NeverStartOver – Dropped-call continuity on WhatsApp

Lose the connection, keep the conversation. When a call drops, AWS End User Messaging Social continues it on WhatsApp with full context, so an AI agent gets help moving and no one starts over.

Comment

Inspiration

We're passionate about great customer experience. At Serve7 we spend our days designing contact-centre journeys, and the moment that frustrates customers most is a dropped call. They call back, wait in the queue again, and repeat everything to a new agent. That moment matters most when someone is stuck on the side of a highway with patchy coverage, exactly when a phone call is least reliable.

AWS End User Messaging Social gave us another channel to protect the conversation when a traditional phone call isn't possible. When the call drops, the conversation doesn't. It carries on in WhatsApp, with everything the customer already told us.

What it does

NeverStartOver is a roadside-assistance contact centre where a dropped call never means starting over.

  1. A driver breaks down and calls Serve7 Roadside (Amazon Connect). They enter their membership number and issue and give explicit consent to be contacted on WhatsApp if the call drops.
  2. The call drops while they're on hold. An EventBridge contact event triggers a Lambda, which uses AWS End User Messaging Social to send a WhatsApp message that picks up the conversation with the details already given, plus a "Send location" interactive button. Amazon SES sends an email as a secondary channel.
  3. A Bedrock AgentCore agent (Claude) takes over the WhatsApp conversation through EUM Social → SNS → Lambda. It never re-asks for information it already has.
  4. The customer shares a location pin. The agent confirms it, sends a live map link and dispatches a driver, with the route and ETA from Amazon Location Service.
  5. An interactive WhatsApp dispatch message (driver, truck, ETA, "Track driver" button) and a driver contact card with tap-to-call keep the customer informed. SES sends a case summary email at the end.

How we built it

We built on our contact-centre and WhatsApp experience and treated this as a challenge: show how rich WhatsApp interactive messages can be when they're part of a real service journey, not just a notification. Location requests, interactive buttons and contact cards all come from AWS End User Messaging Social, and Amazon SES handles email.

We used the AI-Driven Development Lifecycle (AI-DLC) with Claude Code agents to move quickly from requirements to working units. That let us build the agent, the infrastructure, the Connect flows, the live map pages and the UI across many AWS services in a short time.

Built on: AWS End User Messaging Social (WhatsApp), Amazon SES, Amazon Connect, Amazon Bedrock AgentCore, Amazon Bedrock (Claude), AWS Lambda, Amazon DynamoDB, Amazon EventBridge, Amazon SNS, Amazon Location Service.

Challenges we ran into

  • Making the story realistic. We wanted a scenario judges would recognise immediately: a real stretch of highway with patchy coverage, a real breakdown and a believable dispatch. Then we built a storyline that shows each capability at the moment a customer would actually need it.
  • Choosing the right technology for each step. Voice stays in Amazon Connect. WhatsApp runs directly on EUM Social, because the agent needs location messages and rich interactive content. SES carries the email. Consent comes before any outreach.
  • Understanding EUM Social's capabilities. We relied heavily on the EUM Social documentation to understand the 24-hour customer-service window, which interactive message types we could send without templates, and how to route inbound events to our agent.

What's next

We're turning NeverStartOver into a repeatable offering for Canadian organisations: roadside assistance, field service, utilities, healthcare. Any business where a dropped call costs customer trust can keep the conversation going on the channel the customer prefers.

Built With

Share this project:

Updates