0a3f71c2-8e04-4b1d-9f27-5c60a884e113
Adviser A · 04 Nov, 09:12
Not reviewedT13/3
T31/3
T41/1
T24/7T2.53
Topic 2.3Topic 2.5
Case study · Life insurance · Australia
From checking 2 to 5% of calls to checking every one.
Life Insurance Direct has compared life cover for Australians since 2006, and almost all of it happens on the phone. We built Metricsense to read every call, score the adviser and the customer, and give the QA team a list ranked by risk.
- The problem
- Only 2 to 5% of calls checked by hand. Feedback 3 to 4 weeks late.
- What we built
- AI that reads and scores every call, then ranks them by risk.
- The result
- Every call checked. QA scaled with no added headcount.
- Status
- Live in production. A person confirms every flag.
The same problem shows up in
- Mortgage brokers
- Intermediaries
- Life insurance distributors
- Advice licensees
- Any regulated contact centre
If you sell, service and get audited on the phone, a small hand-picked sample means most of your calls are never checked.
Before and after.
Before
5steps
A careful, risk based process. But every step was done by hand, so only a few calls per adviser ever got checked.
- 1Pull the call reports
- 2Work out which calls look risky
- 3Pick a handful per adviser
- 4Listen to each one end to end
- 5Write up the scorecard
Most calls were never heard, and feedback reached advisers 3 to 4 weeks late.
After
2steps
Every call is read and scored first. The team then starts with the highest risk calls, not a random pick.
- 1Every call read and scored
- 2Team reviews in risk order
No longer done by hand
- Pull the call reports
- Work out which calls look risky
- Pick a handful per adviser
- Listen to each one end to end
They now pick up customer experiences they would have missed before.
9,218
calls checked
Every call from 16 April to 10 August 2026, adviser and customer side. Not a sample.
2 to 5%
checked before
What their careful manual process could reach.
26
checks on every call
Plain English rules, from the advice boundary to call handling.
0
added QA headcount
QA scaled without hiring, in the CEO's own words.
“Before Metricsense, checking calls was a manual process: we selected files for review. Now the platform analyses every call, and we review by risk. It has identified customer experiences we would not have picked up before, and it lets us scale our quality assurance program without increasing headcount, with our team still the human in the loop. It helps us protect our customers, our business and our agents.”
The problem
Why not just listen to more calls?
They were already doing manual QA, and doing it well. The problem was the calls it never reached.
The maths does not work
About 2,000 calls a month. Checking them all by hand would take a whole department, not one more reviewer.
Feedback came too late
Advisers heard about a call 3 to 4 weeks later. By then they had taken hundreds more the same way.
Samples miss the unexpected
Even risk based picking only finds the calls someone already suspected.
Keywords are not enough
What matters is whether an explanation was complete, or a customer was led. Keyword search cannot see that.
What they asked for
- Check more calls without growing the QA team.
- Keep a person deciding. Always.
And one more: do not just check that advisers stay inside the advice line. Check that they still help the customer understand what they are buying.
So the goal was to coach the whole team, not police it.
In production
What it does now.
Reads both sides
Scores the adviser and the customer, so you see how the call was handled, not just how it ended.
Ranked by risk
Highest risk calls first, then medium, then coaching opportunities.
Coaching, not just compliance
Checks whether advisers still helped the customer understand the product, not only whether they stayed inside the rules.
A score for every adviser
One quality score out of 100, built from accuracy, satisfaction, resolution, compliance and communication.
Training aimed at the gap
See where each person is weaker than the rest of the team, so onboarding and refreshers target that.
New checks in one sentence
Write a check in plain English and it runs on every call from then on. No retraining.
It flags. A person decides. Metricsense never reports anything to a regulator, or anyone else, on its own.
The screen they open in the morning.
Every call already read and already sorted. The only decision left is where to start.
High Risk3
6b18d40f-2ca7-4e59-8d13-b0947fe2a5c6
Adviser D · 04 Nov, 11:47
Not reviewedT13/3
T32/3
T41/1
T25/7T2.53
Topic 2.4
c92e5a77-31bd-4a08-9c64-72f1e0d3b845
Adviser B · 03 Nov, 15:26
Not reviewedT13/3
T31/3
T41/1
T24/7T2.53
Topic 3.3
Medium Risk9
Low Risk34
Not Applicable4
The columns fill left to right as you scroll, which is the order a reviewer works in. Not Applicable is the fourth column for a reason: a call where nothing was triggered still got read.
And the one that decides who gets coached on what.
