Compliance scoring for every protection sale made by phone.
CallGuard AI scores life, critical illness and income protection sales against your firm's own scorecard. The consent points on the application, such as agreement to share health details and agreement to submit, can be consent-gated, so the scorer's rule is that they are met only on the customer's own clear yes. Disclosure answers are scored on what the customer said, and, if your firm scores sales, the calls from the customer's number within the sale window are scored together. Every pass and fail shows the transcript evidence it was decided on, or a note that none was found.
A protection sale is made in a call or two. A non-disclosure surfaces at claim.
The application records the customer's answers. The recording records how the questions were put: each condition asked on its own, or a list read at speed and answered with one "no"; agreement to share health details before the questions, or a pause taken as a yes; a replacement explained, or an old policy cancelled before the new one starts. Under the Consumer Insurance (Disclosure and Representations) Act 2012 the duty is the customer's: to take reasonable care not to make a misrepresentation to the insurer (s2). How clear and specific the insurer's questions were is one of the Act's examples of what may be taken into account in judging that care (s3(2)(c)), and where the adviser reads the questions out, the recording shows how they were put. Sampled QA hears a fraction of those calls.
Sampled QA on a sales floor
- A small sample of calls listened to
- A list of conditions answered with one "no" goes unnoticed
- Consent read into silence or an "mm"
- Findings only for the calls that were sampled
- The call and the application compared only if someone reads both
- Vulnerability cues caught only if someone hears them
CallGuard AI on every protection sale
- Every sale scored against your scorecard, or every call on its own, depending on how your firm is set up
- Disclosure answers scored on what the customer said
- Consent items are met only on a clear customer yes
- Critical fails surfaced as soon as the sale is scored
- Call answers checked against the application, on the Pro plan
- A vulnerability criterion, if your scorecard has one, scored on every sale
Your scorecard, built around the questions that decide a claim.
CallGuard scores what your scorecard says: your firm defines the checks. Firms bring their own QA scorecard, and can add items from our public protection consent-gate checklist. For firms selling protection under ICOBS, a scorecard usually covers demands and needs, which ICOBS 5.2 applies to every sale, advised or not, and, where advice is given, the advised-sales provisions of ICOBS 5.3: the suitability rule (5.3.1R), the FCA's guidance for advice on pure protection (5.3.2G), and, where a personal recommendation is made, 5.3.4R. Our guide to scoring demands and needs labels each criterion as rule, guidance or firm standard.
Consent gates on the application
Agreement to share health details with the insurer before the first health question, agreement to a medical or other search before it happens, confirmation that the answers are true and complete, and agreement to submit. Set up as consent gates, the scorer's rule is that they are met only on the customer's clear yes, not on silence or the adviser's words. Where it is not reliably clear which speaker is the customer, a person decides.
Disclosure answers from the customer
Smoking and vaping status, height and weight, occupation and hazardous activities, and each medical condition asked on its own. Scored on what the customer said, not the adviser's assumption. ICOBS 5.1.4G, which is guidance, gives asking clear and specific questions, and explaining the customer's responsibility to take reasonable care not to make a misrepresentation and its possible consequences, as ways of ensuring a customer knows what to disclose.
Demands and needs
Needs obtained from the customer rather than supplied to them ("so it's just life cover for the mortgage?"), as ICOBS 5.2.2R requires; asking before anything is proposed is a firm standard. The proposal consistent with the customer's demands and needs (ICOBS 5.2.2BR, advised or not). The statement of demands and needs is a file check, and can be kept as a manual item that is never sent to the model.
Advised sales
Existing cover established (ICOBS 5.3.2G(1)(a), guidance), including cover through an employer (firm standard). A need the customer raised that the recommendation does not meet, told to them (ICOBS 5.3.2G(1)(c), guidance). Where a personal recommendation is made, reasons that belong to this customer rather than general praise for the policy (ICOBS 5.3.4R).
