Responsible AI

AI Governance for Wealth Managers: Evidencing Advice Quality at Scale

May 21, 2026
10 min read

Part of the AI on Trial: The Burden of Proof campaign series

The Defence Brief: Wealth & Advice

A UK wealth manager deployed an AI assistant across its adviser team. Meeting recordings, fact-find updates, suitability report drafts, client correspondence. Hundreds of regulated outputs a week. When the FCA’s next thematic review lands and asks how the firm evidenced advice quality across its full client base, the answer will need to be more than a sample-based QA score and an AI policy document. It will need to be every suitability record, every interaction, every vulnerability indicator, all retrievable and all reviewed. Most wealth firms cannot produce that today. The tool was procured for productivity. The governance work came later, or not at all.

This is the defence brief for wealth and advice. Five governance gaps run through the entire AI on Trial series. Three of them apply with particular force to wealth managers operating under Consumer Duty and the post-2025 supervisory environment: audit trail, sampling-based monitoring, and guidance quality. This article translates those three charges into the specific regulatory and operational context of UK wealth management.

Why wealth managers face the sharpest advice quality test in financial services

Wealth managers carry an exposure profile that sits at the intersection of two FCA priorities.

The first is advice quality. A suitability report, a fact-find, an annual review letter, a recommendation rationale, a meeting note. Each one is a regulated document under COBS 9A. Each one carries Consumer Duty obligations. The FCA’s 2025 financial advice firms survey found that 57% of firms had changed or were changing their record-keeping approach in response to recent regulatory scrutiny.

The second is service delivery at scale. The FCA’s February 2025 review of ongoing advice services examined data from 22 of the largest financial advice firms covering seven years of delivery. The headline finding: scheduled suitability reviews were delivered in around 83% of cases. In a further 15% of cases, clients declined the review or did not respond. Fewer than 2% of cases showed firms making no effort to deliver the promised service.

The FCA did not conclude this was systemic. It did, however, say something more pointed. Firms should be able to clearly evidence that they have delivered the services they were required to deliver, whether under contract or under regulatory rules. The regulator will reassess later in 2025 how firms have responded and what actions they have taken.

The October 2025 multi-firm review of consolidation in financial advice and wealth management sharpened the supervisory tone further. The FCA acknowledged that private capital has modernised much of the sector, but warned that growth without governance has created avoidable risks. PE-backed consolidator groups will now be tested to institutional governance standards, with leverage, integration, and conduct controls all in scope.

AI is the operational lever wealth firms are pulling to handle the volume and consistency the new supervisory environment requires. It is also the lever the FCA will look at first when it asks how a wealth firm is supervising advice quality across its full client base.

The three charges that follow are the ones that map most directly onto wealth and advice operations.

For a full breakdown of all five governance gaps, see the complete framework → AI Governance in UK Financial Services: The Accountability Framework

The suitability record under AI: why the audit trail is now six layers deep

When advisers wrote suitability records themselves, the audit trail was the document. The fact-find sat behind it. The recommendation rationale lived inside it. The adviser’s signature closed it.

When AI drafts the record, the audit trail has more parts. It now includes the underlying conversation, the AI’s transcript of that conversation, the data the AI pulled in from the fact-find or CRM, the prompt or rules the AI used to draft the report, the output the AI produced, the adviser’s edits, and the final approved version sent to the client. Six layers. Each one is potentially relevant to a complaint, a FOS adjudication, or a thematic review.

The FCA’s ongoing advice review identified weaknesses at several firms in exactly this area. The regulator pointed to ineffective processes and controls for evidencing that services had been delivered, and insufficient management information for senior management to oversee delivery. A polished suitability report does not satisfy the obligation if the firm cannot show how it was produced, what it drew on, and what the adviser did to it before it went to the client.

A defensible AI-assisted suitability record contains:

The original interaction in retrievable form, whether audio, video, or transcript
The structured data extracted from that interaction, mapped to the relevant fact-find fields
The recommendation logic, including which client circumstances drove which product or fund selection
The disclosed risks, limitations, and any conflicts of interest
The adviser’s review of the AI draft, including amendments and approval
The version sent to the client and any subsequent correspondence

This is what the Financial Ombudsman Service will want to see when a complaint lands four years from now. It is what the FCA will want to see in a Section 166 review. It is what a senior manager will need to evidence under SMCR if their AI tool generated advice that turned out to be unsuitable.

