Call Review
Automated call QA for your AI voice agents
Call Review checks every call your AI voice agents take. Each week it tells you what went wrong, shows you the calls where it happened, and suggests how to fix it.
How it works
From this week’s calls to a fix that works.
Every week, Call Review works through each agent’s calls in four steps. It uses the agent’s live setup, the same one your callers hear.
First, real conversations are separated from test calls, voicemail and silent or spam calls. Every real conversation is then checked for common problems: failed transfers, broken promises, repeated misunderstandings, tool errors and slow replies.
AI reads the calls with a problem first, then a sample from the rest of the week. It checks each one against the agent’s instructions and your pass criteria, and notes why the call passed or failed.
Problems are ranked by how much they affect callers. Each one shows the calls where it happened, what in the agent caused it and what to change.
Click Apply fix and the AI Assistant drafts the change for you. Once you publish it, Call Review checks for the problem again, only on calls made after the fix.
Scoring
Scored against your standards, not generic ones.
You decide what a good call looks like for each agent, and Call Review scores calls against it.
Write your own pass criteria
Describe up to 10 things a good call does, in plain words, such as “The caller booked an appointment or was told how to”. Choose whether a call must meet all of them or reach a weighted score.
Set number limits
Set a limit for the things you can count, like response time, made-up answers and tool errors.
Correct a result
If you disagree with how a call was scored, override that check. It fine-tunes the scoring to your judgement.
Choose how many calls are read
Calls with a problem are read first. You set the share of calls AI reads and the most it reads in a week.
Findings
Every problem comes with proof.
The weekly report is a list of problems to fix, each one backed by the calls where it happened.
Biggest problems first
Problems are ranked by how much they affect callers, so the top of the list is where to start.
Quotes from the real calls
Each problem links to the calls behind it, with quotes checked against the transcript.
What caused it and what to change
Each problem says which part of the agent caused it and what to change.
Problems on your side, marked as yours
Some problems aren’t the agent’s, such as no one being free to take a transfer. These are marked as yours, with steps for your team.
Fixes
Know a fix worked before you move on.
Your team approves every change, and Call Review keeps checking after it goes live.
Fixes drafted by the AI Assistant
Click Apply fix and the AI Assistant looks at the problem and its calls, then suggests a change. If it can’t confirm the problem, it says so. Nothing changes for callers until you approve and publish.
Checked on new calls only
Once a fix is live, Call Review only looks at calls made after it. The problem shows as “checking” until there are enough calls to tell, then as resolved or still happening.
Check now
No need to wait for next week’s report: Check now looks at the calls since the fix straight away. You can also review the last 7 days at any time.
Frequently asked questions
Does Call Review check every call or just a sample?
Both. Call Review sorts every call in the week and runs automatic checks on every real conversation. AI then reads the calls those checks flagged (up to a limit), plus a sample from the rest of the week. You choose the share of calls it reads and the most it reads each week.
Can we use our own QA scorecard?
Yes. Write up to 10 pass criteria in plain words, such as “The caller booked an appointment or was told how to”, and add number limits such as response time or tool errors. You choose whether a call must pass every check or reach a weighted score. If you disagree with a result, you can override that check for that call.
Does Call Review score calls handled by our human agents?
No. Call Review only reviews calls handled by Trillet AI voice agents. It does check whether a transfer from the AI agent reached a person.
How does a fix get made?
Click Apply fix and the AI Assistant opens with the problem and the calls behind it. It suggests a change to the agent, or tells you if it can’t confirm the problem. Nothing changes for callers until you approve the change and publish it. If the problem is on your side, such as staffing for transfers, you get steps for your team instead.
How do we know a fix actually worked?
Once a fix is live, Call Review only looks at calls made after it. It shows “checking” until there are enough calls to tell, then marks the problem as resolved or still happening. To find out sooner, use Check now instead of waiting for the weekly report.
How often does Call Review run?
Every week, for each agent. You can also review the last 7 days at any time. A full report needs enough real conversations, so agents with fewer calls are reviewed over a longer period.
Does Call Review work with HIPAA mode?
Call Review doesn’t read calls from agents with HIPAA mode on. HIPAA mode is set per agent, so other agents in the same workspace can still be reviewed.
See what Call Review finds on your calls.
Book a demo and we’ll walk you through a weekly report, its fixes and how each fix is checked.