Skip to content
Illustrative Case Study

The AI worked perfectly. It just never made anyone want to go anywhere.

TravAI can plan a comprehensive trip in minutes. The sessions found exactly why that never translated into anyone feeling excited to book one, and what a travel app needs to do differently when every user arrives already burnt out.

The brief

Users came looking for a guide. They found a very clever directory instead.

TravAI's AI can plan an entire trip in minutes, comprehensively, accurately, impressively. None of that mattered to users who arrived burnt out from comparing prices across a dozen tabs, because the product they landed in still asked them to do the comparing. The brief was to find out why sophisticated capability wasn't translating into anyone feeling relieved, and what a genuinely guided experience would need to look like instead.

01
Trust on entry scored 2 out of 10. Not because the AI was unreliable, but because it stayed invisible in the background exactly when users most needed to see it working.
02
Users arrived already carrying analysis paralysis from every other booking platform. The product responded with more comprehensive options, recreating the exact overwhelm it was meant to relieve.
03
Budget and personal preferences were requested before the AI had shown any evidence it understood the trip. Vulnerability was asked for before trust had been earned.
The process

Fourteen sessions. The same gap, from every angle.

The First 60 Seconds and Aspiration Gap sessions, run independently, both landed on the same internal monologue: "Is AI trustworthy? Will this help me, or is it just going to spit out generic answers?" The Heartbreak Scale put a number on it, trust at 2 out of 10, confusion at 4 out of 10, users unsure whether the burden had actually moved off their shoulders at all.

Core PrinciplesThe Heartbreak Scale™75 min
DiscoveryThe First 60 Seconds™30 min
DiscoveryThe Aspiration Gap™60 min
DiscoveryThe Identity Shift™45 min
DiscoveryDay One / Day 90™45 min
Brand PersonalityThe Dinner Party™15 min
Brand PersonalityThe Word Sort™20 min
Tone & CopyThe Voice Sort™40 min
Tone & CopyWrite It Wrong™30 min
Tone & CopyRead It Aloud™30 min
Design PrinciplesDesign Principles Builder™60 min
Design PrinciplesThe Anti-Principles™40 min
Visual DirectionOpposite Ends™10 min
Visual DirectionMoodboard Speed Dating™15 min
"Powered by AI." "Seamless." "Comprehensive." Every one of these landed on the Voice Sort's wrong pile. The words that felt right were smaller: guided, understood, relieved, sorted.
Output 01: Strategy & Research

From which one is worth my time to I can do this.

The Feel Factor mapped three layers, first impression, the journey, identity transformation, and found the same problem at every one: sophisticated technology that had forgotten to consider how it made anyone feel. The Aspiration Gap traced the three moments where that gap actually closes. Design Principles turned both into rules specific enough to test any screen against.

The Feel Factor®
The Feel Factor

The Feel Factor®

Three layers, first impression, the journey, identity transformation, mapping a product that functioned perfectly while forgetting to make anyone want to use it.

Trust measured at 2/10. Users left every session feeling like they'd engaged with a comparison site rather than a personal travel guide.
The AI exists but stays invisible exactly when users need it most, the worst possible outcome: all the stress of planning, plus doubt about whether the technology is helping at all.
The desired shift isn't excitement about a feature. It's the quiet, specific belief: "I can do this."
Download document →
The Aspiration Gap™
Discovery & Strategy

The Aspiration Gap™

The distance between a user arriving burnt out and one who books with confidence, mapped across three moments where that gap actually closes.

Users arrive asking "will this help me, or is it just going to spit out generic answers?", a reasonable question after every other platform promised the same thing.
Three thresholds govern the journey: the guided conversation, the personal roadmap, and one-click confidence at the end of it.
The end belief users actually reach for, in their own words: "I don't need to manage my whole holiday, I just need to enjoy it."
Download document →
Design Principles
Design

Design Principles

Three principles, each with a measurable signal, governing the sequence between listening, understanding, and asking for anything in return.

Principle one: listen first, suggest second. Budget questions never appear before the AI has demonstrated it understands the trip.
Principle two: no screen shows more than three options. For an overwhelmed user, reduction is relief, not limitation.
Principle three: the first thing someone asks for isn't always what they need. The AI's job is to notice the contradiction and ask, not execute the request as given.
Download document →
Plus 8 more documents in the full engagement
04
Research & Strategy
Research & Insights Report
05
Research & Strategy
User Personas
06
Brand
Brand Personality
07
Brand
Tone & Copy
08
Brand
Copy Handbook
09
Design
Visual Direction
10
Design
Interaction Principles
11
Design
Animation and Interaction Guide
Output 02: The Design

One conversation that never asks twice.

From the first question through to the trip hub that still has something to say weeks later. Tap through the screens below.

Original
WAA
Output 03: Technology

A stack built around one requirement: the AI has to remember.

Every technology decision was tested against the same product requirement, conversation state and reasoning that persists indefinitely, real-time multi-vendor coordination, and financial transactions that need to be trusted with someone's holiday budget.

Tech Stack Recommendations
Technical

Tech Stack Recommendations

React Native and NestJS for a single conversational codebase, PostgreSQL and Redis for state that has to survive between sessions, GPT-4 and Pinecone for reasoning that stays visible.

Redis isn't just caching here, it's the reason conversation context and the user's own language survive across sessions instead of resetting every time they return.
Stripe Connect handles multi-vendor payouts automatically, since a single booking can mean coordinating flights, hotels, and activities from entirely different suppliers.
Every choice was weighed against proven reliability over novelty, because trust with someone's holiday budget doesn't tolerate architectural experiments.
Download document →
Developer Handoff
Delivery

Developer Handoff

Five non-negotiables, system copy rules for every error and empty state, and acceptance criteria that judge the build on how it feels, not just whether it functions.

No screen may present more than three primary options at once, and conversation state must persist across every session, not just within one.
Every recommendation has to quote the user's own words back to them. Generic reasoning like "based on your preferences" is explicitly banned.
When someone deviates from a suggested path, the AI adapts rather than redirects, no "back to recommendations" prompts, ever.
Download document →
Plus 3 more documents in the full engagement
03
Technical
Technical Architecture
04
Technical
Technical Blueprint
05
Technical
Implementation Roadmap
The result

A product judged on whether someone feels guided, not just whether the AI ran.

Trust at 2 out of 10 explained almost everything about where the existing experience was losing people, capable technology, wrong sequence. The response was a set of decisions specific enough to hold under deadline pressure, and a build handoff that gave them teeth.

A sequence, not a wishlist
Listen, then demonstrate understanding, then ask for anything vulnerable. Every design decision could now be tested against that order instead of argued over on taste.
Three options, never more
A hard constraint, not a guideline. For users who arrived already carrying analysis paralysis, reduction turned out to be the entire product.
Reasoning that quotes the user, not the algorithm
Every recommendation has to reference what someone actually said, in their own words. Generic personalisation language was named and banned outright.
A stack chosen for memory, not just speed
Conversation state and the user's own language persist indefinitely. Returning users are met as someone the AI already knows, not someone starting over.
More work

Other case studies