Designed to survive whatever the business decided.

Cisco's Collaboration Technology Group asked us to swing at the next thing — an AI-powered voice assistant for meeting rooms, anchoring the company's first IoT SaaS offering. The work pivoted hard when the business decided not to ship the SMB device — but what we'd built survived it. Webex Assistant shipped on enterprise video devices with Proactive Join as the headline feature, which became the most-used capability in the Webex Assistant set.

Frontier AI Voice Interaction Methodology

Role

Director of Design, Cisco UC and Devices

Platform

Cisco Collaboration — frontier AI for meeting rooms

Scope

First IoT SaaS bet anchored by voice AI; pivoted to enterprise video

Team

Five designers and researchers plus manager

Duration

2017–2019

The bet

Find what comes next for meeting rooms. The project's mandate was to build Cisco's first IoT SaaS offering — anchored by an AI-powered voice assistant — for SMB meeting rooms. The bar wasn't "build a voice assistant." It was "find what comes next."

Rendered meeting-room scene: a wall display shows a 'Design All Hands' meeting card with the Webex Assistant listening, prompting 'Hi Jane, would you like to start Design All Hands?' in a speech bubble. A sketched voice device sits on the conference table among chairs. — The opening framing — what comes next for the meeting room.

The methodology

This was a frontier-tech bet inside a company that didn't have the SMB go-to-market machinery to back it. The way to run a project like that was to avoid committing early to a form factor, and instead make every piece of the work survive whatever the business ultimately decided.

A small team — five people plus a manager — running design sprints in weeks rather than quarters. Cardboard prototypes for the device itself. User testing in real meeting-room scenarios before any industrial design or hardware architecture had been committed. Hacked-up software prototypes to validate the voice interaction grammar. Diverge-and-converge methodology: go wide on use cases up front — eleven distinct ones — then converge on what user research and the technology actually supported.

Usability research session: two participants seated at a table testing an early cardboard-and-foam prototype of the voice device, which glows blue on the tabletop beside a laptop. Both participants' faces are blurred for privacy. An overlay caption reads 'How does it know I'm here? I wanna play around with it…' — Cardboard prototypes — the research artifact that defined the methodology.

If the experience worked in cardboard, we could find a hardware path. If the experience didn't work, no amount of industrial design was going to save it.

The design sprint method shown as a four-stage loop — Idea, Build, Launch, Learn — cycling clockwise, with a filled accent lens between Learn and Idea marking the rapid prototyping and testing back-and-forth. Below the loop, four process photographs ground the theory in practice: card sorting after stakeholder interviews, parallel remote and in-room sketching, cardboard prototyping before any hardware, and testing with real users. — The sprint methodology — go wide, converge fast, validate cheaply.

When the business pivoted

Eighteen months into the work, the business made a decision: Cisco didn't have the go-to-market machinery to compete in SMB through a nascent IoT SaaS channel. The audio-only standalone device wouldn't ship. Resources would refocus on enterprise.

The work had been architected to survive whichever outcome materialized. When the business decided, the design team didn't have to start over. We had a clear pivot path because the methodology had assumed exactly this kind of pivot was possible from day one. The interaction grammar held. The cardboard-validated experience model held.

What we shipped

Proactive Join: walk into your booked room, the device recognizes you, the assistant offers to join the meeting, you say "yes," the meeting joins. The whole interaction takes seconds and removes the typical 10-minute meeting-start delay.

The "say yes to join" interaction pattern came from observing how users actually responded to early prototypes. Confirmation was the right interaction primitive — not "tap to join," which adds a manual step, and not "the meeting auto-joins," which removes user agency. Confirmation gave the user control while keeping the manual overhead near zero.

The shipped Proactive Join interaction on a Webex video device: a meeting card reads 'User Experience Review, 10:00–11:30, organized by Brenda Song,' with a touch-panel join affordance and a prompt to join from the touch panel. Below, the Webex Assistant hexagon asks 'Hey Jane, would you like to join this meeting?' — Proactive Join — the survived-the-pivot artifact that became the #1 Webex Assistant feature.

Most-used Webex Assistant feature

Use cases explored, before convergence

18mo

From kickoff to business pivot

The design decision I'm proudest of from this project isn't a feature or a flow. It's how the team operated. Empowering a small group of designers and researchers to run real design sprints with hacked-up prototypes in weeks, validate interaction models before hardware commitments, and ship something real from a project that pivoted — that was the actual deliverable.

Why this matters

The principle wasn't "make the work bulletproof to cancellation" — it was "make the work valuable independent of any specific outcome." The job of a design leader on a frontier-tech bet is to architect the work so that all outcomes — canceled product, launched product, pivot, multi-year capability layer — deliver real value.

Next case study