AI Chat for Payments

This project focused on a specific hypothesis: that for users with low digital-literacy or low confidence navigating structured payment forms, natural language was a more accessible interface than traditional UI patterns.

Building on Moneda's existing wallet infrastructure in the Dominican Republic, I designed an AI-powered chat interface that let users initiate payments, manage transactions, and interact with service providers through conversation — without having to navigate menus, forms, or multi-step flows.

The existing payment experience required users to know which biller they were looking for, navigate a categorized list, fill in account numbers, and confirm amounts — a multi-step process that assumed a level of digital confidence many of our users didn't have.

Drop-off rates in the bill-pay flow were high, and qualitative research pointed to a consistent frustration: users knew what they wanted to pay, but couldn't figure out how to find it in the UI.

The design challenge for an AI chat interface isn't building a chatbot — it's defining what the AI should and shouldn't do, and making the boundary feel natural rather than frustrating.

I mapped the full range of user intents in the payment context: bill pay, P2P transfer, balance inquiry, transaction history. For each, I designed conversation flows that handled successful resolution, ambiguity, and failure gracefully — including what the AI says when it doesn't understand, and how it hands off to a human when needed.

Prototyping was conversational: I wrote out dozens of sample dialogues before touching any visual design, then tested them with users to understand where the natural language interpretation broke down.

The AI chat interface acts as a financial assistant embedded in the Moneda app — understanding user intent through natural language, resolving it to the correct biller, amount, and payment method, and confirming before executing.

The design prioritized legibility of the AI's actions over speed. Every transaction the AI initiated was presented as a confirmation before it was processed — with a plain-language summary of what was about to happen. Trust, not velocity, was the metric that mattered most.

Early adoption metrics showed a meaningful increase in completed bill-pay transactions and a reduction in drop-off compared to the structured form flow. User satisfaction scores improved across the board.

More importantly, the project validated the core hypothesis: for this user base, natural language is a more accessible interface than traditional UI patterns. That finding has shaped how we think about all subsequent feature development.

Designing for AI reinforced something I believe about all interaction design: the interface is only as good as the thinking behind it. A chat interface can feel magical or it can feel broken — the difference is how precisely you've defined the scope of what it can do and how gracefully it handles everything outside that scope.

I came away from this project with a conviction that AI features should be designed for trust before capability. An AI that does less but earns trust will always outperform one that attempts more and fails unpredictably.