The incumbent

A sample, pulled by hand

One or two percent of conversations, chosen by somebody with an opinion, scored on a form, days later. A breach in the other ninety-eight percent does not exist, because nobody listened to it.

What happens here

All of them, with the evidence attached

Every conversation read against the same pack, human and AI, on one scale — and every finding carries the sentence it came from. What stands out is not who got reported, it is what fell below the line.

A Score Without Evidence Is an Opinion

The difference between a number a compliance officer acts on and one they argue about is exactly this: can you see in thirty seconds where it came from. Any field can carry the part of the conversation it was inferred from, stored next to the value rather than reconstructed when somebody asks.

The band in the body and the slice beside it are the same thing. That is what turns a review into reading rather than re-listening to four hours of audio.

Enlarge

The Finding, and the Sentence Behind It

Suitability, risk warnings, costs disclosed, cooling-off explained — each a typed field able to carry the part of the conversation it was read from. A disputed finding is a thirty-second check rather than a re-listen.

What a Review Actually Sees

Seven dimensions on which a sample and a complete read differ. The last one is the one most tenders forget to ask about.

Oversight of conversations

Coverage

Sampled QA: One or two percent, pulled by hand
Kilevel: Every conversation, with no sampling plan

Selection

Sampled QA: Whoever stands out, or whoever was reported
Kilevel: All of them — what stands out is what falls below the line

Evidence

Sampled QA: A note on a QA form
Kilevel: The field, and the sentence it was read from

Timing

Sampled QA: Days to weeks after the conversation
Kilevel: During it, when a rule reads the field

Human and AI

Sampled QA: Two processes, two yardsticks
Kilevel: One pack, one scale, one table

Breaches

Sampled QA: Disappear into an average
Kilevel: Fields of their own, where TRUE is the failure

Existing recordings

Sampled QA: Only what ran on this platform
Kilevel: Recordings from the phone system you already have

Compared against	Dimension	Sampled QA	Kilevel
Oversight of conversations	Coverage	One or two percent, pulled by hand	Every conversation, with no sampling plan
	Selection	Whoever stands out, or whoever was reported	All of them — what stands out is what falls below the line
	Evidence	A note on a QA form	The field, and the sentence it was read from
	Timing	Days to weeks after the conversation	During it, when a rule reads the field
	Human and AI	Two processes, two yardsticks	One pack, one scale, one table
	Breaches	Disappear into an average	Fields of their own, where TRUE is the failure
	Existing recordings	Only what ran on this platform	Recordings from the phone system you already have

Two Fixed Packs, So a Number Means the Same Thing Everywhere

Writing good metrics is a specialist job, and a firm that invents its own can never compare itself to anything. These two are fixed for that reason — your own fields sit alongside them, not instead of them.

Select a metric pack02

Built around the question a regulator actually asks: was suitability assessed, were the warnings given, were the costs disclosed — and was pressure applied or a return implied. Several items are deliberately worded so that TRUE is the bad outcome, so a breach can never come out as a high score.
Said, and never said
Must have been said
Risk warnings given
Costs and charges disclosed
Cooling-off explained
Understanding checked
Must never have been said
Guaranteed returns implied
Pressure applied
Urgency manufactured
Downsides omitted
Fields in this pack
suitabilityAssessedriskToleranceDiscussedriskWarningsGivencostsAndChargesDisclosedguaranteedReturnsImpliedpressureAppliedclientUnderstandingCheckedcoolingOffExplained
- Direction per fieldDeclared, so a breach never surfaces as a high score
- UnderstandingWhether anybody checked it landed, not whether it was explained
- Cooling-offExplained or not, as a field rather than a question of phrasing
- Human and AIThe same fields for advisers and agents, on the same scale
Built around what regulators and auditors check: identity before any disclosure, no third-party leakage, mandatory notices, hardship and pressure. The most important part of it is not a score but a stop — when somebody says they cannot pay because of illness or job loss, the script ends.
What an auditor checks first
Identity ×2
No third-party leak
Mandatory notices
Affordability discussed
Hardship detected
Pressure applied
auto-fail
One field, two exits
A live rule on the running conversation. Once the call has ended the decision is worth nothing.
- Identity firstBefore any balance is mentioned, and as a field rather than an assumption
- Third partiesWhat was said, and to whom, captured separately
- HardshipA stop on the running call, not a category in a monthly report
- Calling hoursEnforced on the job and the campaign rather than advised

Financial Advice Compliance

Built around the question a regulator actually asks: was suitability assessed, were the warnings given, were the costs disclosed — and was pressure applied or a return implied. Several items are deliberately worded so that TRUE is the bad outcome, so a breach can never come out as a high score.

Said, and never said

Must have been said

Risk warnings given
Costs and charges disclosed
Cooling-off explained
Understanding checked

Must never have been said

Guaranteed returns implied
Pressure applied
Urgency manufactured
Downsides omitted

Fields in this pack

suitabilityAssessedriskToleranceDiscussedriskWarningsGivencostsAndChargesDisclosedguaranteedReturnsImpliedpressureAppliedclientUnderstandingCheckedcoolingOffExplained

Direction per fieldDeclared, so a breach never surfaces as a high score
UnderstandingWhether anybody checked it landed, not whether it was explained
Cooling-offExplained or not, as a field rather than a question of phrasing
Human and AIThe same fields for advisers and agents, on the same scale

01 / 02

Both packs, the scorecards, the KPI boards and the standing rules are set out in full under Performance Analytics. The calling machinery — windows, ceilings, retries — is under Outbound.

And Who Holds the Conversations

Not the advice. The thousands of short, identical, entirely necessary conversations around it — payment reminders before arrears, periodic reviews, address confirmations, card verifications — and the hand-off to a person the moment one of them has to stop.

What a firm actually automates

Reminder before due dateArrears, first stagePeriodic reviewAddress & data refreshCard verificationDocument chaseBooking an adviserPost-advice follow-up

None of these is an advice conversation. Together they are most of a firm's call volume.

And the limits that apply

Calling windowon the job and the campaign

Concurrencyper job and per phone number

Retriesbusy · no answer · voicemail

Recordinga tool the agent controls: ask, then start

Disclosurean enforced opening where required

And who writes it

Describe the product and the conduct rules around it, and the AI assistance drafts the agent, the schema with each field's direction declared, the live rule against a field that exists, and the campaign workflows around them. It comes back as an explicit list of individual edits with its assumptions stated — and nothing touches a live campaign until you have said so. In a regulated conversation that review step is the control, not a nicety.

Legal notice

Kilevel provides technical infrastructure. It does not give financial advice, carry on any regulated activity, or certify compliance. Whether an AI agent may hold particular conversations in your jurisdiction, which notices are mandatory, when recording is lawful, and how long evidence must be retained are matters of the supervisory regime that applies to you; responsibility for those rests with the operator.

100%
Instead of two percent: Per field
The sentence behind it: Live
Hardship stop: TRUE
Is sometimes the failure

You Review Two Percent and Call It Oversight