was also my idea to never show participants the URL. Prolific only creates the token when they click Start study, and the study tool checks it before letting them in.
We didn’t build any new pages, flows or custom controls. We also didn’t design a screen for a failed check, because these customers had a direct line to the integrations team and problems went straight to them. The help centre had nothing explaining URL parameters, so every state links to the integration guide.
It shipped in 3 weeks, and uptake started small
We shipped it in 3 weeks instead of months, and the only change from the design was how the switch looked in Storybook. At first only one of the customers who asked for it used it, now and then, and uptake has grown since. Starting small meant Prolific didn’t spend months on it before the demand was there.
I took Participant Finder, an AI matching tool, from a blank page to one Prolific’s services team uses
Only study setup could tell you if Prolific had the right people
“Do they have the participants I need?” is one of the top three questions new researchers ask. Before Participant Finder, only study setup could answer it, and the answer sat near the bottom of the form. Prolific’s services team help customers find the people they need for research, and the tool was built mainly for them. It’ll also ship to researchers on the platform.
The decision I’m most proud of is that it works alone and drops into study setup later
We gave it a page of its own, with a card on the dashboard. You get a live count of who matches as you choose. The most-used screeners come first, and premium ones like verified domain experts stay visible because they earn Prolific money.
I designed it to work on its own page and to drop into study setup later without a redesign, and that shaped every layout and component in it.
In the second version, describing who you need became the first thing on the page. AI picks the screeners, then shows how many people match and what to pay, ready to create a study. That can take about 15 seconds, so the wait needed designing too.
The services team use it, and I don’t have a measured result for it
I took it from a blank page to a working tool, across discovery, flows and dev-ready specs. There’s no conversion number to show you here, so judge this one on the decisions.
I set up Prolific’s tokens so a new brand could be remapped in, and that brand update is paused
Prolific had a brand update coming, so I set up the tokens and variables ready for it: the name groups, primitives, typescale, typefaces, density and themes. The older colours went in under the new token names, so the new brand could be remapped onto them. That update is paused.
Density is the part I’d point to first. Compact sits one step down the spacing ladder and Comfortable one step up, so the whole scale moves together.
The pill shows how far down that goes. Its height, padding, text, icon box and icon stroke all change with the mode, so nobody has to build size variants by hand.
Icons got their own collection too, with stroke weight that grows with size and sizes that pair with text.
The system is now ready for when the new brand comes in, so the components can be reskinned easily.
Some good work never shipped
Each of the four product managers meant starting again and rebuilding trust, and priorities moved with them. We had a polished refer-a-friend system ready to go, and it got deprioritised. We designed a customer satisfaction feature to capture sentiment at key points in the journey, and it never shipped.
What I learnt
My time at Prolific wasn’t what I’d expected. It was messier, longer and more chaotic, and it was also some of the most fun and valuable work I’ve done.
What Prolific and my team said
In the London office, in front of the product team, Sara, the VP of Product, said “Ben has been the single best ROI our team made all year.” I went beetroot red. I was a contractor, but in that moment I felt like part of the team.
