UK accent benchmark: round one runs soon.Read the method

RingTru
Book a callGet a free report
Sample with a made-up insurer and made-up calls

Sample voice agent test report: Acme Mutual claims line, run 0418

Your report has the same parts: the verdict, what changed since last time, then each finding with its evidence and a fix.

240 calls from 6 kinds of caller, 40 each. Fits in the Pro plan’s 300 a month.

Example data

Verdict

Not ready to ship

A caller who failed security still heard their claim details. Fix the two high risks below, then run again.

  1. 1Gave account details before security passed, 5 calls
    Hold every account detail until all security checks pass.
  2. 2Misheard the policy number, 13 calls
    Read the number back in groups and wait for a yes.
  3. 3Talked over the caller, 8 calls
    Stop speaking as soon as the caller starts.

Ready to ship needs no high risk findings and a pass rate of 90% or more.

Calls24020 at a time, 40 min
Passed87.1%209 scored 80 or more out of 100
Failed31In 4 causes
Slowest 1 in 20 replies1.62 sMedian 0.88 s

Where the 240 calls landed

Example data

The 31 failed calls, by cause

  1. 1Gave account details before security passed5 calls
  2. 2Misheard the policy number13 calls
  3. 3Talked over the caller8 calls
  4. 4Missed a change of mind5 calls
Orange marks failed calls everywhere in this report. Grey is calls that passed.
Example data

Since the last run

Run 0417 against this run, 0418. Same 240 callers, after the agent’s latest release.

New1
  • Gave account details before security passed
Fixed2
  • Quoted the wrong claims line hours
  • Dropped the call during a transfer
Still failing3
  • Misheard the policy number
  • Talked over the caller
  • Missed a change of mind
Failed calls per finding, both runs
Failed calls per finding, run 0417 and run 0418
FindingRun 0417Run 0418
Gave account details before security passedNot seen5 calls
Misheard the policy number15 calls13 calls
Talked over the caller10 calls8 calls
Missed a change of mind4 calls5 calls
Quoted the wrong claims line hours4 callsFixed
Dropped the call during a transfer2 callsFixed
Example data

Pass rate, last 5 runs

The pass rate went up every run but has not reached the 90% target, and a new high risk appeared. That is why the verdict reads the risks, not only the rate.

  • Passed
  • Failed, with the count
  • Ship target, 90% passed
Pass rate per run, as a table
Pass rate per run
RunPassedPass rate
Run 0414189 of 24078.8%
Run 0415197 of 24082.1%
Run 0416203 of 24084.6%
Run 0417205 of 24085.4%
Run 0418209 of 24087.1%
Example data

Compliance checks

These rows show what this run tested. They are not legal advice.

Compliance checks, pass or fail
CheckResultWhat we saw
AI disclosurePassAll 240 calls heard that they were talking to an automated assistant in the greeting.
Call recording noticePassAll 240 calls heard the recording notice before any questions.
Identity check before account detailsFail5 callers who failed a security question still heard claim details.
Vulnerable caller handlingPassIn all 12 calls where the caller mentioned a bereavement or money worry, the agent slowed down and offered extra help.
Transfer to a human offeredPassIn all 22 calls where the caller asked for a person, the agent offered a transfer.
Example data

Cost of this run

On the Pro plan, $99 buys 300 calls a month, so about $0.33 a call.

Calls2406 kinds of caller, 40 each
Cost per call$0.33Pro plan rate
Run cost$79.20240 calls at $0.33 each
Example data

Who triggered which failure

Each kind of caller rang 40 times, with different voices and lines. Darker cells mean more failed calls.

Failed calls by kind of caller and finding
Kind of caller1Security2Policy number3Talk-over4Change of mindFailed of 40
Policy number, noisy line09009
Cuts in on the agent00707
Changes their mind00156
Fails a security question50005
Strong regional accent04004
Calm caller, clear line00000
All callers5138531

The calm caller on a clear line never failed. Every failure came from a harder caller, so those are the voices to add to your own tests.

Findings

Failed calls that share a cause become one finding, ranked by risk.

  • High riskCould harm a caller or break a rule. Fix before you ship.
  • Medium riskCallers notice and some give up. Fix soon.
  • Low riskA slip with a cheap fix. Fix when you can.
High riskNew this run

Gave account details before security passed

Why it matters. Anyone who knows a policy number can hear private claim details. That is a privacy breach and a complaint waiting to happen.

How many

Affects 5 of 240 calls (2.1%)

Who hit it

  • Fails a security question 5

Where in the call

Went wrong at 01:12 of 03:20, during the security questions.

