Intelliflo’s 2026 survey found that 74% of financial advice firms are now using AI in some part of their operations, up from 43% the year before[1]. That is a fast move. But adoption is not transformation. Most of what those firms are doing sits at Level 1: a prompt here, a template there, a ChatGPT subscription that a couple of advisers use on their own initiative. The gap between that and a genuinely AI-enabled firm, one where AI is embedded in workflows, overseen properly, and delivering measurable capacity, is still significant.

The good news is that the gap is closable in twelve months, if you go in with a plan rather than a series of experiments. This is that plan.

Why a phased approach matters more than you might expect

Most firms that stall on AI do so not because they lack ambition but because they try to do too much at once. They buy a tool, run a pilot, get inconsistent results, and conclude that AI “isn’t ready” for their firm. What they ran into was not an AI problem. It was an implementation problem.

A phased approach also matters for a second reason: the FCA is watching. The regulator has been clear that AI in regulated firms must be governed, auditable, and human-supervised[2]. Firms that layer AI onto broken or undocumented processes end up with broken, undocumented AI. The sequence below is designed to give you a foundation before you build on it.

The discipline is not to ask “what can AI do?” but “where is human time spent on work that AI can support, with a human still responsible for the outcome?”

Months 1 and 2: map what you actually do before you automate any of it

The first two months are not about AI. They are about your processes.

First, conduct an honest workflow audit. Walk through a client from initial enquiry to annual review. Write down every step. Note who does it, how long it takes, and whether there is a documented process or whether it lives in someone’s head. Pay particular attention to the handoffs: between adviser and paraplanner, between paraplanner and compliance, between back office and client-facing teams.

Second, identify the highest-friction points. These are usually: meeting preparation, suitability report drafting, client communication, compliance checking, and data entry across systems that do not talk to each other. You are looking for work that is high-volume, relatively structured, and currently eating time that could go to clients.

Third, assess your data hygiene. AI is only as good as what you feed it. If your client records are inconsistently completed, if your CRM is a mess, or if key documents live in a shared drive with no naming convention, fix that first. This is unglamorous but it determines whether Month 6 works.

By the end of Month 2, you should have a written map of your top five workflow bottlenecks and a clear picture of where your data is and how clean it is. That document is the foundation for everything that follows.

Months 3 and 4: build internal capability at Level 1

This is where AI enters, but it enters carefully.

Choose two or three of your mapped bottlenecks and address them with tools your team already has access to, or low-cost additions. This is Level 1: education and prompting, not integration or custom builds. The goal is to build confidence and establish what good AI-assisted work looks like at your firm, before you automate it.

The two most productive Level 1 wins for most advice firms are:

Meeting preparation and follow-up. A well-crafted prompt, working from your client record and agenda, can produce a structured meeting brief in minutes. A prompt working from a voice transcript or notes can produce a draft follow-up letter or action list. Both save time. Neither removes adviser judgement from the process.

First-draft suitability sections. AI can draft component sections of a suitability report from structured client data. The adviser or paraplanner reviews, edits, and is accountable for the final document. It does not draft the advice. It drafts the description of the client’s circumstances and the explanation of the recommendation, which the professional then checks against the actual advice position.

One important tool choice here: for regulated, client-facing outputs, use a grounded AI tool wherever possible. NotebookLM, for example, restricts responses to user-uploaded sources only, which significantly reduces the risk of the model introducing information that is not in your files[3]. Tools that can draw on open training data alongside your uploaded files, including ChatGPT Projects and Claude Projects in their standard configurations, carry a materially higher risk of generating content that goes beyond your source material. That distinction matters for compliance. Whichever tool you choose, human review of every output before it touches a client remains the primary control. Choose your tools deliberately, not by default.

By the end of Month 4, you should have two or three working prompt workflows that your team uses consistently, a clear policy on which outputs require human review before they go anywhere near a client (the answer is all of them), and a short internal log of where the tools have helped and where they have not.

Months 5 and 6: move to Level 2 integration where the evidence supports it

By now you know which workflows AI can genuinely support. Month 5 and 6 are about making those connections more durable, by integrating the tools into your existing stack rather than relying on manual copy-paste.

Level 2 integration typically means two or three tools talking to each other: your CRM, your document management system, and an AI layer in between. This is configuration work, not engineering. Tools like n8n, Make, or Zapier handle most of it. The builds typically take days to weeks and cost a few hundred to a few thousand pounds.

A realistic example for a mid-sized advice firm: a workflow that pulls a client record from your CRM, runs it through a structured prompt, and drafts a meeting brief into your document system, ready for the adviser to review and use. Nothing is sent to the client. Nothing is finalised. The human is still at every consequential step. But the work that used to take thirty minutes takes five.

One practical note on digital onboarding: a vendor study in regulated financial services found that digital onboarding can reduce KYC processing time by 34%[4]. That is a meaningful benchmark for a firm where the new client pipeline is a bottleneck, though figures from vendor research should be treated as indicative rather than independently verified. If onboarding is on your list of bottlenecks, this is the right phase to address it.

Months 7 and 8: establish governance before you scale

This is the phase most firms skip, and it is the reason most firms eventually stall or run into problems.

By Month 7, you likely have several AI-assisted workflows running. Before you add more, you need to answer these questions formally:

Who is accountable for each AI-assisted output? In a regulated firm, “the AI did it” is not an answer the FCA will accept. Every suitability recommendation, every client communication, every compliance document has a responsible human. Your AI governance framework needs to name those people and make their accountability explicit.

