Three things that don't line up
The claim is for a theft from a van. The adjuster has the first notification, a schedule of items, and three details that sit awkwardly together: a time that doesn't match a receipt, an item bought more recently than the customer remembered, and a policy that was upgraded eleven weeks ago.
Every one of those has an innocent explanation. Most of the time, all three do.
The adjuster is about to make a decision that shapes everything downstream, and it isn't the decision he thinks it is. He thinks the question is is this genuine? It isn't — that assessment belongs to people whose job it is, with access he doesn't have. The only question in front of him is whether he can obtain a complete, accurate, well-recorded account.
An adjuster who sets out to catch someone will produce a worse account, a worse referral, and a real exposure for his employer.
Detection is not the interviewer's job
This is the central discipline of an insurance claim interview where something doesn't reconcile, and it runs against a strong cultural current.
The assessment sits elsewhere. Specialist teams, with data, cross-referencing and authority the adjuster doesn't have. His contribution is inputs. Contaminating the inputs by pursuing a conclusion makes their job harder, not easier.
Inconsistency is normal in truthful accounts. Human memory is reconstructive, people misremember times and sequences routinely, and stressful events are recalled less precisely rather than more. Treating discrepancy as evidence of dishonesty is a poor inference and everyone who works with witness accounts professionally knows it.
Reading deception from behaviour is unreliable. The confident belief that hesitation, gaze or nervousness reveals dishonesty is widely held and poorly supported. An adjuster acting on it will mostly be identifying anxious honest people, which is a fairness problem and an accuracy problem simultaneously.
And an accusation is an act with consequences. Implying dishonesty to a customer — directly or by tone — creates exposure for the insurer that a careful account, referred properly, does not.
How to get a complete account
1. Free narrative first, uninterrupted. "Take me through the whole day, from when you got up." Don't interject, don't correct, don't ask for clarification while they're talking. The uninterrupted account is the most valuable thing the interview produces and the most commonly destroyed.
2. Then go back to the beginning and fill in detail. Chronologically, at your pace, not theirs. Detail obtained after a free narrative is more complete and more useful than detail obtained by interrupting one.
3. Ask open questions and never a leading one. "What happened next?" not "and was the van locked?" A question that contains its own preferred answer damages the account's value whichever way the person answers it.
4. When something doesn't reconcile, ask once, neutrally, and own it as your error. "I've got the time down as about six — can you help me line that up with the receipt? I may have written it wrong." It's genuinely often true, it produces an explanation rather than a defence, and it keeps the interview usable.
5. Don't bluff about what you have. Implying evidence you don't hold is a technique from a different profession, it is not available here, and it can convert an ordinary claim into a complaint with real teeth.
6. Record the questions, not just the answers. What you asked shapes what you got. A record showing "customer said the van was locked" is much weaker than one showing the open question that preceded it.
7. Refer on facts, through the process. What was said, what doesn't reconcile, what's missing. Not "I had a feeling about this one." Referral is a routine step, not an allegation, and adjusters who understand that refer earlier and more usefully.
And if it's straightforward, say so
The opposite failure is real: an adjuster who notices three anomalies, says nothing, records nothing, and settles — because raising it felt like accusing someone.
Noticing and recording is not accusing. The whole design of this conversation is that it lets you do the first without doing the second.
Four ways it goes wrong
The detective, who sets out to catch someone and produces a defensive, incomplete, contaminated account.
The accuser, whose tone or wording states the suspicion. The version with genuine legal exposure attached.
The bluffer, who implies evidence that doesn't exist.
The smoother, who sees the discrepancies, doesn't want a confrontation, and leaves them out of the file entirely.
Why this isn't trained
Indicator training exists; interview training doesn't. Adjusters learn what patterns to be alert to, and almost nothing about how to conduct the conversation that follows — which is the part that determines whether the alertness produces anything usable.
The "catch them" framing is culturally embedded. It's flattering, it feels like the interesting part of the job, and it's a misallocation of the role.
Deception folklore persists. Beliefs about behavioural cues survive in claims culture long after the evidence for them stopped supporting the confidence placed in them.
And peer role play can't do it. A colleague can't produce the register of a person who is upset at being questioned about a real loss, or the ordinary, innocent inconsistency of someone recalling a bad day. The whole difficulty is telling those apart, and a colleague playing "the dodgy one" removes the difficulty entirely.
