Results
L·AI·C launched and scaled quickly. The numbers reflect what good conversation design does when paired with the right data infrastructure — it removes friction so completely that members don't think of it as using a chatbot. They just got their answer.
Results
L·AI·C launched and scaled quickly. The numbers reflect what good conversation design does when paired with the right data infrastructure — it removes friction so completely that members don't think of it as using a chatbot. They just got their answer.
Conversation design — every word is a decision
Conversational UX is different from screen UX. There's no layout to guide the eye, no button hierarchy to signal priority. The entire experience lives in the words and the order they appear. Every response has to do three things at once: confirm the AI understood, provide the answer, and make the next step obvious.
I designed the conversation flows for all four core journeys, mapping both the happy path and the edge cases — class full, intent unclear, conflicting information, wrong club. The hardest design problem wasn't the happy path. It was designing graceful fallbacks that didn't make members feel like they'd hit a wall.
Discovery
I started with discovery: stakeholder interviews with product, engineering, and ops; an audit of existing member support channels; competitive analysis of how other AI tools handle similar contexts; and a deep dive into the edge cases — what happens when the bot doesn't know, gets it wrong, or needs a human.
The research surfaced four core journeys that covered the vast majority of member needs: class booking, account management, location info, and general support. Everything else was an edge case layered on top of those four.
Class booking — find available classes by type, time, club; confirm booking in one turn
Account management — pause, cancel, update billing; 80% self-serve without human involvement
Location info — real-time hours, pool status, amenities across 230+ clubs
General support — policy questions, complaints, issues that need a human — with warm handoff
L.Ai.C.
Context
Life Time operates 230+ fitness clubs serving over a million active members — all of whom expect instant, accurate support at any hour. I led the UX discovery and conversation design for L·AI·C (pronounced "lay-see"), an AI assistant built natively into the Life Time app that handles member inquiries, books classes, and knows when to hand off to a human.
My Role: UX Designer
Scope: Discovery, research, conversation design, wireframes
Collaborators: Engineering, product strategy, visual design
Status: Live in the Life Time app
The Challenge
With 1M+ active members across 230+ locations, Life Time's support team was fielding thousands of daily inquiries — class schedules, membership questions, billing issues, facility hours. Most of it was repetitive. Peak hours created bottlenecks. And members expected answers at 11pm just as much as they did at 11am.
The support team was doing their best, but the math didn't work. You can't scale humans infinitely to meet 24/7 demand.
Discovery
I started with discovery: stakeholder interviews with product, engineering, and ops; an audit of existing member support channels; competitive analysis of how other AI tools handle similar contexts; and a deep dive into the edge cases — what happens when the bot doesn't know, gets it wrong, or needs a human.
The research surfaced four core journeys that covered the vast majority of member needs: class booking, account management, location info, and general support. Everything else was an edge case layered on top of those four.
Class booking — find available classes by type, time, club; confirm booking in one turn
Account management — pause, cancel, update billing; 80% self-serve without human involvement
Location info — real-time hours, pool status, amenities across 230+ clubs
General support — policy questions, complaints, issues that need a human — with warm handoff
Conversation design — every word is a decision
Conversational UX is different from screen UX. There's no layout to guide the eye, no button hierarchy to signal priority. The entire experience lives in the words and the order they appear. Every response has to do three things at once: confirm the AI understood, provide the answer, and make the next step obvious.
I designed the conversation flows for all four core journeys, mapping both the happy path and the edge cases — class full, intent unclear, conflicting information, wrong club. The hardest design problem wasn't the happy path. It was designing graceful fallbacks that didn't make members feel like they'd hit a wall.
Results
L·AI·C launched and scaled quickly. The numbers reflect what good conversation design does when paired with the right data infrastructure — it removes friction so completely that members don't think of it as using a chatbot. They just got their answer.
Conversation design — every word is a decision
Conversational UX is different from screen UX. There's no layout to guide the eye, no button hierarchy to signal priority. The entire experience lives in the words and the order they appear. Every response has to do three things at once: confirm the AI understood, provide the answer, and make the next step obvious.
I designed the conversation flows for all four core journeys, mapping both the happy path and the edge cases — class full, intent unclear, conflicting information, wrong club. The hardest design problem wasn't the happy path. It was designing graceful fallbacks that didn't make members feel like they'd hit a wall.
Discovery
I started with discovery: stakeholder interviews with product, engineering, and ops; an audit of existing member support channels; competitive analysis of how other AI tools handle similar contexts; and a deep dive into the edge cases — what happens when the bot doesn't know, gets it wrong, or needs a human.
The research surfaced four core journeys that covered the vast majority of member needs: class booking, account management, location info, and general support. Everything else was an edge case layered on top of those four.
Class booking — find available classes by type, time, club; confirm booking in one turn
Account management — pause, cancel, update billing; 80% self-serve without human involvement
Location info — real-time hours, pool status, amenities across 230+ clubs
General support — policy questions, complaints, issues that need a human — with warm handoff
L.Ai.C.
Context
Life Time operates 230+ fitness clubs serving over a million active members — all of whom expect instant, accurate support at any hour. I led the UX discovery and conversation design for L·AI·C (pronounced "lay-see"), an AI assistant built natively into the Life Time app that handles member inquiries, books classes, and knows when to hand off to a human.
My Role: UX Designer
Scope: Discovery, research, conversation design, wireframes
Collaborators: Engineering, product strategy, visual design
Status: Live in the Life Time app
The Challenge
With 1M+ active members across 230+ locations, Life Time's support team was fielding thousands of daily inquiries — class schedules, membership questions, billing issues, facility hours. Most of it was repetitive. Peak hours created bottlenecks. And members expected answers at 11pm just as much as they did at 11am.
The support team was doing their best, but the math didn't work. You can't scale humans infinitely to meet 24/7 demand.
L.Ai.C.
Context
Life Time operates 230+ fitness clubs serving over a million active members — all of whom expect instant, accurate support at any hour. I led the UX discovery and conversation design for L·AI·C (pronounced "lay-see"), an AI assistant built natively into the Life Time app that handles member inquiries, books classes, and knows when to hand off to a human.
My Role: UX Designer
Scope: Discovery, research, conversation design, wireframes
Collaborators: Engineering, product strategy, visual design
Status: Live in the Life Time app
The Challenge
With 1M+ active members across 230+ locations, Life Time's support team was fielding thousands of daily inquiries — class schedules, membership questions, billing issues, facility hours. Most of it was repetitive. Peak hours created bottlenecks. And members expected answers at 11pm just as much as they did at 11am.
The support team was doing their best, but the math didn't work. You can't scale humans infinitely to meet 24/7 demand.