Collective IQ

Organization: Microsoft
IC3 · Microsoft 365 Copilot
Role: Design lead
Sole designer
Collaborators: Zheng Ni — engineering
Legal and compliance
Product
Duration: 2026 – Present
Private preview · NDA

Collective IQ is a 0-to-1 AI knowledge product built on the Microsoft 365 stack. People build up a lot of context in meetings — who talked to an account, what got promised, what's still open — and it stays with whoever was in the room. This puts it somewhere the rest of the organization, and Copilot, can reach it.

I'm the design lead and the only designer on it. I proposed six full design approaches across different surfaces and modalities. One went to MVP and the rest are feeding phases 2 and 3. The interaction model is mine end to end.

On paper it's a sharing feature. Building the button was never the hard part; getting anyone to press it was. Nobody hands their notes to the wider org unless they're confident it won't come back on them later, and the org can't stand behind what Copilot surfaces unless a person vouched for it. Every screen here is working on that problem.

What I did

Interaction model: Took it from a loose strategic direction to the contributor experience that shipped. Six approaches proposed, one taken to MVP, the rest carried into later phases.
Contributor experience: Designed how someone shares what they know without exposing what was said — the summary, the preview before anything moves, and scoping to specific people.
Trust and governance: Shaped the sharing model, the opt-out mechanics, and the audit surfaces, with legal in the room throughout for card wording, states, and when consent fires.
Management platform: Designed the web surface behind the feature: the contribution queue, review states, expiring deadlines, and bulk controls for whoever administers it.
Cross-surface integration: Made the same contribution model work in Teams, Outlook, and SharePoint, and structured it to survive the move past meetings into email, channels, and chats.
Direction under uncertainty: Designed on top of a product definition still being argued out: what shipped in MVP, who owns the sharing action, how topics work. I sat in on the validation sessions that pointed the way.

Selected work

The Collective IQ card in the Teams feed after a meeting — The feed cardAfter a meeting, an AI summary appears and asks whether to contribute it. The whole product turns on this one moment.

Reviewing the AI summary before sharing it — Preview before shareYou read what's about to move before it moves. A conversation is full of half-formed thinking, so what gets vouched for is the summary, not the transcript.

Scoping a contribution to specific teammates — Scoped sharingNarrow a contribution to specific teammates instead of the whole org. Too many knobs and nobody contributes; too few and nobody trusts it with anything real.

Declining to contribute a summary — Opting outDeclining is easy to find and easy to do. The wording, the states, and when the prompt fires were legal decisions as much as design ones.

The contribution review queue in the web management platform — Contribution queueThe management platform behind the feature: summaries across accounts with pending review, expiring deadlines, and bulk action.

The same contribution model appearing in SharePoint — Beyond TeamsThe same contribution model in SharePoint. Knowledge gets made in one place, shown in another, and pulled back much later somewhere else entirely.

Where it stands

In private preview with hundreds of internal users, expanding to thousands. We started with sales because the gap is so obvious there: before a meeting that matters, the answers usually already exist, trapped in someone else's meeting history.

The product is unreleased and under NDA, so there are no published results yet. The full case study covers the reasoning behind the model.

0→1

No spec, no design system, and no prior design history when I joined as design lead

Design approaches proposed; one shipped to MVP, the others feeding phases 2 and 3

100s

Internal users in private preview, expanding to thousands