Most wealth firms cannot reconstruct all six layers today. Many cannot reconstruct any of them in a structured way. The AI sits in a tool the adviser uses. The output lands in the CRM. The intermediate steps disappear.

For a deeper look at the audit trail problem, see Count II: AI Advice Without an Audit Trail.

Consumer Duty at wealth scale: why sampling fails the firms with the most to prove

Most wealth managers QA between 3% and 5% of client interactions. The sample looks fine. It always does.

The problem with sampling at that rate in a wealth context comes down to the asymmetry between what the sample shows and what the FCA expects to see. Consumer Duty asks firms to monitor outcomes across the client base, not across a sample of the client base. The cross-cutting rules apply to every interaction.

When a wealth firm relies on a small sample, three categories of risk stay hidden.

The first is the long-tail vulnerability indicator. The FCA’s March 2025 review of firms’ treatment of vulnerable customers found that only 42% of vulnerable consumers had disclosed their circumstances to any of their financial services providers. The other 58% are present in the data but require detection through patterns of language, hesitation, or financial behaviour. A 3% sample will catch a handful of obvious cases and miss the rest entirely.

The second is the systemic drift. When an AI tool starts producing recommendations that subtly favour certain products, or when a particular adviser’s notes start showing patterns of unsuitable assumptions, those problems compound across hundreds of clients before they show up in a sample. By the time the sample catches them, the redress exposure is meaningful.

The third is the ongoing service gap. The FCA’s ongoing advice review found 15% of clients either declined a review or did not respond. A subsequent TCC poll of wealth managers found that only 23% had a structured indicator in their practice management system showing the client had received their review. Sampling does not surface a problem like this, because the problem is in the cases that did not happen, not the cases that did.

The structural answer is full coverage. When wealth firms monitor every interaction rather than a sample, several patterns surface that were previously invisible.

Vulnerability indicators across a full client book reveal which segments and which advisers are encountering vulnerable customers most often. That data feeds back into training, supervision, and product design.

Consistency of advice across advisers becomes measurable. If one adviser’s recommendations diverge meaningfully from peers handling similar client profiles, the supervisory layer can investigate before the FCA does.

Service delivery against contractual commitments becomes quantifiable. Firms can show, file by file, that every client in the ongoing service tier has been contacted, reviewed, and documented.

For a deeper look at the sampling problem, see Count III: Why Sampling 3% Falls Short of Consumer Duty.

Guidance quality and the vulnerable customer test: where wealth firms have the most to lose

Consumer Duty product and service outcome standards apply to every customer interaction. They do not apply on average. They do not apply to the median client. They apply to each one.

When an AI tool gives poor guidance, it does so at a scale no human workforce could match. A misinterpretation of a client’s capacity for loss can replicate across thousands of suitability reports before anyone notices. A flawed assumption about a vulnerable customer’s circumstances can propagate through every annual review the AI touches.

The FCA’s FG21/1 guidance on vulnerable customers requires firms to take particular care that vulnerable customers receive outcomes as good as those of other customers. The regulator’s data suggests around half of UK adults show one or more characteristics of vulnerability. The 42% disclosure rate means firms cannot rely on customers telling them. They have to detect it themselves, in the conversation, before the advice is given.

This is the test wealth managers should be applying to any AI tool that touches advice. Not “does the output look good” but “can we show, for every vulnerable customer who came through, that the AI flagged the indicators and the adviser acted on them.”

Most QA in wealth management is retrospective. The conversation happens. The note gets written. Someone reviews it later. By then, the moment to support the customer has passed.

Real-time guidance changes that. When an AI monitors the live interaction, identifies vulnerability indicators or compliance risks as they emerge, and surfaces guidance to the adviser in the moment, the adviser still makes the decision. The AI flags what to consider. The customer gets the slower question, the adjusted recommendation, the proper care.

