Bringing Back the Thrill of the Auction
Overview
At the end of 2024, Catawiki made a deliberate strategic reset. A prior push into buy-it-now and fixed-price listings hadn't delivered the trade-off it promised: those objects sold at a lower rate than auctioned ones in almost every category, and existing buyers were visiting and bidding less often than before. User research pointed to the reason: when buyers described what they actually loved about Catawiki, the answer wasn't the objects for sale, it was the act of bidding itself: the competition, the suspense, the moment of winning. So Catawiki recommitted to the auction as its core experience, the format its own buyers had told it they wanted.
Bringing that energy back meant investing in three connected levers: a stronger sense of urgency throughout the auction lifecycle, credible social proof and competition, and nudges that kept buyers engaged mid-bidding rather than losing them after a single bid. Placing a second or third bid mattered just as much as placing the first one.
The object detail page was where most of this had to converge. It's the surface where a buyer decides whether to trust the object, the seller, and the platform enough to bid, and by 2025 it had accumulated years of usability debt on top of the strategic drift: unclear hierarchy, cluttered information, trust signals that weren't landing. Redesigning it wasn't a separate initiative from the thrill-of-the-auction strategy, it was the infrastructure that made the strategy liveable. Timers, urgency cues, and social proof needed a page built to hold them with intention, or they would have piled onto an already disorganised surface and diminished each other's impact.
As Head of Design, I translated the strategic shift into concrete design direction across every surface throughout the buyer's journey, secured organizational buy-in, and coordinated the workstreams and designers responsible for each from discovery through to delivery.
Role & Approach
My role had two dimensions that ran simultaneously throughout the initiative. The first was organisational: before any design work began, I ran discovery workshops with the executive team and with 15 experts and category leads from the commercial department, compiling their insights into the strategic foundation the whole program worked from. I paired this with the voice of the customer, drawing on what buyers were already telling us about what made bidding worth coming back for. Getting leadership and commercial stakeholders to a shared understanding of what "the thrill of the auction" actually meant in practice made every subsequent conversation about priorities and trade-offs across every surface significantly smoother.
The second dimension was creative direction across the surfaces the strategy touched: the object detail page, category and auction pages, and the bidding process itself, including notifications. Nearly all of this was approached mobile-first, since that's where the majority of bidding activity happened, with the app carrying its own specific additions on top, like live auction updates. I ran design jams to establish a north star vision, then worked with the PMs and director of product to slice that vision into testable, iterable releases across the teams responsible for each surface.
My day-to-day involvement was hands-on. Inside the design jams I wasn't just facilitating, I was proposing ideas and working through solutions directly alongside the designers and PMs, and I stepped into the design work myself when it was needed. We validated our thinking with unmoderated user tests via Maze and rigorous A/B testing, since not every idea that looked promising on paper held up once it was in front of real bidders.
Success was measured primarily by winning and bidding rate, with bids per user and return rate as the next layer of signal.
Key Decisions
These decisions weren't arbitrary starting points. The buyer priorities behind the strategy were explicit: existing buyers wanted better auction discovery and more desirable objects, they wanted the thrill of the auction itself, urgency, social proof, competition, and they wanted reasons to keep coming back. Each decision below sits inside one of those buckets.
Mobile mattered throughout, though not in the same way for every decision. Each surface was designed and tested on its own platform, and a version performing well on desktop didn't always hold up on mobile, or the reverse. But since 70-75% of users bid on mobile, mobile performance was the bar any design ultimately had to clear, whatever happened elsewhere.
Once the priorities were set, turning them into an actual page meant taking the ODP apart piece by piece and checking each part against what a buyer needed to feel or do at that exact moment: understand the auction, want the object, feel ready to act, or want to keep looking. The section where a buyer actually decided whether to bid, the price, the timer, the bid history, the button itself, was where the existing page worked against itself the most: the timer looked identical whether an auction closed in five days or five seconds, pricing information was scattered across the panel instead of sitting together, and the bidders listed in the bid history read more like placeholder data than real people.
What follows is how each of those specific problems got solved.
1. Making the timer reflect the auction.
The existing timer was flat: "Closes in 4h 39m 43s," styled identically whether a lot had days left or was about to close. It gave buyers no reason to feel the moment mattered, and it meant an early bidder who'd moved on had nothing pulling them back when it counted. The fix made the timer's appearance track the actual state of the auction: it grew larger and more insistent as the close approached, colour and animation shifting in the final hour, an "Ending today" treatment replacing the static label, paired with outbid signals so a lost lead was as visible as the time running out.
Before
After
The same treatment extended past the page itself. A live activity on the lock screen and home screen showed the same information, time remaining and the buyer's current bid, animated the same way, so the countdown didn't stop mattering just because someone closed the app. It mattered most in the final four hours before a lot closed, the exact window the on-page timer was already built to escalate around.
