AMA — Agentic Agent for Travel

Problem

Hours lost to manual knowledge work

Travel agents spent 5–30 minutes per query hunting through fare manuals, GDS docs, and Slack threads for answers that already existed — just scattered. Shadowing sessions and ticket analysis showed agents lost up to 40% of their day this way, and 80% of questions were variations of the same 20 scenarios.

"Agents didn't want to be replaced by AI — they wanted a knowledgeable colleague on demand."

Approach

Designing for trust, not just speed

An AI that gives wrong fare advice can cause real financial harm — so the core challenge wasn't the chat interface, it was making agents trust the answer enough to act on it.

01

Discovery

Shadowing sessions, ticket analysis, 12 agent interviews.

02

Framing

Defined the AI's voice — confident but citable. Scoped MVP queries.

03

Design

Iterated on conversation UI, source citations, suggested questions.

04

Validate

Weekly usability tests measuring trust, speed, and error rates.

Solution

From hours of search to seconds of certainty

Agents type a question in natural language and get a structured, source-cited answer in seconds — showing exactly which manual or GDS doc it came from. That single change is what earned their trust.

Before

5–30 minutes per query
3–5 browser tabs of manuals
Ask a senior colleague on Slack

After

Answer in under 5 seconds
Source cited inline
Full chat history, anytime

Outcomes

Faster answers, more confident agents

<5sAvg. time to answer

80%Recurring queries covered

↓Senior-agent interruptions

"AI interfaces aren't just about speed — they're about earning enough trust that people stake their professional judgment on the answer."

Designing an AI agent for travel