This matters under Consumer Duty because the obligation is to deliver good outcomes, not to record poor ones accurately. Catching a vulnerability signal in the conversation lets the adviser slow down. Catching it in the file review after the fact lets the firm document the problem after it has already happened.

For a deeper look at the guidance quality problem, see Count IV: When AI Guidance Goes Wrong at Scale.

What this looks like in a wealth firm’s operating model

For a wealth manager working through the AI governance question, the practical implications break down into four areas.

Adviser workflow. Every AI-assisted interaction needs to produce a record that can be retrieved, reviewed, and defended. The tool the adviser uses has to capture the conversation, structure the data, draft the regulated output, and preserve the source material in a way that connects back to the final document.

Compliance monitoring. QA at 3% will not satisfy the FCA’s emerging expectations on outcome monitoring. Wealth firms need to move toward 100% coverage on the interactions that matter most: advice meetings, annual reviews, vulnerability-sensitive conversations.

Senior management oversight. Senior managers are personally liable under SMCR for the AI tools deployed in their area of responsibility. That liability becomes meaningful evidence only if the manager can produce a documented record of how the tool was selected, tested, supervised, and reviewed.

Consolidator-specific obligations. PE-backed groups operating under the post-October 2025 consolidator regime have to demonstrate that conduct controls scale across acquired firms. AI is the operational lever for that scaling. The AI tools deployed across the group need to produce consistent records, consistent monitoring, and consistent reporting. Otherwise the governance gap the FCA highlighted in the consolidator review becomes the firm’s exposure.

What the regulatory timeline looks like for wealth firms in 2026

The pressure is concentrated. Several deadlines fall in the same window.

The FCA has said it will return later in 2025 to assess how firms have responded to the ongoing advice review and whether appropriate remedies are being applied. The October 2025 consolidator review brought bilateral follow-up with named groups, with industry-wide expectations now clearly set. The Mills Review is examining what oversight senior managers must have over AI decisions, with comprehensive guidance expected by the end of 2026. The targeted support regime, relevant to firms delivering AI-driven recommendations to retail customers, came into force in April 2026.

For wealth firms reading this article, the practical question is which deadline arrives first, and what the firm’s defence brief looks like when it does. Firms that arrive at a thematic review or a Section 166 with documented AI oversight, 100% interaction monitoring, and audit-ready suitability records will pass. Firms that arrive with a governance policy, a 3% sample, and a generic AI assistant behind their advice process will not.

How Aveni helps wealth firms build the defence brief

The case-by-case work of building the infrastructure described above is what Aveni is built for.

Aveni Assist handles the adviser-facing layer. It records meetings, drafts suitability reports, updates CRM records, and produces client correspondence. Every output references back to the source. The product is built around the six-layer audit trail described earlier in this article, so a wealth firm using Assist can produce the full chain of custody for any AI-assisted suitability record on demand. Assist integrates with Intelliflo Office, Xplan, MS Teams, Zoom, Google Meet, and Webex.

Aveni Detect handles the compliance and operations layer. It monitors 100% of customer interactions, identifies risk and vulnerability indicators, prioritises cases for review, and produces the management information senior managers need under SMCR and Consumer Duty. For a wealth firm moving from sampling to full coverage, Detect closes the gap between policy and evidence.

Both products run on FinLLM, the domain-specific language model Aveni co-developed with major UK financial services partners including Nationwide and Lloyds Banking Group. Think of FinLLM as Exhibit F: provenance evidence. When a regulator or an EU AI Act assessor asks what the model behind the firm’s advice tools was trained on, FinLLM has the answer on file.

The firms that will answer the FCA’s questions most convincingly are the wealth managers that stopped treating governance as a policy exercise and started treating it as an infrastructure problem. Aveni is the expert witness for the defence. Not the prosecution. We help wealth firms build the case, not fear the courtroom.

Where does your wealth firm stand?

The three governance gaps covered in this article are the ones that apply most directly to UK wealth managers and advice firms. The other two charges in the AI on Trial series cover senior manager accountability and model provenance, and may also apply depending on your AI deployment.

See how Aveni helps UK wealth firms build the defence brief →

This article is part of Aveni’s AI on Trial: The Burden of Proof campaign series.

Read the full series: