Skip to content

Evaluation modes

An Agentic Eyes run is scored against a question. The evaluation mode is that question. Four modes exist because “did the visit go well” means four different things depending on what the site is for — and silently scoring research against a conversion bar produces confidently wrong reports.

Mode Question it scores Use for
Conversion Did the persona reach a conversion action? Marketing sites, signup funnels, e-commerce
Pre-signup intent Did the persona develop intent to sign up — without any credential steps? Product sites where the sale happens later
Information discovery Did the persona find and understand the information they came for? Documentation, knowledge bases, research pages
Authenticated task Did the persona complete a task in the logged-in product? App onboarding, core workflows

The modes guard against two specific scoring failures:

  • A blocked signup is not a failed research visit. A persona researching your product who read everything and left informed succeeded under information discovery — and “failed” under conversion. Same journey, opposite verdicts, both defensible only under their own mode.
  • Pre-signup intent exists because personas don’t have credentials. Agentic Eyes never fills login forms with invented data, so a conversion-barred run against a gated product measures only the gate. Pre-signup intent scores the real question for growth sites: did the visit build intent to come back and sign up?

Omitting the mode defaults to conversion — the historical behavior. That default is flagged in run output precisely so a research workflow never silently reports zero conversions. Pass the mode explicitly on every run; it costs one parameter and removes the single most common misreading of Agentic Eyes reports.

When a journey spans modes — discover the product, then sign up, then activate — run each phase under its own mode and read the sequence: discovery success feeding intent, intent feeding conversion. That funnel view is more actionable than one blended score.