Replacement business
Where a new policy replaces existing cover, the customer confirms they understand any gap in cover during the switch (a firm standard, and one of the checklist's ten items); explaining that risk first is a further firm standard. No ICOBS 5 provision deals with replacement business as such.
How the sale was made
No pressure or false urgency. Understanding checked, and questions invited. A sign of vulnerability, such as a recent diagnosis or bereavement, acknowledged and acted on rather than talked over. Why non-disclosure starts on the call.
Built for a team that sells from a dialler and a CRM.
Calls arrive from your own dialler: through the CloudTalk connector, SFTP pickup of any dialler's recording export, or upload over the API. A sale in your CRM starts sale scoring, and scores are written back to your QA module; Zoho CRM is the connector already built, and for any other CRM CallGuard builds the same connection during setup. By default, names and other personal, payment and health details are replaced with tags in the transcript before it is stored or scored, and health details are kept readable only where we set that up with a firm, with a DPIA. Audio is encrypted at rest with AES-256-GCM, and customer data is stored in the UK. Our DPA and sub-processor list are published, and our plans are quoted per seat for your team.
The whole sale, not one call
If your firm is set up to score sales, the calls from the customer's number within the sale window are scored together as one compliance unit, so a sale is judged on those calls together, not failed call by call for gaps covered on another.
The call checked against the application
Reconciliation compares what the customer said on the call with the answers submitted on the insurer's application, and flags the ones that don't match for your team to raise with the insurer. It runs once we've switched it on for your firm and the insurer's application PDF is on the sale in your CRM. Health details a customer names are redacted by default, so an answer that depends on them is usually marked "Could not verify", not a match; plain yes or no answers are still compared. It is on the Pro plan.
Evidence a reviewer can check
Pass or fail per criterion, with the transcript evidence it was decided on. A criterion with nothing relevant on the call is recorded as "no relevant evidence found". A supervisor looking at a critical fail reads what was said, agrees or disagrees, and moves on.
Shown your firm's own corrections
When your compliance officer overturns a verdict with a reason, that correction becomes the record, and up to five of the most recent corrections on that criterion are shown to the AI as examples when it next scores that criterion for your firm. More on scoring calibration.
Common questions, answered.
Is there a ready-made protection scorecard?
Our protection consent-gate checklist is public: ten items, each described in full on the page, and the CSV is emailed on request. You can bring your own QA scorecard or manual, which we set up as your scorecard, and add any of the checklist's items to it. Nothing we supply is FCA-endorsed, and none of it replaces your own compliance judgement or your insurers' application questions.
What does consent-gated mean in practice?
A consent-gated item carries a stricter rule for the scorer: it is met only on the customer's clear yes. A vague or inaudible answer is not a pass, and neither silence nor the adviser simply saying the words is agreement. Where it is not reliably clear which speaker is the customer, the item goes to a manual review queue and a person decides, because a false pass on a consent item is the worst result the scorer can produce.
Our advisers take more than one call to close a sale. How is that scored?
If your firm is set up to score sales, as one sale. When the sale is recorded in your CRM, the calls from that customer's number within the sale window are scored together as one compliance unit. A sale where needs were covered on the first call and consent to submit on the second is judged on those calls together, rather than failing each call for what happened on the other.
We have elected to apply COBS to our pure protection business. Does ICOBS still apply?
Not to that business, apart from ICOBS 4.6 (ICOBS 1 Annex 1 Part 2, 3.1R), so the ICOBS references on this page don't apply to it. CallGuard scores your firm's scorecard rather than a fixed rulebook: you write your scorecard to the rules you work under.
Which diallers and CRMs does it work with?
Calls arrive from your own dialler: through the CloudTalk connector, SFTP pickup of any dialler's recording export, or upload over the API. A sale in your CRM starts sale scoring, and scores are written back to your QA module. Zoho CRM is the connector already built; for any other CRM, CallGuard builds the same connection during setup.
Can we see it on our own calls?
Demos use synthetic calls, so nothing of yours is needed to see how it works. Scoring your own recordings comes after a data processing agreement is in place, as it does for every firm.