I’m designing a low-fidelity research prototype for a tool that compares current behavior with an expected pattern and helps security or operations teams investigate deviations. Early research suggests “normal” varies by task, and people need the reference period, source coverage, change context, and a clear next step before trusting a finding. How would you test whether this deserves its own investigation surface or belongs within an existing workflow? What decision should participants make in the prototype?
I'd start by keeping the finding fixed and varying where it lives. Build two low-fi versions of the same deviation: one shown inside the tool the team already uses (an alert row, a panel next to the ticket), and one as a dedicated investigation surface. If you only prototype the new surface, you'll mostly learn that people can use it. When you start with the solution, you may never find something that works better, and that includes the solution of "this needs its own screen."
Before you test anything, figure out how often these investigations happen and how long they take. Ask whether a case happens once, twice, or more, and ask both the PM and the dev or QA lead, because you'll likely get different answers. Rare, quick checks usually belong in the existing workflow. Frequent investigations that pull together multiple sources over time are a better case for a separate surface. Keep in mind that people don't hate change, they hate change that takes away their current knowledge. A new surface has to earn the cost of leaving the place they already know.
So what decision should participants make? Have them make a triage call on each finding: dismiss it as expected behavior, escalate it as a real issue, or hold it because they need more evidence (and say what evidence). Then have them rate their confidence and explain why. That one decision tests everything your research surfaced. Did they check the reference period? Did they notice missing source coverage? Did they connect the deviation to a recent change? Did the suggested next step help them act or get ignored?
Seed the cases deliberately. Include one real anomaly, one deviation explained by a recent deploy or config change, one where a data source dropped out, and one where "normal" depends on the task. If people escalate the deploy-driven case or trust the finding with the coverage gap, the trust cues aren't doing their job, wherever they live. We used a similar structure in an internal security investigation concept, where each case laid out observed activity, possible cause, and a next check ahead of human review:
Watch where people go to get context. Observational testing shows you where the issues are, and user diaries over time help you find where people fall back to previous tools or processes. If participants using the dedicated surface keep jumping back to their logs or change tracker, it probably belongs inside that workflow. If people in the embedded version keep wishing for more room to compare periods and sources, that's your signal for a separate surface. And users don't care what data lives in which tool, but they do care about too many clicks on tasks they do regularly.
One more thing to plan for: you'll probably see a split in how much reasoning people want to see. With agentic tools, some users want the interface to focus on results while others say seeing the work is essential for checking it. Over time we've leaned toward rolling it up and making it accessible when you need it. So show the reference period, coverage, and change context in summary, and let people expand them.
Put simply, judge each version by whether people make the right triage call quickly and confidently, and let that decide where the investigation lives.