Read the launch here
For millions of people, WhatsApp is where conversations happen, and community is built.
It’s a trusted platform where families stay connected across borders. Sending money home is rarely just a transaction. For most customers, it’s a conversation that’s already happening on WhatsApp with family.
Remitly meets them there, offering a guided experience from onboarding to international transfers, inside a platform that already feels like home.
Role: Conversation Designer Type: Agentic AI Experience Platform: WhatsApp
Timeline: Q1–Q2 2026 Status: Designed and ready for A/B testing
Remitly’s early WhatsApp experience was built like a structured workflow and not a conversation.
Customers who wanted to send money abroad had to navigate a rigid, step-by-step flow that didn’t understand natural language, didn’t adapt to how they spoke, and didn’t recover gracefully when something went wrong. The system used a workflow tool (Cognigy) that treated every interaction in a structured format, whether the customer typed “hi” or “I want to send $100 to my mom in Mexico.”
A first attempt at an agentic upgrade made things worse, not better.
Customers were being handed off to human support agents at scale, not because they needed help, but because the system couldn’t handle ambiguity. Latency averaged ~15 seconds per response. Language mismatches went unresolved.
<aside> 💡
The product needed to meet customers in their language, on their terms, without making them work hard to send money home.
</aside>
I was the sole Conversation Designer on this project, embedded within the SCOUT team. I worked alongside a Product Manager, Engineers, and a Design manager to define the content experience for Remitly’s migration to Sierra, a more powerful, flexible AI agent platform designed to deliver a faster, more natural conversational experience.
My work spanned the full content lifecycle: auditing the existing experience, defining voice and tone, building conversation guidelines, creating the AI prompt framework, and stress-testing the prototype for content issues before A/B testing.