Three AI Agents, One Wealth Management Firm, 64% More Client Reviews Completed

A four-person Manchester wealth management firm was processing 28 annual reviews a quarter. Their lead adviser was spending 31 hours a week on review preparation, portfolio commentary, and client reporting — work that had to be done but kept her away from the relationships that drove referrals. Three AI agents changed the maths. Twelve weeks later, they completed 46 reviews in the same quarter, recovered 26 hours of weekly capacity, and spent £85 a month to run the entire system. <figure> <img src="https://images.unsplash.com/photo-1507679799987-c73779587ccf?w=1200&q=80" alt="AI agents helping a UK wealth management firm complete 64% more client reviews with the same team size" width="1200" height="630" loading="lazy" /> </figure> <h2>The Problem: Compliance Admin Was Eating the Practice</h2> <figure> <img src="https://images.unsplash.com/photo-1579621970563-ebec7560ff3e?w=1200&q=80" alt="Wealth management compliance burden — FCA Consumer Duty, suitability letters, and annual review preparation consuming adviser capacity at a UK wealth management firm" width="1200" height="800" loading="lazy" /> </figure> The firm — four people, 140 clients, £62m under advice — had a problem that looked like a capacity problem but was actually an admin problem. Their lead adviser, Sarah, was technically billing at 70% capacity. But when she mapped her week, 31 of her working hours were disappearing into work that had nothing to do with advice. Annual review preparation consumed the most time. For each review, Sarah pulled portfolio performance data from two platforms, drafted a suitability letter that referenced both the current portfolio and the client's stated objectives, prepared a summary of market conditions, and updated the firm's internal risk record. Each review took between two and three hours of preparation before she sat down with the client. With 28 reviews per quarter, that was up to 84 hours of preparation work every quarter — largely repetitive, largely templated, almost entirely automatable. The FCA's Consumer Duty rules had added another layer. Every client communication now needed a documented rationale. Portfolio commentary — explaining what happened in markets and why, in plain English — had to be attached to every quarterly statement. Sarah was writing these by hand, reviewing regulatory guidance, and then having her operations manager check each one. The process worked. It was also consuming her at a rate she could not sustain. <blockquote>Sarah was spending more time preparing for client conversations than having them. The work itself was important, but 80% of it followed the same pattern every time. That pattern is exactly what AI agents are designed to handle.</blockquote> The firm had looked at CRM automation tools before, but nothing connected review prep to portfolio data to suitability letter generation without significant manual intervention. What they needed was not a feature. They needed agents that could work across their actual systems. <h2>Agent 1: The Annual Review Preparation Agent</h2> <figure> <img src="https://images.unsplash.com/photo-1677442135703-1787eea5ce01?w=1200&q=80" alt="AI review preparation agent workflow for wealth management — connecting CRM, portfolio platforms, and suitability letter generation automatically" width="1200" height="800" loading="lazy" /> </figure> The first agent tackled the highest-friction point: review preparation. Built in n8n and connected to the firm's two portfolio platforms via API, the agent runs automatically for every review scheduled in the next 14 days. When triggered, it pulls the relevant client record from the CRM, retrieves the last 12 months of portfolio performance data, compares the current allocation to the client's recorded risk profile and objectives, and identifies any drift that needs addressing. It then drafts a review pack: a performance summary, a meeting agenda, and a first-draft suitability letter that references the client's stated goals, their current holdings, and any changes being recommended. Sarah reviews the draft before each meeting, edits anything that needs personalising, and approves it. The approval triggers the letter to be saved to the client record. What used to take two to three hours per review now takes 20 minutes — largely a review-and-personalise workflow rather than a create-from-scratch one. The agent also handles post-review actions: when Sarah marks the review complete and records the outcome, it updates the client record, logs the regulatory documentation, and schedules the next review in the CRM automatically. This is the same logic that drives every effective <a href="/blog/automate-client-reports-ai-agent">AI reporting agent</a>: identify the pattern that repeats across every instance of a task, encode it into an agent, and leave the judgment calls — personalisation, nuance, relationship context — to the human. The pattern handles itself. The adviser handles the exceptions. <h2>Agent 2: The Portfolio Commentary Agent</h2> The second agent addressed the Consumer Duty compliance burden: specifically, the quarterly portfolio commentary that had to accompany every client statement. Previously written by hand for each of the 140 clients, this was 140 unique documents per quarter, each requiring a review of what had happened in markets, why, and what it meant for the client's specific portfolio. The commentary agent pulls market data from three sources each quarter — bond yields, equity indices, and economic indicators relevant to the firm's client base — and uses Claude to generate a plain-English summary calibrated to three complexity levels: simplified for clients who requested accessible communication, standard for most clients, and detailed for the handful who wanted full analysis. It then generates a client-specific paragraph for each client's record, referencing their actual portfolio performance against the market context and noting any quarters where their portfolio materially outperformed or underperformed the relevant benchmark. The operations manager reviews a sample of 15 to 20 commentaries each quarter before approving the batch for distribution. The whole process, which previously took 11 hours per quarter, now takes 45 minutes of human time. <blockquote>This is exactly the kind of work where AI operating systems justify their cost. The task requires genuine knowledge — understanding markets, interpreting performance, writing clearly for non-specialist readers. But 90% of the pattern repeats. The agent handles the pattern. The team handles the edge cases and the sign-off.