AI can crunch the usability data, but a researcher still needs to own the findings.

The analytical structure (code themes, compare groups, prioritize) is repeatable, but each study brings unique participant language, product-specific flows, and context that shifts how themes are drawn. It's not a one-size-fits-all pipeline.

Success criteria are underspecified: what counts as a valid theme cluster, how severity is weighted, and what 'annotated wireframe recommendations' means are all judgment calls. An agent cannot reliably know when the work is done to the client's standard.

The raw notes and metrics exist and can be fed to an agent, but wireframe tools (Figma, etc.) and the actual app screens are likely not in scope, making the annotation deliverable hard to fully automate. Structured metric data is accessible; unstructured notes require careful ingestion.

Misclassified themes or wrong severity rankings could send the design team down the wrong redesign path, wasting sprint cycles and potentially shipping a worse product. In fintech, poor UX in authentication or transaction flows has real user trust and compliance consequences.

Deciding which usability failures are truly severe versus merely frequent, and translating findings into wireframe-level recommendations, requires domain knowledge of fintech UX conventions, understanding of the startup's constraints, and researcher credibility with the client team.