The real library of checks, at its real depth. Check names are hidden at the client's request.
Agent Scorecard
| Concept ↓ · Adviser → | Adviser A | Adviser B | Adviser C | Adviser D | Adviser E | Adviser F | Adviser G | Total |
|---|---|---|---|---|---|---|---|---|
| Topic 1(3) | ||||||||
| Topic 1.1 | clean | |||||||
| Topic 1.2 | 1 | 1 | 2 med | |||||
| Topic 1.3 | ||||||||
| Topic 2(7) | 2 | 1 | 1 | 2 | 1 | |||
| Topic 2.1 | ||||||||
| Topic 2.2 | 2 med | |||||||
| Topic 2.3 | 2 med | |||||||
| Topic 2.4 | 2 med | |||||||
| Topic 2.5(3) | 1 | 1 | ||||||
| Topic 3(3) | ||||||||
| Topic 3.1 | clean | |||||||
| Topic 3.2 | ||||||||
| Topic 3.3 | ||||||||
| Topic 4(1) | ||||||||
| Topic 4.1 | clean | |||||||
| Call | 15 | 8 | 8 | 6 | 5 | 4 | 4 | 50 |
Read a row and you are looking at one concept across the whole floor. Read a column and you are looking at one person's training plan.
Trust and data
Safe for a licensed business.
Life Insurance Direct holds an Australian financial services licence. Their calls are the sale, the service, and what a regulator would read. So the system had to be careful before it was clever.
The rules it runs by:
Evidence on every flag
Each score links to the exact moment in the call, with the audio.
A person decides
The system suggests. The team confirms. It reports nothing on its own.
Stays in Australia
Runs in their own cloud account. Personal details are removed first.
Never shared
Their calls never train a model used by anyone else.
The process
How we worked.
- 01First
Checks for every call first
Checks that apply to every call went live first. Checks for specific call types were added on top.
- 02A few days each
Their QA rules became the checks
Their scripts and procedures were turned into plain English checks the system runs on every call.
- 03Weekly, still running
Their QA lead marked it
Each week their QA lead reviewed calls beside the system and told us where it was wrong. We tuned against that.
- 04Ongoing
New checks in days
Spot something worth watching, write it as one sentence, and it runs on every call from then on.
What the first round changed
Not just "did they stop?" but "what did they say next?"
An adviser who says "I cannot give you advice" and stops is compliant, but may have lost the customer. So the check now looks at whether they followed up with what they can do, and whether the customer left understanding the product.
This was the client's idea, not ours. It changed the product, and we rebuilt the checks around it.
The hard part
The obvious build was the wrong one.
They wanted to check that the insurer's application questions were read out correctly on the call. The obvious way: give the AI the whole form and the transcript, and ask it to compare.
Why that fails
AI skims the middle of long documents
On an insurance application, the middle is where the medical questions sit. That is exactly the part that must be checked word for word.
What we did instead
Give it less, and the right part
Each check only sees the one section it needs, with personal details removed. As their QA lead put it: the less we give it, the less it makes up.
A wrong flag costs more than no flag. If reviewers keep opening flags that turn out fine, they stop trusting the system.
“The thing I have enjoyed the most is the responsiveness and the willingness to challenge yourselves and how you might have deployed a solution. Open to challenging us as well, and giving us different options to consider, and options that we never even thought of.”
Scope
What we left out on purpose.
The application check above is still a pilot, not running on every call. Their customer record holds around ten fields while the application collects around a hundred, so there was no clean record to check the call against.
We will not ship a check that guesses. It stays a pilot until it is something we would defend in front of their regulator.
Six months from now
How it stays accurate.
Before any check changes, it is tested against calls their QA team has already reviewed and marked. Nothing goes live unless it still agrees with them.
- 1Does it still flag what reviewers flagged?
- 2Does it stay quiet on calls they cleared?
- 3Does every score still link to its moment?
- 4Does a person still make the decision?
- Goes live
Without this, accuracy slips quietly and reviewers stop trusting the list.
An example
Early on, some good calls landed in the high risk column. Tightening the rule fixed that, but risked missing other calls. Testing against the marked calls let us measure that trade instead of guessing.
Still improving
Their CEO says the platform still needs ongoing training, and that they are well placed to keep improving it as it matures with the business. The testing above is how that happens safely.
Start with fifty of your own calls.
A short discovery session, then a pilot on your real calls with your own checks. Compare our results with your own QA before you commit to anything.
Book an AI KickoffOr email hello@avestalabs.ai