The managing partner of a 22-person UK accountancy practice has a quarterly utilisation report on his desk. The numbers are fine. The numbers were fine last quarter. He has been at a regional roundtable that morning where two of his peers mentioned in passing that their reconciliation queues are now mostly automated. Both peers run firms within ten miles of his. Both are within five staff of his headcount.
He has done the maths in his head on the way back to the office. What reconciliation automation does to a peer firm’s margin structure. What it does to his firm’s margin if peers can take work at prices he cannot match. He is not panicking. He is calculating.
What does the Wolters Kluwer fourfold jump actually mean?
The headline is from Wolters Kluwer’s Future Ready Accountant report. AI adoption at accounting firms nationwide jumped from 9 percent in 2024 to 41 percent in 2025, more than fourfold in a single year. 77 percent of firms surveyed said they plan to increase AI investments. That is the largest one-year jump in any sector tracked across the 2024 to 2025 cycle.
The shape of the jump matters more than the absolute numbers. The Big Four and national firms are using proprietary tooling and increasingly agentic AI to handle audit, tax, and advisory work at scale. Mid-size and smaller firms are slower adopters. They face barriers including privacy concerns and the practical issue that AI tools require substantial structured data to be effective, and many smaller practices have data scattered across multiple systems and formats.
For the 10 to 50 person practice, the most-cited entry point is Microsoft Copilot. It integrates with the Office and Excel tooling smaller firms are already using. It automates routine work without firm-wide data migration. It is priced inside an existing licence rather than as a new line item.
The CPA.com 2025 framework names four phases of AI adoption (Discover, Pilot, Operationalise, Scale). Most smaller firms sit in phase 1 or 2. The Big Four are in phase 4. The gap between those two is what the regional roundtable conversation is about.
Why is Microsoft Copilot becoming the SME default?
Microsoft Copilot has become the de facto AI choice for 10 to 50 person accountancy firms because it sits inside the Office and Excel licence those firms already pay for. The marginal cost is roughly £20 per user per month, the marginal training cost is small, and the integration risk on existing workflows is close to zero.
The trade-off is depth. Copilot handles general productivity work well: drafting, summarising, exploring data in spreadsheets, surfacing patterns. Specialist platforms like Trullion are built for accounting-specific workflows: AI-powered lease accounting, audit workflow automation, audit-trail support with domain-specific compliance features. They cost more, take more setup, and pay back faster on firms with high-volume routine work.
The pattern across the data is consistent. Smaller firms adopt generic tools first, then migrate to specialist platforms as ROI is demonstrated and compliance burden justifies the spend. That is rational behaviour. It also means the smaller firm is two phases behind the larger firm at any given moment, and the gap is closing only if the smaller firm actively closes it.
Three use cases that work at 10 to 50 person practice scale
The three use cases that produce measurable ROI at this firm size are bank reconciliation and transaction matching, generative AI for client memo and tax position drafting, and AI-assisted tax return review with risk flagging on outliers. Each operates on structured data, integrates with existing accounting systems, and keeps the accountant in the final-judgment position.
Bank reconciliation is the highest-volume routine workflow in most practices. AI tools (built into Copilot, QuickBooks, or specialist platforms like Trullion) match transactions against records, flag discrepancies, and suggest adjustments. A 10 to 50 person firm processing 20 or more bank accounts saves 2 to 4 hours per account per month. At £50 per hour loaded cost, that is £2,000 to £8,000 per practice per year. The payback maths on Microsoft 365 with Copilot turn positive in the first month.
Tax memo and client communication drafting is the second use case. Accountants prepare memos explaining tax positions, audit findings, or advisory recommendations to non-specialist clients. A generative AI system drafts the initial memo. The accountant reviews, refines, and signs. The accountant remains the professional in the loop, the writing time drops materially.
The third use case is AI-assisted tax return review. AI flags outliers (an unusually high deduction relative to prior years, an inconsistency between schedules) for human review. The CPA.com 2025 framework calls this “agentic AI” with the accountant as reviewer rather than doer. Throughput rises while professional accountability stays intact.
