GoFood's ordering funnel had three steps. The first two held up fine. The last one didn't. Almost half the users who made it to checkout — people who had already browsed, chosen a restaurant, built a cart — never placed an order.
This isn't a friction problem. It's a trust problem that only surfaces after the user has already invested.
For years, checkout was treated as a confirmation screen. We started treating it as a conversion screen.
We ran a drop-off study to understand why users abandoned checkout. The top four reasons were all about price and promo. Deals existed. Users couldn't see them.
Promo application had stagnated at ~60% for some time, with new users under-applying despite qualifying. The problem wasn't friction in the form — it was discoverability.
"The problem wasn't friction. It was promo discoverability and relevancy."Drop-off study finding, n = 903
Over Jul–Sep, 75,422 cancellations traced back to information that should have been confirmed at checkout. This reframed the work entirely. Checkout isn't just a conversion screen — it's the last reliable point where an order can still be made correct.
I rolled the symptoms into three improvement bets. Together they pointed to a north star: "Help users accurately and quickly check out while having the best deal possible."
"Help users accurately and quickly check out while having the best deal possible."North star, GoFood Checkout redesign
The redesign consolidates five components. Milestone 1 ships proven components; one moves to A/B testing because usability research deliberately split on it.
The delivery options component was where UT didn't converge — and we resisted the urge to pick by taste.
Driven by the components that did the quietest work — clarified visual hierarchy, compact address, restructured SKU cart, pre/post-savings price. The changes nobody would describe as flashy were the ones that moved the funnel.
This was the cohort the drop-off study said was hurting most. The lift landing specifically on new users is the satisfying confirmation: we designed for the right segment.
From the floating offer bar. Smaller in absolute terms — but pointing somewhere important. Users who weren't promo-sensitive ignored it, which is the right failure mode for a component like this.
The segment-specific lift tells a bigger story: our highest-leverage next step isn't designing one better screen — it's designing the right screen for each user type.
"The interesting finding wasn't the lift itself. It was that the lift was segment-specific — which means our biggest next opportunity is designing the right screen for each segment."Post-launch reflection
The floating offer bar's segment-specific lift is the clearest signal in the dataset. The highest-leverage version of this work going forward isn't a universal redesign — it's a checkout that adapts to who's looking at it.
The +2.9% for new users came from visual hierarchy and separation work, not from the components I expected to headline. Clarity at the structural level is usually undervalued vs. clever new components.
The real test isn't this launch — it's whether the next checkout addition respects the framework or breaks it. That requires design and product alignment to hold the line.
Conversion lift was measurable inside the existing experiment infrastructure. Cancellation reduction requires connecting checkout-time edits to downstream cancellation reasons. I'd invest in that instrumentation upfront next time.
Pre-experiment, I'd propose the segment cut as a primary metric, not secondary. The story the data tells is clearer by segment than in aggregate — but we set up experiments thinking aggregate would be the headline. That framing cost us a sharper readout.