Co-Design: Real-Time Concept Iteration, Built Into the Interview

How it works

During an interview, a participant reacts to an image stimulus, a concept, a package, an ad, a screen, and describes what they'd change. Outset generates a revised version on the spot. The participant reacts to that version, and can ask for another change if it's still not right, all without leaving the session.

The researcher leaves with a revised artifact, not a list of complaints to interpret after the fact.

The latest generation of AI image models finally got fast enough to keep pace with a live conversation. Co-Design is built on a model layer designed to always run on whichever backend is fastest, so that stays true as the underlying models keep improving.

Where teams are using it

Concept and packaging testing. A product concept, a package design, an ad creative. Participants show you the direction they'd actually take it in, not just what missed.

Message and ad testing. Headlines, claims, and framing, tested and revised in the same round instead of a re-fielded follow-up.

Landing page testing. Hero sections, calls to action, trust cues. Participants reshape the layout instead of describing it in the abstract.

Single-screen product feedback. A dashboard, a settings screen, a form. Co-design works on one static screen at a time today, so the best fit is a single layout or hierarchy question, not a multi-step flow.

What it saves

Co-Design is a new capability so we’re still quantifying its impact, but the shape of the savings is clear: a design change that used to require a re-fielded round or a new participant now happens inside the session that's already running. No new fieldwork. No second round of recruiting. The iteration and the validation happen in the same sitting, not weeks apart.

Reporting that shows the whole journey, not just the final answer

A single co-designed image is a session artifact. The real value shows up when you can see the pattern across every participant who worked through the same question.

Every Co-Design turn is captured in full: the original stimulus, the participant's edit request, the version Outset generated, and their reaction to it, preserved as a version history you can step through, not just a before-and-after.

Across a study, that becomes a report that surfaces the themes participants pulled a concept toward, ranks the directions that came up most, and pulls out the probing-question responses that explain why.

You're not left sorting through a folder of individual images by hand. You get the common thread across dozens of divergent, personal edits, plus the verbatim reasoning behind them, in one place.

What this means

Outset was first to bring AI moderation into qualitative research. Co-Design is the next chapter of that same idea: research that moves at the speed of the models underneath it, without taking the researcher, or the participant, out of the loop.

As consumer behavior keeps compressing into shorter and shorter cycles, the tools research teams use have to compress with it.