Case study · Conversation design
The Fabric Matchmaker
Designing a conversational AI that scales expert knowledge without losing the human touch.
Try the live prototypeHow it started
Selvedge & Bolts was closing its physical shop and moving fully online. The owner wanted to do “something with AI.” She just didn't know what.
I looked at the website, the social media, spoke to the owner and read her customers' comments. The shop had worked because the owner was in it. The website wasn't doing that job and a chatbot built on top of it wouldn't either.
So I started with the content. New product descriptions, a naming system, a voice framework to scale her expertise across the catalogue. Read that piece here.
Then the chatbot. One job: help customers find the right fabric for what they're making. Everything else followed from that.
What it refused to be
message deleted
An open door answers anything and owns nothing. One job, matching you to the right fabric, is what gave her focus and a voice.
Five versions deep
Where it broke, and what fixed it.
What broke
Early versions failed in specific ways. v2 asked two questions at once and overwhelmed people. v3 overcorrected into a rigid script that felt robotic. v4 handled a near-miss backwards, leading with what we didn't have instead of what we did.
What I got right
I tested across five iterations and let each failure name its own fix. One question per message. Loosen the script. Never open with the gap. By v5, testers across every experience level said they felt confident enough to buy.
The temptation
The easy build was a friendly bot that opens with “How can I help?” and answers anything.
The line I held
I refused the open door. One job, match the customer to a fabric, and a guardrail that redirects everything else. The focus is what made it convert instead of just deflect.
The problem
The expertise had nowhere to go.
In the shop, customers would walk in with a project. “I'm making a dress, what do you have?” The owner knew exactly what to say. That conversation was the product. Moving online meant losing it.
What customers asked in store
“What can I make with this?” “Will I need a lining?” “Is it easy to sew?”
What made the shop work
The way the owner talked about fabric. Her voice, her knowledge, her storytelling. Warm, generous, specific. That was the thing worth keeping.
The persona
The Fabric Matchmaker is the shop floor, personified.
Most retail bots have warm, generic human names. Friendly, but vague. This bot has a job to do and the name says it upfront. The Fabric Matchmaker tells you what she does before you've typed a word. It also tells you what she needs from you: your project, your context, your constraints.
Three traits define her voice.
Generous
She shares knowledge freely. She'll warn you that a fabric is slippery before you cut it. She'll tell you when a lining makes sense. Expertise as service, not gatekeeping.
Honest
She never oversells. If a fabric isn't right for your project, she won't pretend it is. “It's not a solid black, but honestly it reads more sophisticated.”
Focused
One job: matching you with the right fabric for what you're making. Everything else, she redirects.
How she speaks
Short, sensory sentences. “This moves like water.” “Cool against the skin.” “Born to be a bias-cut slip dress.” She never opens with “How can I help?” She opens with “Hey, what are you making?” Always.
What she won't do
Pretend. Oversell. Gatekeep. Get distracted.
The conversation design
Ask before recommending.
The first thing the Fabric Matchmaker asks is “What are you making?” Not “How can I help?” Not “What are you looking for?” The conversation starts focused and stays focused. By the time a fabric is named, the customer has already been heard.
The opening question. Focused from the first word.
Evocative, not clinical. “Garden party romantic, or farmer's market crisp?”
Narrows the brief without narrowing the options.
Context that changes everything. A garden party dress and a work dress are different conversations.
Only asked when relevant. Never makes the customer feel like a beginner.
The recommendation rule: connect back to what they said before naming the fabric. Context first. Product follows.
The voice framework
Four pillars. One consistent voice.
The product descriptions had a voice framework. The bot uses the same one. But conversation has different rules. You can't lead with texture for three sentences when someone is waiting for an answer. The framework had to work in a single breath.
Tactile · Make them feel it without touching it
“This has a water-like drape that designers kill for. It's cool against the skin.”
In Motion · How it lives when you wear it
“Moves with a fluid, expensive swing.”
Creative Spark · Show them what it could become
“Born to be a bias-cut slip dress. Or an oversized French-tuck shirt.”
The Expert Friend · The shop floor knowledge, now in the copy
“A word of warning: this fabric is slippery to cut. Take your time, use weights, not pins, and you'll be fine.”
The recommendation above shows all four working together. Tactile, in motion, a specific moment, and the expert detail. In a single breath.
The conversation in action
A complete journey, start to finish.
A user came looking for fabric for a wedding guest outfit. The bot asked the right questions, built context, checked experience level naturally, and landed on a specific recommendation with a product link. This is the final version of the prototype doing its job.
Note: the bot was originally built in Voiceflow and has since been rebuilt using Claude.
What broke and how I fixed it
Five versions. Each one broke differently.
Tested with 5 to 6 users across five iterations. Each version surfaced a specific design failure.
Asked two questions at once. Users got overwhelmed. Fixed: one question per message, always.
Overcorrected into a rigid script. Felt robotic. Fixed: loosened the structure. Let the conversation breathe.
Near-miss handling was backwards. Led with the gap, not the product. Fixed: never open with what you don't have. Lead with what you do.
Finally right. Users across all experience levels said they felt confident enough to buy.
v4. Wrong
I don't have a true black fabric in the current collection, but...v5. Right
The Monochrome Garden stretch cotton is doing something interesting for that. Tiny black and white florals that read graphic, not sweet. It's not a solid black, but honestly it reads more sophisticated...The guardrails
Focused by design.
The Fabric Matchmaker has one job: match you with the right fabric for what you're making. Every conversation starts there. “Hey, what are you making?” Not “How can I help?” because “How can I help?” invites anything. That's the ChatGPT-fication of a bot: open door, no focus, no identity.
She's not the Google of sewing. She won't answer general questions about fabric care, clothing construction, or the brand. She doesn't get distracted.
When users push off course, she redirects. When a recommended fabric isn't available, she stays in voice, holds onto the original brief, and comes back with a new recommendation. The guardrail isn't a limitation. It's what makes the rest of it work.
What I delivered
A live, working system.
Voice framework built to scale the owner's expertise into every conversation
Conversation system built around context-building, not quick recommendation
Product briefing notes structured for AI use. Facts only, no pre-written copy.
System prompt iterated across five versions, each addressing a specific design failure
Working prototype, live and publicly accessible at a permanent URL
Content guidelines for ongoing use as new stock arrives
By v5, every participant across all experience levels said they felt confident enough to buy. Quantitative outcomes, including conversion rate, time on page, and click-through from bot to product, will be added once the site relaunches. Tracking implemented, awaiting sufficient traffic post-relaunch.
What testers said
Occasional Maker
Professional Sewist
Browser
Why this isn't just a fancy FAQ
A FAQ answers questions customers already know to ask.
This bot helps customers figure out what they need. It doesn't add to cart. It doesn't need to. It makes people confident enough to buy. That's the difference between deflection and conversion.