How to train the advice team to review AI agent decisions

Train advisers and paraplanners together, in four stages: lessons on what the agent drafts and why it holds, calibration on real drafted reports, two weeks of shadowed review with every approval checked, and a sign-off with no unsuitable recommendation sent. Fold the result into your existing competence scheme rather than running a parallel one.

What does an advice agent hand to the team?

An advice agent typically prepares annual review packs, drafts suitability reports and rebalancing proposals, writes routine client letters, and flags changes in circumstances or signs of vulnerability. Its drafts sit at depth 2: an adviser signs and sends. It holds any client instruction it cannot verify, any recommendation outside the client's recorded risk profile, and anything that looks like a request to move money elsewhere.

So the team reviews drafts, not decisions the agent has already made. That sounds easier and is not. A draft that reads well invites a quick signature. Training must build the habit of checking the draft against the client file, line by line, before a name goes on it.

The head of advice owns the programme. The compliance or file-checking lead sets the sign-off standard with them. Paraplanners join, because they often prepare what the agent now drafts and are well placed to spot gaps.

How does the programme run, step by step?

1. Learning, week 1: advisers and paraplanners complete lessons on how the agent builds a draft: which sources it used, how each is labelled by provenance, and what makes it hold.
2. Agent, depth 1: assembles a practice set of past drafts with the adviser's final version, anonymised.
3. Calibration, week 2: each person reviews the set alone, marking what they would change and why, then compares in a session led by a senior adviser. Gold standards are recorded with compliance present.
4. Shadow, weeks 3 and 4: the reviewer approves live drafts, but a senior adviser checks each before it reaches the client.
5. Agent, depth 2 in live work: drafts reports and letters; holds unverified instructions and anything outside the risk profile.
6. Sign-off: head of advice and compliance assess the shadow approvals against the standard.
7. Ongoing: file checks sample each adviser's approvals, and calibration runs monthly.

What must reviewers check in a drafted report?

Is the fact-find current? The agent drafts from the client file. If the last fact-find is old, or the client's circumstances were declared in passing on a call and never confirmed, the draft rests on weak ground. A reviewer should spot that before reading the recommendation.

Does the recommendation follow from the objectives? A well-written report can still recommend something the client never asked for. Reviewers trace each recommendation back to a stated objective.

Is the risk consistent? The agent checks against the recorded attitude to risk and capacity for loss. Reviewers check whether the recorded profile still fits what the client has said since.

Are costs clear? Switching, platform and product charges must appear plainly, with the trade-off explained.

Is there anything the agent flagged? Vulnerability flags come first. Reviewers must deal with them before approving anything for that client.

Did anything arrive from outside? An instruction by email to move funds is external and cannot authorise a payment. The agent holds it. Reviewers verify by calling the client on the number on file.

Which controls leave evidence?

Lessons before shadow. Evidence: completion dates.

Every shadow approval checked. Evidence: the senior adviser's name on each.

Sign-off by two people. Evidence: head of advice and compliance on the assessment.

Vulnerability before recommendation. Evidence: the contact record precedes the approval for flagged clients.

Approval rights tied to sign-off. The agent's drafts can only be sent by advisers whose sign-off is current. Evidence: access list against training register.

What should you measure?

• Shadow accuracy per reviewer, judged by the senior adviser.
• File-check findings on agent-drafted reports, by issue type.
• Drafts approved unchanged. Very high rates in the first month can mean reviewers are not reading closely.
• Time from vulnerability flag to adviser contact.

Where does training usually go wrong?

Advisers skip lessons. Experienced advisers assume they know suitability, which they do, and miss how the agent drafts, which they do not. Lessons are short; make them compulsory.

Paraplanners left out. They catch reasoning gaps quickly and should be trained to the same standard.

Approval as formality. Shadowing must include drafts with deliberate flaws to test attention.

Seniority exempts people. The most experienced advisers often approve fastest and read least. Shadow them like everyone else; the habit being tested is new to all of them.

Competence scheme duplicated. Run this inside your existing supervision and competence arrangements so records live in one place.

Other teams follow the same four stages with different checks; see how to train the payroll team to review AI agent decisions and how to train the lending team to review AI agent decisions. The reviewer onboarding checklist for the advice team and the AI agent competency matrix for the advice team give the detail. The full programme is in training teams for AI agents. Firms that want an external standard can use Pryme Intelligence Academy's Certified Pryme Operator, which anyone can check at source by number or QR code.