2. Trusting the data over the instinct.
The intuitive bet was that bigger, more immersive photos would build more confidence to bid: the image gallery got a full redesign, larger pictures, thumb-swipe, a progress bar. On desktop, it worked as expected and shipped with no downside. On mobile it consistently hurt the metrics that mattered, serious bidders down, bids per user down, across iOS and Android, even though people engaged with the new gallery more, scrolling further into it than before. The instinct that buyers wanted bigger images wasn't wrong, it just wasn't converting into more bidding.
The first iteration failed: bidding activity dropped on mobile, even as gallery engagement itself went up. Rather than over-invest in diagnosing the gap between the two, we deprioritized it, focused on higher-impact problems, and flagged it for a later return.
Vision for the gallery
Reverted gallery
3. Placing social proof where it actually helped.
Alongside the timer, this was the other core piece of bringing the thrill of the auction back: buyers weren't just choosing an object, they were racing other people for it. But that only worked if the competition showed up in the right place on the page. Bid activity, how many people were bidding, how many were watching, originally sat lower down, treated as background information alongside the object description and seller details. It didn't belong there, it was information about the competition, not the object, and it needed to live where the buttons were, the part of the page focused on justifying and acting on a bid.
The first version overcorrected: bold red copy on a coloured background, pushing the sense of competition hard and constantly. It read as aggressive rather than exciting, and bidding dropped. The fix toned it down considerably: a plain count of active bids and how many people were following the object, with the full bid history only expanding when someone tapped to place a bid on mobile, the exact moment competition actually mattered to them. This version worked, GMV won per participant on web rose over 6%, without the pushiness that had worked against the first attempt.
4. Holding the quality bar on AI-generated content.
Two new pieces of the page ran on AI-generated content: a one-sentence object summary meant to help buyers grasp what made something worth bidding on without reading a full seller description, and a generated expert bio meant to reinforce trust in the object's selection.
The summary worked from the start, bidders rose by over 1.6%, and a later iteration on the model and prompts pushed it further still.
The expert bio didn't. Digging into why revealed the generated content itself was subpar. It went through prompt improvements and changes to the underlying content logic, then relaunched successfully.
Generating content with AI isn't something you can fully judge before it ships. What the model actually produces at scale, across thousands of different objects, isn't fully predictable in advance. Stopping to reassess, adjusting the model and the logic behind it, and being willing to revise after launch mattered as much as the original decision to build it.
These four decisions are representative, not exhaustive. Alongside them, a much larger set of changes shipped across the ODP and beyond: trust signals like free shipping and no-reserve-price labels, a redesigned seller block and page, human seller names replacing anonymized ones, quick bid buttons at the point of confirmation, outbid notifications and CRM improvements, and a "Stay in the Game" section on Home surfacing time-sensitive auctions buyers were already engaged with.
Outcome & Impact
What shipped
- Consistent improvements to bidding rate and winning rate across most experiments launched during the initiative.
- A dynamic timer, on the page and as a lock screen live activity, replaced a static countdown and became one of the clearest signals of urgency in the redesigned experience.
- Social proof improvements in the bidding panel meaningfully increased GMV won and reserve-price winners, after the failed first version was stopped, diagnosed, and relaunched.
- AI-generated object summaries improved bidder rates from the start, and improved further after a later content iteration.
- AI-generated expert bios initially underperformed due to content quality issues, then improved bidding confidence after a round of prompt and logic fixes.
- 61,000 anonymized seller names replaced with human-readable ones, contributing to higher bid values and improved payment completion rates.
Leadership impact
- The discovery workshops I led before design began created a shared language around the strategic direction that held throughout a long, complex initiative spanning multiple teams and surfaces.
- The discipline around stopping underperforming experiments, the gallery, the first social proof version, and holding the quality bar on AI content, meant that what shipped was meaningfully better than what it replaced.
- Directed design across the ODP, the bidding experience, and adjacent surfaces like notifications and live activities from a single point of accountability, rather than each shipping in isolation.
Reflection
This project reinforced something I already believed but hadn't seen proven at this scale: that alignment on shared language is foundational work. The workshops I led before any design work began did more than inform the direction. They created a common vocabulary around what "the thrill of the auction" actually meant, and that made every subsequent conversation about roadmap, priorities, and trade-offs significantly smoother, well beyond the product and design teams.
It also taught me something more specific about designing with AI-generated content: it's not something you can fully judge before it ships. What a model actually produces at scale, across thousands of different objects, isn't fully predictable in advance, so testing, catching quality issues early, and being willing to revise after launch has to be part of the process from the start, not a sign that something went wrong.
If I did this again, I'd push harder for isolating one variable at a time before testing. The gallery redesign bundled a larger image gallery, a smaller title, and new button placement into a single test, which made it harder to tell which specific change was hurting bidding on mobile, and slower to get a clear answer. I'd rather ship a narrower test first and add complexity once I knew which single change was actually doing the work.
- For: Catawiki
- When: 2025
- What: Design leadership, discovery workshops, creative direction, A/B testing, Maze user testing