Life Insurance Direct, Australian life-insurance broker and AFS licensee
The shift
A risk-based approach used to mean manual selection and reports, to work out where to look. Now the calls arrive already triaged, and the review starts at the top.
Before
You pick the calls worth a listen, and find out later whether you picked the right ones.
After
Your team works the high-risk calls first, then the medium, then the ones that are coaching opportunities.
In August 2025 ASIC wrote to life insurance chief executives naming the review of only a small percentage of sales calls as a deficiency, and pointing to firms using speech analytics to monitor 100% of them. Read the letter.
Both sides
Every call is scored on the customer's side and on the adviser's side, because the expensive failure is not always the one that breaks a rule.
The one you already fear
One warm sentence at the end of a good call, from your best adviser, meaning well. It surfaces months later as a complaint, and by then it is a remediation event rather than a coaching conversation.
The one nobody counts
Over-warned, hedged, and lost a customer who was ready to buy. Nothing improper happened, so nothing is reported, and it never reaches a compliance pack because there is no breach in it to find.
What we cannot show you is the sale that did not happen: it leaves no recording. What we can show you is where caution is running hot across the floor, and that is the part you can actually coach.
Nuance
The hard calls are not the obviously bad ones. They are the ones that are almost right.
The advice boundary
The crossing is rarely blatant. It is one warm sentence at the end of a good call, from your best adviser, meaning well.
General advice
"Stepped premiums start lower and rise with age; level premiums start higher and stay flatter. Here's the PDS so you can compare."
Personal advice
"I'd go with this one if I were you, that's the best one for your situation."
It reads the subtle forms too, and whether the general advice warning was given.
Technical correctness
This is where a technically correct explanation quietly goes incomplete. An error announces itself. An omission makes no sound.
Technically correct
"An age-stepped premium is recalculated each year as you get older. A level premium is averaged to stay flatter: higher now, more stable later."
Incorrect / misleading
"Level premiums never change." (They still move with CPI and indexation.)
Incomplete scores differently from wrong, because on your floor they are different problems with different fixes.
Some calls genuinely sit on the line, and on those the honest output is a flag for a person to read, not a verdict. A concept that sounds certain about an ambiguous call is a concept you should trust less, not more.
Concepts
No tagging project, no model to train, no data science on your side. Find a gap in what you are picking up, and you close it.
How your team judges a call. Your wording, your edge cases, where you think the line sits. That document is the input.
Our team turns your document into a working concept: an AI agent that reads every call for what you described.
Your compliance team checks the output against their own judgement. We refine until it matches. Then it runs on every call, not a sample.
The library grows. Every gap you find becomes another concept, so the set gets sharper the longer it runs. Twenty-five concepts run on Life Insurance Direct's calls today.
Human in the loop
Nothing is auto-actioned. Every flag waits for one of your people to decide what it is.
It arrives with the evidence attached. Your QA team accepts it or dismisses it.
The reason is recorded against the concept that raised it, so the concept can be corrected rather than argued with.
Scores are role-based and built for coaching. We do not rank your advisers on a wall, and we never planned to.
We report nothing to a regulator on our own. A flag enters your existing breach and complaints process, and your framework decides what it is, exactly as it does when a QA analyst escalates something today.
Scale
The constraint on quality assurance was never the will to review calls. It was the hours in the week, and those same hours quietly set the ceiling on how fast you could grow.
The same team now covers every call instead of the ones they had time for, and it keeps covering every call as the floor gets bigger. Quality assurance stops being the line you have to fund again every time you hire. When the same thing recurs across a floor, it is usually one wording problem rather than a hundred separate ones, and that is a single coaching conversation.
Coaching
Once every call is read, you can see how an adviser is working, not just whether one call passed.
The index exists to point a coaching conversation at the right thing, not to rank your floor.
Every call an adviser takes is scored on both sides. The per-agent Call Quality Index sits on top of that. You can see where someone is on their training journey.
The same concepts run across the whole floor. So the area where one adviser is weaker than their peers stands out. Training goes at that area, rather than at everyone.
Onboarding stops being guesswork. You can see which parts of the call journey a new adviser has not got yet, and coach those parts in week one.
Customer story
Where your calls sit
You are not being asked to hand your call recordings to a third party.
Inside your own sovereign cloud: your tenancy, your region, Sydney for the Australian deployment. Your account, your keys, your logs. The deployment goes to the data, so the recordings never leave your environment.