</blockquote> <h2>Agent 3: The Client Communication Agent</h2> <figure> <img src="https://images.unsplash.com/photo-1590283603385-17ffb3a7f29f?w=1200&q=80" alt="Key results from the AI wealth management operating system — 64% more client reviews, 26 hours recovered per week, £85 per month running cost, 34% improvement in client response rate" width="1200" height="800" loading="lazy" /> </figure> The third agent focused on proactive client communication — the touchpoints that happen between reviews and that most wealth managers know they should send but rarely find time to personalise at scale. The communication agent monitors three triggers for each client: significant portfolio events (a position moving more than 5% in a week), upcoming key dates (birthdays, policy renewal dates recorded in the CRM, ISA season), and inbound client queries routed through the firm's contact form. For each trigger, it drafts a personalised message for Sarah or her associate to review and send with one click. The agent does not send communications autonomously. Everything passes through a human approval step — a deliberate choice, and the right one for a regulated firm. But the difference between "Sarah writes this from scratch" and "Sarah reads a well-drafted message and clicks send" is the difference between 12 minutes and 90 seconds. Across 140 clients, that difference is the capacity that makes proactive service sustainable. The <a href="/blog/human-in-the-loop-ai-agents-uk">human-in-the-loop architecture</a> is not a concession to caution — it is what makes the system trustworthy enough to rely on at scale. The firm's client response rate to proactive communication increased by 34% in the first quarter after launch. More clients engaged. More referrals followed. The effect was not from sending more messages — they sent roughly the same volume as before. The effect was from sending better-timed, better-personalised messages without Sarah spending three hours a week composing them. <h2>The Numbers After 90 Days</h2> <figure> <img src="https://images.unsplash.com/photo-1554224155-8d04cb21cd6c?w=1200&q=80" alt="90-day results for the wealth management AI operating system — reviews per quarter from 28 to 46, 26 hours recovered weekly, portfolio commentary time from 11 hours to 45 minutes" width="1200" height="800" loading="lazy" /> </figure> Twelve weeks after deploying all three agents, the firm tracked the results against the quarter before launch: <ul> <li>Reviews completed per quarter: 28 → 46 (64% increase)</li> <li>Weekly capacity recovered: 26 hours (review prep from 31 hours to 5; compliance documentation from 11 hours to 2; client communication drafting from 8 hours to 1)</li> <li>Portfolio commentary production time: 11 hours per quarter → 45 minutes</li> <li>Client response rate to proactive communication: +34%</li> <li>Monthly running cost: £85 (n8n cloud plan plus Claude API usage at current volumes)</li> <li>Regulatory near-misses logged: Zero — the structured documentation workflow had actually improved the firm's compliance record.</li> </ul> The recovered capacity went directly into business development. Sarah used the first 12 weeks to pursue six high-net-worth prospects she had not had time to meet properly before. Two became clients. The new AUM added in the quarter exceeded £8m. The connection between agent-recovered time and new revenue was direct and measurable. <blockquote>An AI operating system for a wealth management firm does not replace the adviser. It removes the structural drag that prevents the adviser from doing what only they can do: building trust with clients, exercising judgement on complex cases, and developing the relationships that drive referrals.</blockquote> The firm is now planning a fourth agent — a client onboarding agent that will handle the documentation-heavy early phase of bringing on a new client. Based on the pattern established by the first three, they expect it to reduce new client onboarding time from 14 hours to under three hours per client, making the economics of taking on smaller clients viable for the first time. <h2>What This Means for Your Practice</h2> Wealth management is one of the UK professional services sectors where AI agents generate disproportionate returns — not because the work is simple, but because a large proportion of it is systematically patterned. Review preparation follows the same structure for every client. Portfolio commentary follows the same structure every quarter. Compliance documentation follows the same structure for every interaction. Agents are built for exactly this: high-volume, pattern-rich work that requires knowledge to do well but does not require unique creativity every time. The same three-agent structure — review preparation, portfolio commentary, client communication — works for any regulated financial advisory business. IFAs, discretionary investment managers, financial planning firms, pension advisers. The specific system connections change; the underlying architecture does not. The build cost for this system was £3,600 all-in. The monthly running cost is £85. The first two new clients Sarah converted using her recovered capacity paid that build cost inside three months. The <a href="/blog/ai-financial-planning-firm-case-study">IFA case study</a> covers a similar pattern at a financial planning practice — if you want to see how the same agent architecture adapts across firm types and regulatory contexts, that is a useful companion read. And if you want to see how this would work for your practice specifically — what the three agents would look like, what they would connect to, and what the realistic capacity recovery would be — <a href="/contact">get in touch</a>. We build AI operating systems for UK professional services firms, and the gap between where most practices are now and where they could be is almost always a systems problem, not an advice problem. We fix the systems.