How might we redesign vehicle capture to be faster and more forgiving than a 20-photo checklist, without breaking down in a driveway that's too tight or a lot that's too big — for two users who couldn't be more different in how often they do this?

Video loosened the angle requirement, but it didn't remove distance from the equation. Users still needed enough space between themselves and the car. On top of that, this would ship across multiple white-labeled clients, each with their own workflow expectations, and to two user groups with almost opposite needs: dealers doing this daily, consumers doing it once or twice in their life.

I started by meeting with the PM and engineering manager building the underlying scan technology — what we were building, what "working" would actually mean, and how we'd realistically test it once it existed.

From there, I brought in the team that owned OEM tooling — the dealer-facing flow where a vehicle gets photographed for valuation right after a trade-in, before it goes up as a listing. They wanted video specifically to cut down the time that process took. We built the user workflow and journey together before any screens existed. External clients came through the same process afterward — each one had their own consumer-facing flow for photographing a vehicle's condition at trade-in or lease-end, and each wanted video for the same reasons: efficiency, ease of use, less time spent. Rather than starting from scratch with every client, I took the workflow built with OEM tooling and adapted it with each one.

Testing this kind of tool the way we normally would wasn't an option, either. AI-based video capture can't be validated online. The only way to know if it works is to point a phone at an actual car in an actual driveway. That shaped the approach from the start: design early, get it into a small group's hands, and keep iterating in the field once it's live.

Once I had an initial design, I ran it through review with the AI tool team and OEM team, and through feedback sessions with each external client, iterating the design with all of them before moving quickly into build — so the teams who'd be testing it could get their hands on it as soon as possible.

Most inspections happen outdoors — glare, harsh light, reflections on the car's surface — so accessibility came first: multiple aspect ratios to adapt to different framing conditions, and clear, unmistakable CTAs for record, pause, and stop. A circular indicator showed scanning progress and camera angle in real time, so users could see the system was actively tracking their movement; if someone moved too fast or held the wrong distance, a message appeared at the top of the screen — visible, but never blocking their view of the car.

Dealers use this daily; consumers might do it once or twice in their life. Rather than building two separate experiences, guidance density adjusted to context — visible enough for a first-timer, unobtrusive enough not to slow down someone who's done this a hundred times.

Qualitatively, the tool shipped across multiple businesses under a single white-labeled framework, and the real-time guidance system became a reference point for other capture-based features on the team.

How might we redesign vehicle capture to be faster and more forgiving than a 20-photo checklist, without breaking down in a driveway that's too tight or a lot that's too big — for two users who couldn't be more different in how often they do this?

Video loosened the angle requirement, but it didn't remove distance from the equation. Users still needed enough space between themselves and the car. On top of that, this would ship across multiple white-labeled clients, each with their own workflow expectations, and to two user groups with almost opposite needs: dealers doing this daily, consumers doing it once or twice in their life.

I started by meeting with the PM and engineering manager building the underlying scan technology — what we were building, what "working" would actually mean, and how we'd realistically test it once it existed.

From there, I brought in the team that owned OEM tooling — the dealer-facing flow where a vehicle gets photographed for valuation right after a trade-in, before it goes up as a listing. They wanted video specifically to cut down the time that process took. We built the user workflow and journey together before any screens existed. External clients came through the same process afterward — each one had their own consumer-facing flow for photographing a vehicle's condition at trade-in or lease-end, and each wanted video for the same reasons: efficiency, ease of use, less time spent. Rather than starting from scratch with every client, I took the workflow built with OEM tooling and adapted it with each one.

Testing this kind of tool the way we normally would wasn't an option, either. AI-based video capture can't be validated online. The only way to know if it works is to point a phone at an actual car in an actual driveway. That shaped the approach from the start: design early, get it into a small group's hands, and keep iterating in the field once it's live.

Once I had an initial design, I ran it through review with the AI tool team and OEM team, and through feedback sessions with each external client, iterating the design with all of them before moving quickly into build — so the teams who'd be testing it could get their hands on it as soon as possible.

Most inspections happen outdoors — glare, harsh light, reflections on the car's surface — so accessibility came first: multiple aspect ratios to adapt to different framing conditions, and clear, unmistakable CTAs for record, pause, and stop. A circular indicator showed scanning progress and camera angle in real time, so users could see the system was actively tracking their movement; if someone moved too fast or held the wrong distance, a message appeared at the top of the screen — visible, but never blocking their view of the car.

Dealers use this daily; consumers might do it once or twice in their life. Rather than building two separate experiences, guidance density adjusted to context — visible enough for a first-timer, unobtrusive enough not to slow down someone who's done this a hundred times.