Exit Poll
Ask a random sample of 'resolved' customers three days later. Project the truth.
The job
Measures any AI agent's real resolution rate, a vendor's or OnClarity's own, by asking a random sample of "resolved" customers three days later, in their own language, if the problem was solved. It is certified monthly so it can be used to dispute an invoice.
The moment
Three days after an AI agent marks conversations resolved, a random sample of those customers is asked, in their own language, whether the problem was actually solved.
Replies feed a board styled like an election-night results show, and the margin of error narrows as more answers arrive.
At 11:40 the result is called: the vendor claims 81% resolved, but the poll projects 58%, give or take 4 points, based on 412 replies.
Finance forwards the signed ledger page to the vendor before lunch.
What it does
- Draw the sample
- Open the polls
- Poll by voice instead
- Route a 'No' case
- Draft the vendor dispute note
- Sign and export the monthly ledger
What you see
An election-night results board (a live projection whose margin of error narrows as replies arrive, a swingometer against the vendor's claim, intents called green or red) plus a monthly certified ledger that finance signs
What it moves
The certified resolution rate compared with the rate billed for each bot, and the dollars of outcome fees disputed or recovered each month, with disputes raised within 30 days.
Built for
- Finance
- Frontline staff
- Compliance, complaints & legal
- CX leaders & VoC
