Product Management
chatcollab
Why Teams Kept Screenshotting Their AI Chats
Sequencing the roadmap for a real-time, multi-user collaborative workspace inside ChatGPT
Why Teams Kept Screenshotting Their AI Chats
Sequencing the roadmap for a real-time, multi-user collaborative workspace inside ChatGPT
Timeline
Timeline
5 weeks
5 weeks
team
Team
Team of 4
Team of 4
frameworks
Frameworks
Lean Canvas
JTBD
PRD
Kano Model
Lean Canvas
JTBD
PRD
Kano Model
The build, start to finish:
Opportunity hypothesis → market validation → Lean Canvas & SWOT → competitive analysis → Kano-prioritized PRD → roadmap → resourcing → GTM, UVP & KPIs
The build, start to finish:
Opportunity hypothesis → market validation → Lean Canvas & SWOT → competitive analysis → Kano-prioritized PRD → roadmap → resourcing → GTM, UVP & KPIs
Overview
ChatGPT is where teams already do their thinking, but it was never built to hold more than one thread. We felt this ourselves before we researched it:
23 screenshots and GPT links traded across Slack on one project,
because there was no way to show a teammate what the AI had already told you. That number became the opportunity hypothesis.
ChatGPT is where teams already do their thinking, but it was never built to hold more than one thread. We felt this ourselves before we researched it:
23 screenshots and GPT links traded across Slack on one project,
because there was no way to show a teammate what the AI had already told you. That number became the opportunity hypothesis.
ChatGPT is where teams already do their thinking, but it was never built to hold more than one thread. We felt this ourselves before we researched it:
23 screenshots and GPT links traded across Slack on one project,
because there was no way to show a teammate what the AI had already told you. That number became the opportunity hypothesis.
fuzzy Problem
HMW let a team think inside ChatGPT together, without everyone ending up in their own disconnected thread?
HMW let a team think inside ChatGPT together, without everyone ending up in their own disconnected thread?
Every team we looked at, ours included, wasn't missing a feature so much as a floor: nobody could see who else was even in the conversation. You can't build collaboration on top of that.
The fragmentation wasn't a UI problem, it was a foundation problem -> and that distinction changed the order we built in.
Every team we looked at, ours included, wasn't missing a feature so much as a floor: nobody could see who else was even in the conversation. You can't build collaboration on top of that.
The fragmentation wasn't a UI problem, it was a foundation problem -> and that distinction changed the order we built in.
Research: exploring product direction
Community sentiment backed it, teams outside ours were saying the same thing:
Community sentiment backed it, teams outside ours were saying the same thing:
"I'm beyond disappointed... 'collaboration' means all team members can interact with all chats. I can't believe this isn't a feature."
— Reddit user, on ChatGPT
"I'm beyond disappointed... 'collaboration' means all team members can interact with all chats. I can't believe this isn't a feature."
— Reddit user, on ChatGPT
"I'm beyond disappointed... 'collaboration' means all team members can interact with all chats. I can't believe this isn't a feature."
— Reddit user, on ChatGPT
Market data did too: 62% of organizations were already experimenting with AI agents.
Market data did too: 62% of organizations were already experimenting with AI agents.
We pressure tested differentiation with a Lean Canvas and SWOT, then ran it against the field: NotebookLM, Gemini, Claude, Copilot. Nobody offered live multi user chat, co-creation or authorship attribution. That gap is what we scoped against.
We pressure tested differentiation with a Lean Canvas and SWOT, then ran it against the field — NotebookLM, Gemini, Claude, Copilot. Nobody offered live multi-user chat, co-creation, or authorship attribution. That gap is what we scoped against.
Deciding what to build, and in what order
Deciding what to build, and in what order
Our first roadmap draft led with async collaboration - comments, mentions, the features that demo well. Scoping it exposed a dependency we'd missed:
You can't comment on a thread you don't know exists.
We rebuilt the sequence - presence and access first, then threading, async pushed back.
From there, I ran the feature set through a Kano-based prioritization model to separate what teams would expect by default from what would actually differentiate ChatCollab, then scoped it into the PRD
Our first roadmap draft led with async collaboration - comments, mentions, the features that demo well. Scoping it exposed a dependency we'd missed:
You can't comment on a thread you don't know exists.
We rebuilt the sequence - presence and access first, then threading, async pushed back.
From there, I ran the feature set through a Kano-based prioritization model to separate what teams would expect by default from what would actually differentiate ChatCollab, then scoped it into the PRD
Our first roadmap draft led with async collaboration - comments, mentions, the features that demo well. Scoping it exposed a dependency we'd missed:
You can't comment on a thread you don't know exists.
We rebuilt the sequence - presence and access first, then threading, async pushed back.
From there, I ran the feature set through a Kano-based prioritization model to separate what teams would expect by default from what would actually differentiate ChatCollab, then scoped it into the PRD
Our Solution
Our solution, ChatCollab, introduces a shared ChatGPT workspace with structured collaboration built directly into the chat experience.
Its core capabilities include:
Our solution, ChatCollab, introduces a shared ChatGPT workspace with structured collaboration built directly into the chat experience.
Its core capabilities include:
Multi user shared conversations
Multi user shared conversations
Threads & automatic branching to support parallel thinking
Threads & automatic branching to support parallel thinking
Comments, mentions and AI-assisted revisions
Comments, mentions and AI-assisted revisions
Role-based permissions (view, comment, edit)
Role-based permissions (view, comment, edit)
Project organization (folders, search, filters)
Project organization (folders, search, filters)
Presence indicators + follow mode
Presence indicators + follow mode
Sequencing the roadmap
The rebuilt sequence, resourced and scoped for a phased release: presence and access first, then threading, then workspace organization, async last.
The rebuilt sequence, resourced and scoped for a phased release: presence and access first, then threading, then workspace organization, async last.
Getting it ready to ship
GTM moved through four phases - internal dogfooding, a 50-account enterprise alpha, a broader beta, then market launch. Positioning centered on one line: the only ChatGPT experience built for teams, not individuals.
GTM moved through four phases - internal dogfooding, a 50-account enterprise alpha, a broader beta, then market launch. Positioning centered on one line: the only ChatGPT experience built for teams, not individuals.
Success metrics tied directly back to the fragmentation problem: a 20% increase in shared workspace creation, a 15% reduction in churn, and a 10% increase in team retention — the opposite of what 23 screenshots in one Slack channel looks like.
Success metrics tied directly back to the fragmentation problem: a 20% increase in shared workspace creation, a 15% reduction in churn, and a 10% increase in team retention — the opposite of what 23 screenshots in one Slack channel looks like.
reflections
The least exciting line on our roadmap - can you see who else is here, turned out to be the one everything else depended on. I didn't believe that until we tried to skip it and had to walk the sequence back. Kano gave me a way to separate "table stakes" from "differentiator" instead of relying on gut feel, and that's the part of this process I'd reach for again on any roadmap, not just this one.
The least exciting line on our roadmap - can you see who else is here, turned out to be the one everything else depended on. I didn't believe that until we tried to skip it and had to walk the sequence back. Kano gave me a way to separate "table stakes" from "differentiator" instead of relying on gut feel, and that's the part of this process I'd reach for again on any roadmap, not just this one.
