Payments Home

The architecture of Payments Home came down to how much a single screen could carry. I worked through a two-tabbed structure first, with account management on one side and transactions on the other, and decided against it once I determined the two tabs would compete for engagement and discoverability. To prevent bloat, I settled on truncating sections at 2 items max while showing enough information to communicate intent, and implementing secondary pages to house further actions and content.

The harder call was Shared with you, a section listing the Pix keys people have recently sent you. Keys are tied to transactions, and user research kept surfacing the same behavior: people scrolling back through chats to find someone’s key so they could pay them. Duplicating a full keys UI in both places made no sense, so I split the work between them. Shared with you presents the keys on their own, ready to copy. The transaction list still records them, but only as an entry labeled Shared with you, with a Pix icon over the sender’s photo to mark it as a key rather than a payment. It has become one of the parts people respond to most, and it came from watching local behavior rather than the data model.

Since I was designing for the Brazil market, I led with the belief that local behavior should determine the design, even when stacked against existing markets that were further along in development and traction. India had already shipped its own Payments Home successfully, lifting payment registrations by 8%, so there was real pressure to copy it screen for screen. The two markets are not the same underneath. India runs on UPI, its national instant payment network, while Brazil runs on Pix and did not have native payments within WhatsApp yet. Critically, Brazil also carries far higher rates of fraud and scams. I kept the parts that transferred, including the profile-centric model, a similar approach to transaction history, and persistent education for first-time users, and altered the parts that did not.

The main change was pulling transaction history to the front of the screen rather than one level down with a deprioritized entry point, which India had done. Someone in Brazil who has just been scammed needs to dispute the payment, report the account, or block the sender immediately, and that is not a task to bury. Every divergence took active advocacy in review, and the position I held throughout was that Brazil would not trade away trust and transparency to match another market’s experience.

Interaction patterns and trade-offs between options are sometimes difficult to settle on paper, so I built something stakeholders could experience for themselves. Through a cocktail of Figma, Cursor, and Claude Code, with Obsidian as the brain, I created a working iOS prototype with a native SwiftUI build, complete with transitions and edge cases. Most of the work went into “teaching” the AI how to recognize and hold to our actual design system components rather than use vanilla OS elements, which was critical in keeping the fidelity close to production.

This prototype surfaced error and edge states early in the process, while the spec was still flexible and cheap to change. It also let leadership react to the strategy instead of debating individual interactions, and in 48 hours it produced more usable feedback than the static spec had in two weeks.