What's actually wrong with the two agents right now

Users say they're not interested, the price put them off, or they flatly reject the call — but the sentiment data shows friction building before the call actually ends, suggesting the agent doesn't disengage on the first clear "no".

The agent re-asks fields the user already answered in this same call or a prior one ("Erratic Profile Collection"), one call hit a 0.9 frustration score over profile completion, and income/other fields come back wrong. This reads as a state-sync bug, not a tone problem.

Mostly mild friction — a user asks to be called back later, or is unavailable — but a few calls re-dial sooner than requested or interrupt an inconvenient moment.

Users ask mid-call to delete their profile or cancel, and the qualification script keeps going instead of routing straight to an exit.

This group includes a caller making financial demands and threats, a suspected scam/fake-account attempt, an aggressive user, an emergency-sounding termination, and one abusive caller that hit the absolute floor on both sentiment and frustration (-1.0 / 1.0).

Users directly ask if they're talking to an AI and the agent dodges or denies it; separately, two users outright refuse to continue once they suspect it's automated.

One Hindi-language recognition loop causing repeated mishearing, one caller disputing their own name on file (likely a CRM data-quality issue upstream), one off-topic question the agent handled awkwardly, and one correctly-terminated call with a minor.

This is the single biggest issue across both agents. 28 separate negative conversations are each a fresh "please delete my account/profile" request. The repetition across so many distinct conversations is the real signal: it strongly suggests earlier deletion requests from the same users aren't actually completing, so people come back and ask again, more frustrated each time.

Setup and verification prompts stall users before they ever see a match, and some give up mid-way through the qualification questions.

Camera permission errors, repeated photo requests, and upload failures — a recurring theme going back months, still unresolved on this version.

"Stop messaging me" and "unsubscribe" type requests that aren't immediately honored — this is a consent and compliance risk independent of the deletion-flow issue above.

OTP delivery failures, login errors and a password-security concern — these block the user from doing anything else in the flow, so they escalate fast.

Users have to fight through the conversation to fix a wrong field (gender, in both cases) instead of just editing it.

Users refuse to answer further profile questions and the agent pushes rather than backing off.

One user opting out of AI chat entirely, one escalation/transfer-to-human request, one complaint of unwanted spam calls (-0.8 / 0.9, the sharpest score in this whole group), one complaint of being contacted by "multiple agents", and one unwanted-signup complaint.