The question worth asking of any AI tool is a plain one: what will this do in my firm, on a Tuesday afternoon, when a paraplanner needs it to work? Plenty of firms sit through demo after demo without getting an answer, because demos aren’t designed to give one.

The AI market in 2026 isn’t short of capability. It’s short of honesty about the gap between what a system can do in a controlled setting and what it will do reliably in yours. This guide tries to close that gap for advice and wealth firms, without dismissing AI or overselling it.

What can today’s AI actually do well

Today’s AI is very good at working with text that already exists.

Summarising a long document, drafting a first version of a letter from structured notes, extracting key terms from a contract, reformatting data between systems, answering questions about a fixed set of documents: current AI does these reliably, and the time saved is real and measurable. For a financial planning firm, that covers a good share of the administrative and documentation work currently done by expensive people.

The reason is structural. Language models are trained to predict plausible text. When the task is “produce text that looks like this source material, structured this way”, the model is doing exactly what it was built for. The ceiling is high, and the floor is higher than most people expect.

So the question isn’t whether AI can help your firm. It’s whether you’ve matched it to a task it can do.

Where AI falls down, and why it matters in a regulated context

Today’s AI is unreliable when it needs to be correct rather than plausible.

Language models hallucinate. They produce text that reads as confident and well organised but contains errors, sometimes minor, sometimes fundamental. Vendors aren’t about to fix this. It’s a property of how these systems work, and it will stay that way. The models don’t know what they don’t know, and when they lack information they fill the gap with what sounds right.

For most consumer uses that’s a minor nuisance. For a firm producing a suitability report, a client communication or any document under FCA scrutiny, it’s a serious operational risk. Any AI output used in a regulated document or decision needs human review before it’s finalised. That isn’t optional, and no vendor can responsibly tell you otherwise.

“Grounded” AI does much better: systems that may only answer from a specific, defined set of documents rather than from general training. Google, for example, claims a hallucination rate below 1% for NotebookLM used this way. Tools that work like this have a real compliance advantage for regulated use, but even then the human review step doesn’t go away. It narrows the risk rather than removing it.

The governance gap is the real risk right now

The bigger risk for most firms is governance, not the technology. Decisions about AI deployment, including staffing and process redesign, are often made on an understanding of what AI can do that doesn’t match the reality of running it day to day. What fails then isn’t mainly the technology. It’s governance: the gap between what a firm’s AI tools can do in theory and what its processes, oversight and staff can actually manage.

The FCA and Bank of England issued a joint statement in May 2026 signalling that AI governance for UK-regulated firms is moving from voluntary best practice towards active regulatory expectation.[1] The direction is clear, even if the specific rules are still forming.

In practice, deploying AI without documented oversight, defined human review points and clear accountability for outputs is no longer only a reputational question. It’s the kind of operational control gap regulators will ask about. The UK governance guide covers what that oversight should look like.

What to do, a practical starting point for a firm your size

You don’t need a large project to start, just an honest one.

Map what AI is already in use

This takes more work than it sounds. Staff are already using AI tools (ChatGPT, Copilot, Gemini and others) whether or not a policy allows it, and you can’t govern what you can’t see. A structured conversation with team leads, rather than a formal audit, is usually enough to get the picture quickly.

Separate the tasks where AI is low-risk from those where it isn’t

A first-pass meeting summary from structured notes is low risk: it’s easy to review, and the person checking has the full context to catch errors. Content that goes into a suitability letter with little oversight is high risk: errors are hard to catch under time pressure, and the regulatory exposure is real. Most firms have both kinds of use going on at once, without ever having made that distinction.

Establish a review step for any AI output that touches a regulated document or client communication

This needn’t be a committee or a policy document nobody reads. It’s a practical habit: the person who uses the AI output is responsible for checking it before it goes anywhere, and that responsibility is written down.

Apply the same supplier due diligence to AI tools that you’d apply to any critical third party

Ask where the data goes, who hosts it, and what the retention and deletion terms are. For firms that can’t send client data through public cloud services, on-premise or private deployment is now a normal commercial option, not a specialist purchase. The guide to reading an AI vendor contract lists the clauses to check.

The capability levels worth knowing

I use a simple three-level framework with firms, to avoid both spending too little and building too much.

  • Level 1 (education): The answer already exists inside a tool you pay for. A better prompt, a template, fifteen minutes of guidance. Free or nearly free, and available today. Most firms have a lot of unused capability here.
  • Level 2 (integration): Two or three tools that need to talk to each other. Configuration work, not engineering. Typically days to weeks, and a few hundred to a few thousand pounds. This is where automating routine workflows lives.
  • Level 3 (custom build): Real engineering. Custom pipelines and complex logic over four to twelve weeks, scoped carefully, and only when the problem needs it.

Always try Level 1 first. Quite often, a firm that gets in touch thinking it needs a Level 3 build needs Level 1 or 2. That’s no criticism of the firm. The vendor market tends to blur the distinction, because Level 3 is where the revenue is.

What this means for an advice firm

Current AI is useful for a financial advice firm. It can cut the time spent on documentation, drafting and finding information in ways you can measure. It isn’t reliable as an autonomous decision-maker, it isn’t safe without human oversight in regulated work, and it doesn’t replace the judgement your qualified staff apply to client situations.

The firms getting real value from AI now aren’t the ones with the most tools. They picked specific, bounded tasks where AI helps, built simple review habits around the output, and kept clear of the governance gap that makes regulators nervous.

This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.

Sources

[1] FCA and Bank of England joint statement on AI governance, May 2026, signalling shift from voluntary best practice toward active regulatory expectation for UK-regulated firms.