VR and AI for Conduct and Client Conversation Training in Finance

Financial services spends heavily on conduct training: annual modules, policy attestations, case studies circulated after enforcement action. The training is usually completed. The failures still happen — because a conduct breach is rarely a knowledge gap.

It is a conversation that went slightly wrong. An adviser in a hurry did not ask why the client needed the money. A customer mentioned a recent bereavement and the adviser carried on with the product pitch. A complaint was handled competently but never logged, so the pattern never surfaced.

These are behavioural failures under mild pressure, and they are invisible to multiple-choice testing.

A roleplay with a colleague is the closest existing method — and it rarely happens, because it is expensive, embarrassing and inconsistent. The senior person who could play a difficult client is the same person who is behind on their own work.

VR removes the three barriers at once:

• It is repeatable. The same vulnerable-customer scenario, delivered identically to every adviser in every branch.
• It removes the audience. People rehearse honestly when their manager is not watching them fail.
• It creates a record. A scored session with a transcript is evidence of competence in a way that an e-learning completion flag is not.

It does not replace supervision, coaching or the complaints process. It makes the rehearsal part of training real.

Build the scenarios from your own complaint and conduct data, not from a generic library.

1. Vulnerability cues. The client mentions a health issue, a bereavement, or confusion about the product. The adviser has to notice, adapt the conversation and record it. 2. Suitability pressure. A client wants a product that does not fit their stated objective. The adviser has to challenge, document and, where needed, decline. 3. The complaint. A client is angry and partly right. The adviser has to acknowledge, avoid promising what they cannot deliver, and log the complaint correctly. 4. Escalation to financial crime. The client's explanation does not fit the transaction pattern. The adviser must ask, avoid tipping off, and escalate. 5. The vulnerable customer and the third party. Someone else is prompting the client's answers. The adviser has to separate the two voices.

Each maps to a documented policy clause, which is what makes the training defensible.

A VR conversation without assessment is theatre. The AI layer should score four things, each traceable to a policy line:

• Required disclosures and questions. Did the adviser ask about objectives, affordability, dependants, existing cover — the questions the file must show?
• Cue recognition. Did they respond to the vulnerability signal, and how many turns later? Late recognition is the most common failure we see scored, and the easiest to coach.
• Language risk. Phrases that create expectation, imply a guarantee, or breach the financial promotion rules. Transcribed speech can be flagged in the moment and reviewed afterwards.
• Escalation timing. When the scenario required a referral, how many turns passed before it happened — and did the adviser avoid tipping off?

The output is a session report with timestamps: what was said at 02:14, which policy clause applies, what a better response would have been. That report is also the audit evidence.

Financial services is heavily regulated, and training technology inherits the regulation.

• Do not simulate real client data. Scenarios use synthetic personas, not live customer records.
• Keep the transcript in the training record, not the CRM. The simulation is a training environment; mixing it with client files creates a data protection problem.
• Be explicit with staff. People must know a session is scored and retained, how long it is kept and who sees it.
• Do not use scores as the sole basis for performance action. The defensible use is coaching and evidence of training, with human review before any consequence.

Three design choices separate a programme that survives from one that becomes an annual box-tick:

• Short sessions, frequent. Ten minutes a month beats a half-day once a year, and matches how skills decay.
• Scenario rotation tied to real events. When a complaint theme appears, the next month's scenario reflects it.
• Manager visibility at the right level. Managers see team themes and coaching needs; they do not need to read every transcript. Compliance sees the audit trail.

• Weeks 1–4. Select two scenarios from complaint data. Agree the rubric with compliance and legal before building.
• Weeks 5–8. Build and test with a small group of advisers who will give honest feedback.
• Weeks 9–12. Run a branch cohort alongside existing annual training. Compare not just scores but the quality of escalation notes in the following month.
• Quarter 2. Rotate scenarios monthly and connect completion to the training record.

It will not fix a sales culture that rewards the behaviour the training discourages. If targets and incentives contradict the conduct message, no simulation will change outcomes — and staff will treat it as another compliance ritual.

It will not replace case review, supervision or whistleblowing channels. It gives advisers a safe place to practise the difficult conversation before they have it with someone's pension.