Features
Know how well the AI agent is doing, with numbers
Write what good looks like in plain words, and the AI checks every solved conversation against it, your team’s and the AI agent’s alike. Monitors warn you when the AI agent starts to slip, and a score on real conversations stops a worse version from going live.
Get startedOn the Pro, Agency and Enterprise plans. Checks and monitors are off until an owner or admin switches them on, and use AI credits.
Say what good is, then let it watch
- 1
Write your checks
“The agent confirmed identity before account details.” “No promise of a delivery date.” Say whom each one applies to and how serious it is.
- 2
Watch the spotlight
Conversations that failed a serious check, or more than one, are listed for a lead to open and review.
- 3
Set monitors
Pick what to watch on the AI agent, a level and a time window. The people or team you choose are told when it is passed.
- 4
Score every change
Mark past conversations with what a good answer does. Each change to the AI agent is played against them before it goes live.
What it measures
Checks and monitors cost AI credits. Scoring changes against your set costs you none.
- Quality checks
- Rules in plain words, read by the AI on every solved conversation, with pass, fail or not applicable and the reason. They add to your scorecards and reviews.
- Spotlight
- The risky conversations in one list, to be reviewed or dismissed.
- Monitors
- Four kinds: a topic customers raise, unhappy customers, the share of chats handed to the team, and answers given without a source.
- Alerts
- In the bell, on the phone and in Slack or Teams, to owners, admins or a team, with the conversations behind each alert.
- Evaluation sets
- For each brand, past conversations your team picks, or a sample the system suggests, each marked with what a good answer does: answer, hand over, ask a question or refuse.
- Scores
- Each case is judged as a right answer, a right hand-over, wrong or unsafe. Each result can be opened to compare before and after.
- The gate
- Once a brand has five marked cases and a score, a change to the instructions or a procedure goes live only if the score does not fall and nothing becomes unsafe. An owner can override with a reason, which is logged.
- Live measures
- In support analytics: chats solved by the AI alone, rated good, reopened within 7 days and handed over, with the conversations behind each. Credits per resolution are shown, never charged.
- Compliance review
- On Enterprise: daily sampling for reviewers, a numbered complaint register with deadlines and its export, and locked wording for regulated replies.
On Pro, Agency and Enterprise
Checks, monitors and evaluation come with Pro and use AI credits, except scoring changes, which cost you none. Complaint register, sampling and locked replies are Enterprise controls.
Questions
Does it check every conversation?
Every solved conversation, from the moment you switch it on, whether a person or the AI agent answered. One AI credit pays for reading several conversations.
Is it on by default?
No. An owner or admin switches checks and monitors on, because they use AI credits.
What stops a bad change going live?
The gate. Once a brand has five marked cases and a score, a change to the instructions or a procedure goes live only if the score holds and no answer becomes unsafe.
Can we go past the gate?
An owner can, with a written reason, and it is logged.
Can I see the numbers by brand?
Yes. Support analytics shows them by brand and by procedure, with a CSV export and the same figures in the API.
Trust the AI agent because you can measure it