What does the privacy red line look like in practice?
The hard rule is that client tax data, payroll data, and personal financial information must not be fed into general-purpose AI tools without explicit Data Processing Agreements that prohibit retraining on the data. CPA.com’s 2025 report names privacy-by-design as the new differentiator between profession-specific platforms and generic AI tools. SOC 2, GDPR, opt-in training protocols are the operational vocabulary.
The risk is not academic. A practice that drops a client’s payroll spreadsheet into a public AI tool to “ask Copilot to summarise the year on year change” has just shipped that data to a third party with terms most accountants haven’t read. Whether that constitutes a breach depends on the tool, the configuration, and the jurisdiction. Whether it triggers a regulatory inquiry depends on whether the client finds out.
The fix is straightforward. Firms with Microsoft 365 enterprise contracts can configure Copilot with data residency, no-training, and audit settings that satisfy ICAEW and ACCA expectations. Firms using specialist platforms like Trullion get those features by default. Firms using public ChatGPT or public Claude on client work without enterprise contracts are running an unmanaged risk.
What does the regulatory layer require of you?
The regulator stack for accountancy AI is structured but not yet prescriptive. UK firms work under ICAEW’s 2026 guidance on regulatory compliance risks with AI agents, ACCA Global’s frameworks for governance and conformity, and the FCA where the firm offers regulated investment advice. US firms work under AICPA guidelines for forensic and valuation services and SEC oversight on investment advisory.
The common principle across all of these is professional accountability. Accountants stay responsible for AI outputs even when delegating to an agentic system. The ICAEW 2026 guidance is explicit on this point. If an AI tool flags or misses something material, the accountant cannot point to the AI as the cause; the firm carries the regulatory and professional indemnity exposure.
The implication for a 22-person practice is that AI deployment needs an internal owner. Someone in the firm has to write the policy, sign off on the tools, log the audit trail, and answer to the regulator if asked. In practice that is usually the senior partner with the strongest IT literacy, supported by the firm’s IT-managed services provider on the technical side.
What does the maths look like for a 22-person practice?
Bank reconciliation pays back the fastest. A 22-person practice processing 25 bank accounts at 3 hours saved per account per month yields 900 hours per year. At £50 per hour loaded cost, that is £45,000 of recovered time per year on a tool costing roughly £5,000 per year for a Microsoft 365 with Copilot deployment across the partner and senior staff.
The number assumes the practice actually redirects the recovered time to billable client work. If utilisation is already at ceiling, the saving is in cost not revenue, which still matters but doesn’t compound the same way. Tax memo drafting is slower to pay back, typically 6 to 12 months because of training and template setup. Tax return risk flagging needs more pilot validation; initial setup costs sit in the £5,000 to £15,000 range.
The harder-to-model return is competitive. The roundtable peers automating reconciliation can either take work at lower prices or run higher margins at the same prices. The 22-person practice that doesn’t move loses on one of those axes within the next 12 to 18 months. The data on the 9 to 41 percent jump in a single year suggests the curve is steepening, the gap is not flattening on its own.
What is the actual next move?
The actual next move is to pick one workflow, usually bank reconciliation, run a 60-day pilot with Microsoft Copilot or a specialist platform configured for proper data handling, and audit the time saved against the time spent on AI rework. Concrete pilot, real numbers, written policy. The phase 2 decision comes from those numbers, not from peer pressure.
For most 10 to 50 person practices, the pilot looks like this. One partner takes ownership. The firm’s IT-managed services provider configures Copilot with the right enterprise contract terms. The pilot runs across 5 to 10 client accounts for 60 days. The senior partner audits the time saved per account, the time spent on AI rework, and the audit-trail completeness. The board reviews the numbers and decides what phase 2 looks like.
The 22-person practice managing partner with the utilisation report on his desk is doing the right work. The peer firms in the roundtable are at phase 2. The data says he has 12 to 18 months to be there too, before the gap becomes a margin problem rather than a curiosity. Acting against the right benchmark, the sector at 41 percent rather than the all-business average, is the practical first answer.
If you would like to walk through this for your practice specifically, book a conversation.
