AI for Professional Services Firms

For a small professional services firm, the smart place to start with AI is the high-volume research, drafting, and admin work that fills your week, governed by one firm rule: never put identifiable client information into a public AI tool, and a human checks every output before it leaves the building. Get those two things right and AI becomes leverage. Get them wrong and it becomes a liability.

This playbook is for owners of small agencies, consultancies, accounting and bookkeeping practices, and law firms, the kind of 1 to 50 person business with no IT department. It spans the professions on purpose, because the use cases, the confidentiality duties, and the pricing pressure rhyme across all of them. For the broader question of whether your firm is set up to adopt AI at all, start with the pillar guide, Is Your Small Business Ready for AI?

Where does AI actually fit in a professional firm?

It fits the repeatable knowledge work that sits between you and the billable or advisory work you would rather be doing: research, first drafts, summaries, correspondence, and routine number-crunching. The pattern across professions is the same, even though the tasks differ. Tax research is the top reported use among tax and accounting professionals, used by 77% of them, ahead of return preparation at 63% and advisory at 62% (Thomson Reuters, 2025).

Are professional firms really using AI yet?

Increasingly, yes, and fast. Generative AI use across professional services nearly doubled in a year, from 12% of organizations in 2024 to 22% in 2025. Legal use rose from 14% to 26%, and tax firms nearly tripled from 8% to 21% (Thomson Reuters, 2025). Adoption is still early enough that a deliberate small firm can lead rather than scramble, but the direction is not in doubt.

Adoption is running ahead of governance, which is the risk. Among legal firms, 48% still have no formal GenAI policy, and across professions 64% of users have had no training on it (Thomson Reuters, 2025). That gap between using AI and governing it is exactly where avoidable trouble lives.

How much time does AI actually save?

Enough to matter, on the right tasks. In a controlled trial published in Science, professionals using generative AI finished writing tasks 40% faster and produced 18% higher-quality work (Noy and Zhang, Science, 2023). In a study of more than 5,000 support agents, AI lifted output by about 14% on average and 34% for the least experienced (Brynjolfsson, Li and Raymond, NBER, 2023). Thomson Reuters estimates the cumulative effect could free up nearly 240 hours per professional per year, an average value of about $19,000 each (Thomson Reuters Institute, 2025). Note the pattern in the research: AI helps the newest staff most, on well-defined tasks. That is a training and capacity story, not a redundancy one. Turning those saved hours into actual return is its own discipline, covered in the companion guide on the real ROI of AI for a small business.

The one rule: protect client confidentiality

This is the rule that separates safe AI use from a complaint to your professional body. The duty is the same whether you are bound by legal privilege or accounting confidentiality: client information must not go into a tool that could expose it. California's State Bar guidance states a lawyer must not input client confidential information into an AI tool that lacks adequate confidentiality protections, and should anonymize client details (State Bar of California, 2023). AICPA guidance tells CPA firms to prohibit sharing confidential client and firm information with generative AI tools, and to treat anything entered with the same care as a public social media post (Journal of Accountancy, AICPA, 2023). A practical way to write the rule is by tool tier.

The second duty is competence: you must understand the tool well enough to use it responsibly, which means reviewing what it produces. The American Bar Association's Formal Opinion 512, the first US ethics guidance on generative AI for lawyers, ties both duties together, holding that lawyers must safeguard confidentiality and verify AI output rather than rely on it (ABA Formal Opinion 512, 2024). AICPA puts the same point plainly: supervise and review AI output as you would the work of any other team member, because the tool is not a licensed professional. If your firm has staff using AI, this belongs in a short written policy, the subject of the companion guide on whether your small business needs an AI policy.

What happens when AI output goes unchecked?

The cautionary cases are now numerous and expensive, and they all share one cause: AI output sent onward without verification. Generative AI invents facts, citations, and quotations that read as completely authentic.

The Mata sanction came from a US federal court in 2023 (reported by CNBC, 2023); the $110,000 penalty, the largest of its kind to date, came from a federal judge in Oregon (ABA Journal, 2026). The point is not that lawyers are uniquely careless: Deloitte refunded part of a government contract after a report it delivered was found to contain AI-fabricated references (ACS Information Age, 2025). Any firm that forwards AI output as finished work carries the same exposure.

AI output is a draft to verify, never a final answer to send. In professional work, the human review step is not optional polish. It is the job.

What does AI do to the billable hour?

This is the strategic question most firms have not faced yet. If a task that took four hours now takes one, and you bill by the hour, your revenue on that task just fell by three quarters, and the client keeps the saving. Commentators at the Thomson Reuters Institute describe this as an unsustainable long-run setup, and argue generative AI is the catalyst pushing firms from hourly billing toward fixed-fee and value-based pricing (Thomson Reuters Institute, 2024). The regulators reinforce the squeeze: ABA Opinion 512 says a lawyer generally cannot bill a client for the time spent learning to use an AI tool.

You do not have to overhaul your pricing overnight. But before you roll AI across client work, decide how the firm will capture the efficiency, through fixed fees, packaged services, or freed capacity used to take on more clients, rather than handing every gain back through a smaller invoice.

How should a small firm actually start?

Start narrow and internal. Pick one high-volume task where no identifiable client data is required, such as first-draft research or summarizing your own documents, and run it for a few weeks with a human checking every output. Prove it saves real time, write a one-page rule on data and review, then widen use one task at a time. The same disciplined, one-task-at-a-time approach in the companion guide on where to start with AI applies directly here, with the confidentiality rule layered on top.

Common questions about AI for professional services firms

Sources

• Thomson Reuters, 2025 Generative AI in Professional Services Report, April 2025 (survey of nearly 1,800 professionals).
• Thomson Reuters Institute, The Future of Professionals 2025, June 2025 (240 hours / $19,000 per professional estimate).
• Thomson Reuters Institute, Pricing AI-driven legal services and the billable hour, 2024.
• American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools, 2024.
• State Bar of California, Practical Guidance for the Use of Generative AI in the Practice of Law, 2023.
• Journal of Accountancy (AICPA), Generative AI and risks to CPA firms, 2023.
• CNBC, Judge sanctions lawyers over AI-written filing with fake citations (Mata v. Avianca), 2023.
• ABA Journal, Oregon federal judge hands down $110,000 penalty for AI errors, 2026.
• ACS Information Age, Deloitte to refund government over AI errors, 2025.
• Noy and Zhang, Experimental evidence on the productivity effects of generative AI, Science, 2023.
• Brynjolfsson, Li and Raymond, Generative AI at Work, NBER, 2023.