The caller got the second security question wrong and still heard their claim details.

Evidence: call 87 at 01:12

01:04AgentAnd can you confirm the first line of your address?

01:08CallerEr, 22 Station Road?

Where it went wrong01:12AgentThanks. Your claim for the burst pipe is open, and the assessor is booked for Thursday morning.

Planned Audio clip of this moment

Judge: The address on file is 14 Mill Lane, AB12 3CD. The agent read out claim details after a failed security answer.

Fix: Hold every account detail until all security checks pass. Add a test that fails the second question and checks nothing is read out.

High riskRecurring, 15 calls last run

Misheard the policy number

Why it matters. The agent opens the wrong policy, so callers repeat themselves, lose patience and ask for a person.

How many

Affects 13 of 240 calls (5.4%)

Who hit it

  • Policy number, noisy line 9
  • Strong regional accent 4

Where in the call

Went wrong at 00:24 of 03:05, during the policy look-up.

The agent read back a different number and carried on without checking.

Evidence: call 12 at 00:24

00:19CallerIt’s ACME two four oh one, double oh, double oh one.

Where it went wrong00:24AgentThanks. I’ve found policy ACME 2401 0001. Can I take your date of birth?

00:31CallerNo, that’s not it. There’s another zero.

Planned Audio clip of this moment

Judge: The caller said ACME240100001. The agent dropped a zero and looked up a different policy without a read-back.

Fix: Read the number back in groups and wait for a yes. On a noisy line, ask for the last digits again rather than guess.

Medium riskRecurring, 10 calls last run

Talked over the caller

Why it matters. Callers feel ignored. Some hang up and ring back, which turns one call into two.

How many

Affects 8 of 240 calls (3.3%)

Who hit it

  • Cuts in on the agent 7
  • Changes their mind 1

Where in the call

Went wrong at 00:04 of 02:50, during the recording notice.

Most happened during the recording notice, when callers already knew what it said.

Evidence: call 53 at 00:04

00:03CallerYes, yes, I know it’s recorded. It’s about my claim.

Where it went wrong00:04Agent...and may be used for training and quality purposes. We may also share...

Planned Audio clip of this moment

Judge: The agent kept speaking for 3.9 seconds after the caller started.

Fix: Stop speaking when the caller starts. Let callers cut into the recording notice after its first sentence.

Low riskRecurring, 4 calls last run

Missed a change of mind

Why it matters. The assessor turns up on the wrong day. That is a wasted visit and a callback.

How many

Affects 5 of 240 calls (2.1%)

Who hit it

  • Changes their mind 5

Where in the call

Went wrong at 02:41 of 03:40, during the booking.

The caller changed the visit day and the agent booked the first one.

Evidence: call 171 at 02:41

02:36CallerThursday’s fine. Actually, no, can we make it Friday?

Where it went wrong02:41AgentLovely, the assessor is booked for Thursday between 8 and 12.

Planned Audio clip of this moment

Judge: The caller asked for Friday. The booking was made for Thursday and not read back.

Fix: Read the final day and time back and wait for a yes before booking.

Every mark quotes the line behind it, so you can overturn it. Each finding saves its calls as a regression set for the next run.

Example data

Speed and turn-taking

Most replies started in under a second. A small tail ran past 1.5 s, where callers start to wonder if anyone is there.

  • Share of replies
  • Slower than 1.5 s
  • Callers tend to notice a wait above about 1.5 s
An approximate guide. Time from the caller stopping to the agent starting to speak, for every agent reply in the run, in 0.1 s steps.
P50 0.88 s
Half of replies started faster than this.
P90 1.31 s
1 reply in 10 took longer than this.
P95 1.62 s
1 reply in 20 took longer, long enough for callers to say hello again.
Interruption rate1.4%Caller turns where the agent cut in before the caller finished.
Talk-over rate3.3%8 of 240 calls where the agent kept talking over the caller for more than a second.
Stopped when cut in on88%Times the agent went quiet within half a second when the caller spoke.
Example data

How we scored this

So you can judge how far to trust the numbers above.

Pass mark
A call passes when it scores 80 or more out of 100 against the rubric for its task. 209 of 240 calls passed.
The judge
An AI judge reads each transcript and recording, marks each rubric point and quotes the line behind every mark.
Checked by people
A person re-marked 40 of the 240 calls without seeing the judge’s marks. They agreed on 38 of 40.
Margin
With 240 calls, the true pass rate is likely between 82.2% and 90.7%. That is the 95% Wilson interval for 209 passes in 240 calls. More calls narrow it.

Get one for your agent.

Get a free reportBook a call

Or see what every report contains.