T-MOBILE

An AI assistant working in tandem with a live agent

T-Mobile's AI Assistant is built to resolve what it can and connect customers to a live agent whenever they ask.

AGENCY

Kettle

ROLE

Lead Interaction Designer

MY CONTRIBUTIONS

UX/UI
IA
AI Design
Conversational Design
Prototype

Before the AI Assistant, every support question went to a live agent, which left customers waiting until one was available. The brief was to cut that wait and let the AI resolve what it could on its own. At any point, a customer could ask to speak with a live agent, and the AI Assistant would seamlessly connect them.

THE CHALLENGE
Infobot hero image

33%

FEWER DAILY LIVE-CARE CHATS

4.5M+

CHATS HANDLED

One conversation, two visual languages for AI and a live agent

A customer should always know whether they're talking to the AI or a person. With that in mind, I designed the live agent chat with the familiar text-message look: a name and chat bubble. The AI chat was deliberately built to look different: a logo and text, no bubble.

Infobot, section 3

The handoff between AI and agent, one continuous thread

I led the design of one chat thread that carries customers between the AI and a live agent. A customer can ask for a live agent at any point. Just as important, a customer can end the conversation with the live agent and have the AI pick back up, without restarting the chat or opening a new conversation.

AI Assistant has reduced daily live-care chat volume by roughly 33%

The AI Assistant is now T-Mobile's primary entry point for support conversations, having managed over 4.5 million chats. Most customers start there, and roughly a third of conversations are resolved without a live agent.

Measured roughly two months after launch.