Test the prototype before you build the product
Experience Intelligence Studio is an AI-assisted usability testing MVP. A designer or product manager uploads a runnable HTML prototype, defines one task, shares a link and reviews how participants performed.
Why I built it
I could build interactive prototypes quickly with AI, but learning whether people could actually use them still meant coordinating a study and piecing together feedback. I wanted a lightweight way to answer one question before engineering started: is this interaction ready to build, or should we change it first?
Start research without buying a testing license
The pilot was designed to run on free-tier infrastructure with participants the team recruits itself. Participants open a link without creating an account. The study captures task outcomes, interaction events and feedback in one place. Zero cost refers to pilot software spend within free-tier limits; recruiting, incentives, team time and scaling can still cost money.
My role: Product framing, UX strategy, requirements, prioritization and AI-assisted implementation.
One study: Can a support lead find the interaction that needs review?
01 · Upload the prototype. Add a standalone HTML file, or a ZIP with index.html and its assets. Preview the working support dashboard before sharing it.
02 · Create and share the test. Ask the participant to find a low-scoring interaction and open its evaluation. Define what counts as success, select the metrics and share the participant link.
03 · Capture the attempt. The participant consents, tries the task, and completes it or reports that they could not. Optional feedback adds context to the recorded activity.
04 · Review the dashboard and decide. Check task outcomes, then open individual sessions to see where people slowed down or took a different path. Change the design and retest the same task.
The screenshots show the working MVP with demo data. The diagram follows the same researcher and participant journey.
Read the results: what each metric actually measures
Task success rate = successful sessions ÷ all started sessions × 100. Check how success was detected and distinguish unfinished attempts from confirmed failures.
Average successful time = total recorded duration of successful sessions ÷ successful sessions. This MVP timer includes feedback time, so it is directional and should not be used to claim faster task performance.
Average clicks = total captured clicks ÷ all started sessions. Compare it with the expected path and inspect sessions before treating extra clicks as friction.
Hesitation and possible misclicks are counts of signals: a pointer pause of at least 1.2 seconds, or a click with no detected supporting event in the next two seconds. Reading, exploration or delayed feedback can trigger them too.
Ease and confidence are averages of submitted ratings: ease on a 1–7 scale and confidence on a 1–5 scale. Check how many people responded and compare ratings with actual completion.
Session replay and maps help explain the totals by reconstructing recorded events and showing where clicks or pauses occurred. Replay is event-based, not a screen recording.
Example: make one decision before engineering starts
Suppose 4 of 5 people open the correct evaluation (80% success), yet several hesitate before choosing a row and report low confidence. Inspect the sessions for a recurring pattern. If people consistently miss the same action, make the “Review evaluation” button more prominent and retest before handing off to engineering. This is an illustrative decision, not a pilot finding.
Business value and next steps
Value hypothesis: If the team can set up a study faster without weakening its quality, it can test earlier and catch avoidable rework before implementation. Compare setup time, launch errors and decision usefulness with the team’s existing workflow. Adoption, savings and ROI have not been measured.
Priority 1: Verify event capture, separate task time from feedback time, and test recovery and data retention. Priority 2: Pilot with designers and PMs who bring their own participants; measure setup time, unassisted publishing and time to the first evidence-backed finding. Priority 3: Explore editable persona and Job-to-be-Done scenario drafts, with researcher review before publishing.