Metrics question

What metrics would prove EVI's emotional intelligence adds real value?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Tests metrics design for proving a novel, hard to quantify capability actually delivers real value over a non emotional baseline.

Define the comparison clearly: outcomes with EVI's emotional intelligence enabled versus the same task handled by a standard, non emotion aware voice agent.
Choose outcome metrics tied to the use case: for customer service, that could be resolution rate, escalation rate, and customer satisfaction, compared head to head between the two conditions.
Add a perception metric: whether users report feeling understood or heard, since that is the direct value proposition of emotional intelligence, not just task completion.
Run a controlled comparison, ideally an A/B test with similar users and tasks split between emotion aware and standard versions, to isolate the effect of emotional intelligence specifically.
Watch for false positives: a user rating high satisfaction because of general product quality rather than the emotional intelligence feature specifically, and try to isolate that signal.
Confirm with the interviewer whether the target use case is customer service, mental health adjacent support, or something else, since the outcome metrics differ.

Proposes a direct controlled comparison against a non emotional baseline, which is the only rigorous way to isolate whether emotional intelligence specifically adds value.
Includes a perception metric like feeling heard alongside task metrics like resolution rate, capturing both the functional and experiential value emotional intelligence claims to add.
Explicitly plans to guard against attributing general product quality improvements to the emotional intelligence feature specifically.
Ties the metric set to a specific use case rather than proposing a generic, one size fits all measure of emotional value.

Proposing only a general satisfaction score without a controlled comparison against a non emotional baseline.
Ignoring the risk of attributing unrelated quality improvements to the emotional intelligence feature specifically.
No use case specificity, making the metric too abstract to actually act on.

How would you design a fair A/B test between emotion aware and standard voice agents?
What would you do if resolution rate improved but users did not report feeling more understood?
How would you measure this differently for a mental health adjacent use case?