Rethinking Job Search for Skilled Tradespeople

With a stale product and a backlog of unresolved research, one thing missing was an understanding of how competitors were serving candidates in the trades space. I mapped what they were doing across navigation, search, profile building, and job posting quality. I also used NotebookLM to collect and synthesize all existing research — asking questions of it quickly rather than combing through documents manually.

The competitive analysis wasn’t a separate effort — it was one of three inputs to a focused design sprint. The product team had identified the candidate experience as the area with the most friction — but not where in that experience. Before any design work could begin, we needed to find the seam. I ran a focused design sprint to locate it: structured activities, real constraints, a clear output. That sprint defined the problem space the rest of the project was built around. The full team — UX research, product, and engineering — reviewed the research together and dot voted on where we had the most open questions and risk. Three areas rose to the top.

Development had slowed around Thanksgiving, and the product manager and I saw it as the perfect time for discovery on the areas the team had flagged during the sprint.

For each area, I used Gemini to help conceive the research plan — iterating on it until the approach felt right, then building competing prototype pairs in Figma Make. Each pair tested a different hypothesis about what candidates actually needed. Where testing surfaced unclear findings, we iterated and ran another cycle. We put the prototypes in front of real tradespeople, watched how they responded, synthesized the findings, and presented back to the team. The prototypes themselves were more interactive than anything Figma alone could have handled cleanly — Figma Make made the complexity manageable. The three areas consumed about two and a half weeks — work that would typically stretch across a month or more.

Our bar for a meaningful signal wasn’t a perfect score — it was 80% of participants responding with genuine, unprompted enthusiasm. Prompted responses carry inherent social bias and tell you what people will tolerate. Unprompted ones tell you what they actually want.

One usability finding stood out. Early prototypes asked candidates to rate their years of experience directly alongside a skill selection — two questions in close proximity. Participants consistently glossed over the experience input. Separating them into sequential steps changed the behavior immediately.

One direction generated the strongest signal of the entire study. An AI-assisted profile builder produced 100% unprompted enthusiasm across every participant. That finding directly shaped the design direction that moved into development.