Cloudfox · Payments and checkout for Brazilian merchants
Cutting checkout abandonment from 68% to 26%
Cloudfox was a checkout and payments platform for Brazilian merchants, acquired by Vega after processing R$ 23.5M (about US$ 4.5M). As its first designer, I decided what to cut, bet on a checkout merchants could call their own, and designed the recovery and reports around it.
The merchant dashboard: revenue, approved sales and balances first, then the sales trend and the latest transactions.
Results
Four flows, four measured changes
Checkout and recovery figures are measured on the value of closed sales.
The problem
Ten features at ten percent
When I joined, the dashboard listed ten features, each roughly ten percent complete. There was no design system and no navigation logic tying them together. Merchants saw breadth, but nothing they could rely on end to end.
I was the only designer, with no product manager. Deciding what to finish first was not a precondition for the design work. It was the design work.
The approach
Cut the widget, own the checkout
I reframed the roadmap around what merchants needed to do, not the features already started. Three jobs gave the team an order of work, and they structure the rest of this case.
- Understand the businessFinance, sales and marketing reports
- Earn moreCheckout editor, buyer checkout and payment recovery
- Manage moneySales ledger, balances and statements
The hardest call was what to stop. We cut the embeddable checkout widget. Instead of a platform-branded checkout like Hotmart’s, we put the effort into a transparent, white-label checkout that ran under each merchant’s brand, and into an editor that made it look like theirs. The Brazilian market was not offering that at the time.
We also had to choose which number to own. Chargebacks and checkout abandonment both cost merchants money, but we had far more control over abandonment. I proposed making it the team’s first metric, and every scope decision after that was weighed against it with the CEO, who spoke for merchants, the Tech Lead, who set what was feasible, and the CMO, who owned the brand behind white labelling.
Reports
Three numbers first, depth on demand
Merchants could not improve what they could not see, and they did not have time to read a report. Each module opened on three numbers and nothing else. In finance: net revenue, average net ticket and approval rate. Everything else stayed one click away, never in the way.
Analytics opens on three numbers, then net revenue by day, by payment method and from gross sales to net.
Marketing showed conversion across Brazil’s states, because that is how merchants already compared regions and planned campaigns.
Checkout editor
Remove the fear of publishing
A white-label checkout is only an advantage if merchants actually make it theirs, and most did not. They were afraid of publishing something broken to live customers. The fix was not more options but more certainty: one editor for the look and the rules, where every change is visible before it reaches a buyer.
One editor for the look and the rules: a one-page checkout, a full-width banner and a theme checked on mobile, then a wallet added, a carrier switched off and free shipping limited to three states.
- Start from a theme.Ready-made themes give merchants a safe starting point before they touch a single colour.
- Rules beside the look.Payment methods, discounts and free shipping by state sit in the same editor as the logo and colours.
- Nothing goes live unseen.A live preview, on desktop and mobile, shows every change before a buyer does.
Sixty days after launch, merchants using at least one customisation rose from 18% to 64%.
Buyer checkout
Ask for the least costly information first
The merchant editor only mattered if buyers finished paying. The order of the buyer checkout was a business decision as much as a usability one. Contact details came first, so a merchant could follow up with a buyer who left. Delivery came next, to calculate shipping. Payment came last.
The buyer checkout opens on contact details, with the order summary always in view.
The editor also offered a one-page checkout, but three steps stayed the default. On a single page, a buyer who leaves halfway leaves nothing behind. With contact first, every abandoned checkout was still a lead. Checkout abandonment fell from 68% to 26%.
Payment recovery
A failed payment is not the end of the order
In Brazil, many buyers pay by boleto, a bank slip that is often shared and paid through WhatsApp. That made the confirmation page part of the payment, not the end of the flow. Redesigning it to help buyers complete payment raised boleto completion by 24%.
A declined card needed a next step too. Instead of an error, buyers got two alternatives: Pix with a 5% discount, or a boleto. Drop-off after a declined card fell from 24% to 12%.
After a declined card, the order stays on screen and the buyer gets two ways to finish: Pix with 5% off, or a boleto. Here the buyer picks Pix and pays R$ 581,15 instead of R$ 610,90.
Sales ledger
One place to answer “where is my money?”
For a merchant, a sale is not finished until the money arrives. The ledger brought every transaction into one table, with its payment method, status, dates and commission, and filters for date, order ID and customer.
The period, status tabs, payment filter and search work on one dataset: change any of them and the cards, the chart and the table answer together.
Statuses carried most of the meaning. Approved, awaiting, in review, disputed, refunded and cancelled each needed a different action from the merchant, so each got a distinct colour and label.
What I learned
The scope decision mattered more than any single screen
The measurable gains came from the editor, the checkout and recovery. They existed because we stopped building a widget and ten partial features, and picked one metric we could actually move. Choosing what not to build was the highest-leverage design decision on the project.
The recovery work taught me the second lesson. The most valuable moments were not on the happy path. A declined card or an unpaid boleto was where a clear next step was worth the most.
What I would do differently: set up measurement before each release, not after it. With a baseline for every flow, each result in this case could be tied to the decision behind it.