A de-identified transcript reaches the model, with PII stripped before that happens. The model runs on Amazon Bedrock, inside your own boundary and your own region, so the transcript does not leave your environment to be scored. Health and underwriting data stays where it is.
Avesta Labs, who build and operate Metricsense, holds a SOC 2 Type I attestation, scoped to the company including the Metricsense platform. The Type II observation period is under way.
Your reviewer will want more than three answers. Read the governance page. It answers where your data sits, what leaves your environment, and what the software decides against what you decide, without a form and without talking to us.
The pilot
Free. One week to the readout. Nothing to integrate, and no IT ticket.
Around 50 calls, audio or transcripts. AWS Connect and CSV are built in, and anything with an API can be connected.
You document how your team judges a call. We assess that document, build the concepts from it, and run them across the whole sample inside your own sovereign cloud.
One week from the sample landing to the readout, every flag pinned to the moment behind it. You measure us against your own QA, not against a number we printed on a website.
The pilot costs nothing. If the readout does not hold up against your QA, you keep the readout and we go away. Pilot data is deleted on request when it ends.
No self-serve tier and no per-seat fee. Pricing runs on call volume, scoped to your concepts and your security requirements, and quoted as one package. You will have the shape of it before you have to take it to anyone internally.
What 50 calls will not do is tell you what is happening on your floor. It is enough to check whether our flags match your QA, and that is all we claim for it.
Questions
Call compliance monitoring software reads the conversations your advisers have with customers and checks them against the standards your business is licensed to meet. Metricsense reads every call rather than a sample, scores both the customer side and the adviser side, and pins every flag to the moment in the call that produced it, inside your own sovereign cloud.
No. What ASIC did, in its August 2025 letter to life insurance chief executives, was name the review of only a small percentage of sales calls as a deficiency, and point to firms using speech analytics to monitor all of them. That is a description of what good looks like rather than a rule, and it is the comparison your next review gets read against.
No. Your telephony stays exactly as it is, no script changes, and your advisers do nothing differently on a call. Metricsense reads the recordings or transcripts your current system already produces, where they already sit inside your environment, and any source with an API can be connected. Nobody is asked to migrate a phone system to be able to review its own calls.
Only where a recording or a transcript already exists. We read what your business keeps, audio or text, including an archive nobody has opened since it was made. If a channel is not recorded at all there is nothing on it for us to read, and switching recording on is a change inside your own telephony and your own disclosure process rather than something we do for you.
Yes. You start with whatever your quality process already cares about most, and the library grows from there as gaps turn up. Twenty-five concepts run on Life Insurance Direct's calls today, and that is not where they began either. The pilot deliberately runs on a narrow set, because it exists to test whether our flags match your judgement, not to cover everything at once.
The pilot runs on around 50 calls you already hold, and the readout comes back inside a week. Past that we will not quote you a go-live date before we have seen what your security review requires and how quickly your team can document the standard it judges calls against, because both of those sit on your side rather than ours.
That sits with you. Metricsense runs inside your own tenancy, so who can open a call and read its score is governed by your own identity and access controls, the same ones that already govern the recordings. Inside the product the scores are role-based and built for a coaching conversation, and there is no floor-wide ranking of your advisers in it.
It stays where it has been the whole time. The recordings, the transcripts, the scores and the evidence all sit inside your own cloud account, so ending the engagement takes our software out of your environment rather than taking your data out of your business. Pilot data is deleted on request when the pilot ends.
Written answers to the questions they actually have, against your deployment rather than in general, and if they work from a security questionnaire we complete it in their format. Our governance page answers the first few without a form and without talking to us. We would rather answer what your reviewer asks than send a generic pack that does not.
There is no self-serve tier for calls and no per-seat fee. Pricing runs on your call volume, scoped to the concepts you need and your security requirements, and quoted as one package. The pilot costs nothing: send around 50 calls you already hold and the readout comes back inside a week, whether or not you go further.
Send around 50 calls you already hold. We run your concepts across all of them, inside your own sovereign cloud, and send back a risk readout inside a week with every flag pinned to its quote.
Not ready to book? That is the step before the pilot, not instead of it: your reviewer can read it without a form and without talking to us.
Or reply to hello@metricsense.aiLast reviewed 9 September 2026. This review rebuilt the page around what our customers actually say about the work, and moved the regulatory detail to the governance page, where the reviewer who needs it will look.