What non-accusatory interview training can rehearse
A simulation can present accounts containing anomalies — most of them innocent — and score the interview's technique rather than its conclusion: whether a free narrative was obtained, whether questions were open, whether discrepancies were raised neutrally. Foretell AI supplies the counterparty configuration, transcripts and rubric-based scoring; referral criteria, investigation policy, specialist team processes and all legal and regulatory requirements stay with the insurer.
Four to build:
- The innocent inconsistency, where the anomaly resolves completely and the correct outcome is a clean settlement with the point recorded.
- The offended honest claimant, who reacts badly to being questioned at all.
- The one who volunteers the explanation late, testing whether the adjuster asked in a way that allowed it.
- The one who gives an account that keeps changing, where the correct behaviour is to record carefully and refer, not to challenge.
Design caution — read before building, highest tier in this series. Nothing here describes fraud indicators, detection techniques or behavioural cues, and scenario libraries must not contain them either — content of that kind is both unreliable and inappropriate for a training vendor to author. Referral criteria, investigation procedures and what an adjuster may say or do are set by the insurer and vary by jurisdiction. Modules must make clear that assessment sits with the insurer's specialist function, must not train adjusters to reach conclusions about dishonesty, and confer no assurance of any kind. Scenarios should be weighted so that most anomalies resolve innocently, which is both realistic and the correct thing to train.
Designing the module
Pass one — the narrative. Score whether an uninterrupted free account was obtained and how many times the adjuster interjected during it.
Pass two — the questions. Score the proportion of open to closed and leading questions.
Pass three — the discrepancy. Score whether it was raised once, neutrally, without signalling, and whether the adjuster attributed the possible error to himself.
Rubric on observable behavior: Was a free narrative obtained? Interruptions during it. Ratio of open to leading questions. Was any discrepancy raised more than once? Was suspicion signalled in tone or wording? Was any evidence implied that the adjuster didn't hold? Were the questions recorded alongside the answers? Was the anomaly recorded whatever the outcome?
Leading-question ratio is the measure. It's countable, it's the single biggest determinant of whether an account is worth anything to the people who assess it, and virtually no claims quality framework looks at it.
The operator case
Your specialist function's inputs are produced by people with no interview training. Referral quality is bounded by the conversation that preceded it, and that conversation is currently unassessed.
Accusation exposure is created entirely at the front line. A careful interview and a proper referral carry no such risk; an implied allegation in a phone call does, and it is the individual's improvisation rather than the insurer's policy that produces it.
Measure referral quality, not referral volume. Volume targets produce weak referrals; usable accounts produce outcomes. The specialist team can tell you which referrals are workable and that feedback rarely travels back.
And honest customers treated as suspects leave and complain. Most anomalies are innocent. An interview technique that is neutral by design protects the large majority as a matter of course, rather than relying on each adjuster's judgement about who deserves it.
For insurance and customer experience programmes, this is a good case of role clarity: the adjuster who believes his job is to reach a verdict does his actual job worse, and the discipline is in declining a question that isn't his.
Frequently asked questions
What should an adjuster do when a claim doesn't add up? Obtain a complete account using open questions, raise the discrepancy once in a neutral way, record both the questions and the answers, and refer through the insurer's process on the facts. The assessment itself belongs to the specialist function.
Can you tell if someone is lying in a claims interview? Not reliably from behaviour. Confidence in behavioural cues is widespread and poorly supported, and acting on it mostly identifies anxious honest people.
Is inconsistency evidence of a false claim? On its own, no. Memory is reconstructive and people routinely misremember times and sequences, particularly around stressful events. Discrepancies should be recorded and explored neutrally rather than treated as findings.
How should you raise a discrepancy with a claimant? Once, neutrally, and framed as your own possible error — "I may have written this down wrong, can you help me line it up?" It produces an explanation rather than a defence and keeps the account usable.
The short version
Three things don't line up, and all three probably have ordinary explanations.
His job is not to work out whether this is genuine. It's to get the fullest possible account — free narrative first, open questions, discrepancy raised once as his own error — write down what he asked as well as what he heard, and pass it to the people whose job the assessment actually is.
The adjuster who decides to catch someone gets a worse account, a weaker referral, and occasionally a complaint the insurer can't defend. Declining the question is the skill.
Foretell AI lets insurers build conversational simulations — including account-taking, neutral discrepancy handling and non-accusatory interviewing like the one above — with configurable counterparties, transcripts, recordings, and rubric-based evaluation. If your specialist team's referrals are only as good as the interviews behind them, we're happy to walk through how other insurers have structured it.