How are you logging AI use? If the FCA asks what role AI played in a specific client outcome, you need to be able to answer. That means keeping records of which workflows used AI assistance, which outputs were AI-assisted, and what human review took place. This does not need to be elaborate. It needs to be consistent and retrievable.

What is your review cadence? AI tools change. Providers update models, change terms, and occasionally pivot their products[5]. A workflow that works correctly in Month 6 may behave differently in Month 10 if the underlying model has changed. Build in a quarterly review of your AI workflows as a standing item.

A practical format for this: a one-page AI register that names each active workflow, the tool it uses, the human accountable for its outputs, and the last date it was reviewed. That document is your governance record.

Months 9 and 10: measure what has actually changed

You cannot manage what you cannot measure, and by this point you should have enough months of data to be honest about what AI has and has not delivered.

The metrics that matter for an advice firm are not impressionistic. They are: time per suitability report, from instruction to signed-off document. Time per meeting cycle. Volume of new clients processed per paraplanner per month. Error rate in client documents, measured by the number of compliance corrections raised. If you were logging these figures before you started (which the Month 1 to 2 audit should have captured), you now have a baseline to compare against.

Structured AI peer review approaches, applied to document checking in professional services contexts, have shown error reductions of around 50% and review time reductions of around 30%[6]. Whether your firm achieves similar results depends on how well your workflows are designed and how consistently your team uses them. But those figures give you a benchmark to test your own results against.

This is also the phase to be honest about what has not worked. If a workflow is not being used, ask why. Usually it is one of three things: it does not save enough time to be worth the change in habit, the output quality is not reliable enough, or the person accountable for the outcome does not trust it. Each of those has a different fix, and none of them is “add more AI”.

Months 11 and 12: position for what comes next

The final phase is not about adding more tools. It is about embedding what you have built and deciding deliberately what Level 3 looks like for your firm, if anything.

Level 3, in my framework, means real engineering: bespoke pipelines, orchestration across multiple agents, complex deterministic and probabilistic logic. For most advice firms of under 50 staff, nothing they need actually requires Level 3. The risk in Month 12 is that, having built confidence, a firm reaches for complexity it does not need.

The question to ask at this stage is not “what else can AI do?” It is: “where is the highest-value human time in this firm still being spent on work that AI could support?” The answer to that question, informed by twelve months of data, is where to focus next.

Two developments worth tracking as you plan beyond Month 12: intelliflo has signalled that its next platform generation will be an agentic open platform that orchestrates processes across firm systems, third-party tools, and its own applications[1]. If your practice management runs on intelliflo, the integration landscape is going to change materially. And the FCA has been consulting on reform of the Advice Guidance Boundary, with the aim of creating space for differentiated advice journeys[7]. Both of those are worth watching, not acting on yet, but watching.

What to do from here

If you are at the beginning of this, the most useful single action is the workflow audit in Month 1. Not a technology decision, not a tool evaluation. A map of where your people’s time goes and where the friction is. Everything else follows from that.

The firms that will be genuinely AI-enabled in twelve months are not the ones that move fastest. They are the ones that move most deliberately: mapping first, building on evidence, governing before they scale, and measuring honestly. That sequence is available to any firm that decides to follow it.

If you want to think through what this looks like for your specific firm, a discovery call with Cordrey Consulting is a sensible starting point.

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] Intelliflo, ‘Intelliflo Innovate 2026: Building the agentic open platform for financial advice professionals’, Intelliflo Insights, 17 June 2026. Available at: https://www.intelliflo.com/insights/thought-leadership/intelliflo-innovate-2026-building-the-agentic-open-platform-for-financial-advice-professionals

[2] FCA, ‘Rethinking regulation in the age of AI’, Financial Conduct Authority, 26 June 2026. Available at: https://www.fca.org.uk/news/speeches/rethinking-regulation-age-ai

[3] Everyday AI, ‘Ep 778: Codex Goes Remote Control, Claude Goes Small, NotebookLM Gets Super Powers’, 15 May 2026. [Cited for NotebookLM grounded response architecture.]

[4] Vendor-sourced. [Cited for 34% KYC processing time reduction figure. This is vendor research and should be treated as indicative rather than independently verified.]

[5] Digital Applied (2026) ‘ChatGPT Drops Below 50%: AI Assistant Market Share 2026’, Digital Applied. Available at: https://www.digitalapplied.com/blog/ai-assistant-market-share-2026-chatgpt-below-50-percent-analysis [Cited for AI platform volatility and market concentration risk.]

[6] Zen van Riel (2026) ‘AI Peer Review With 35% Error Reduction & 30% Faster Reviews’, Zen van Riel, AI Engineer Blog. Available at: https://zenvanriel.com/ai-engineer-blog/ai-peer-review-error-reduction-faster-reviews [Cited for error reduction and review time figures. Note: findings are from engineering workflows; analogous gains in regulated professional services contexts have not been independently verified.]

[7] Intelliflo, ‘Will hybrid advice dominate the next decade?’, Intelliflo Insights, 23 May 2026. Available at: https://www.intelliflo.com/insights/thought-leadership/will-hybrid-advice-dominate-the-next-decade [Cited for FCA Advice Guidance Boundary Review context.]