AI in Professional Services Firms: What’s Working in Law, Accounting, and Consulting

Law firms, accounting firms, and consultancies sell expertise, mostly by the hour, and a large share of those hours goes to reading, researching, drafting, and analyzing. That’s exactly the work generative AI is good at. It makes professional services firms both the most exposed to AI and among the best placed to gain from it.

This guide looks at what the early adopters did, what the research says about where AI helps and where it hurts, the failures worth learning from, and a practical roadmap for firms of any size.

How fast firms are adopting AI

Adoption in the professions has moved quickly. Clio’s 2024 Legal Trends Report found that the share of legal professionals using AI jumped from 19% in 2023 to 79% in 2024. Its analysis also estimated that up to 74% of hourly billable tasks, such as gathering information and analyzing data, could be automated with AI.

Thomson Reuters’ Future of Professionals 2025 report, based on 2,275 professionals in legal, tax, accounting, risk, and compliance, found that respondents expect AI to save them about 5 hours a week over the next year, roughly 240 hours a year, worth an average of $19,000 per professional. The same research found a sharp divide: organizations with a visible AI strategy were twice as likely to see AI-driven revenue growth as those adopting AI ad hoc, and 3.5 times as likely to realize critical benefits as those with no significant plans. Only about a quarter had such a strategy.

What the early movers did

The largest firms moved first, and publicly:

• Allen & Overy (now A&O Shearman) trialed the legal AI platform Harvey from November 2022. By the end of the trial, about 3,500 of its lawyers had asked it around 40,000 questions, and in February 2023 the firm announced a firmwide rollout across 43 offices.
• PwC US committed $1 billion over three years to AI in April 2023, and its tax and legal business struck its own deal with Harvey. KPMG and EY announced large AI programs of their own later that year.
• Garfield.Law became the first AI-driven law firm authorized by the Solicitors Regulation Authority, of England and Wales in May 2025. It helps businesses recover unpaid debts through the small claims process, starting at £2 for a reminder letter. Before approving it, the SRA reviewed its safeguards on confidentiality, conflicts, quality, and hallucinations, and the system acts only with client approval at each step.

The pattern is instructive. The winners didn’t hand AI a job and walk away. They picked specific, high-volume work, kept professionals accountable for every output, and built the governance before scaling.

The jagged frontier: where AI helps and where it hurts

The most useful study for firms comes from Harvard Business School, working with Boston Consulting Group. In a preregistered experiment, 758 BCG consultants did realistic consulting tasks with or without GPT-4.

On 18 tasks within the AI’s capabilities, consultants using it completed 12.2% more tasks, 25.1% faster, at more than 40% higher quality. On one task deliberately chosen to fall outside its capabilities, consultants with AI were 19 percentage points less likely to reach the correct answer than those without it.

The researchers called this the “jagged technological frontier”: AI is excellent at some tasks and quietly wrong at others that look just as hard. Its mistakes look as polished as its good work, which is what makes them dangerous. For firms, the lesson is to map the work task by task. Drafting, summarizing, and first-pass review often sit well within the frontier. Novel judgment calls, unusual fact patterns, and anything where a plausible-sounding answer can be wrong need a professional in charge, with AI in a supporting role at most.

The failure mode: confident fabrication

The best-known warning is Mata v. Avianca. In June 2023, a federal judge in New York sanctioned two lawyers and their firm $5,000 for filing a brief that cited six court decisions invented by ChatGPT, complete with fake quotes and citations. The judge found they had acted in bad faith by standing by the fake cases after they were questioned.

It was far from the last case. Researcher Damien Charlotin’s AI Hallucination Cases database tracks court decisions that address AI-generated content in filings, such as invented citations. As of September 2026 it lists 2,086 of them, most in the United States.

Consulting isn’t immune either. In 2025, Deloitte Australia partially refunded the Australian government for an A$440,000 review after academics found fabricated references and a made-up quote from a court judgment. The corrected version disclosed that generative AI had been used in writing it, and Deloitte refunded about A$97,000.

In each case, the model behaved as models do. What failed was the workflow: nobody checked the output against its sources before it went out under a professional’s name.

Ethics and client obligations

Regulators have mostly applied existing duties rather than inventing new ones. The American Bar Association’s Formal Opinion 512 (July 2024), its first ethics guidance on generative AI, covers:

• Competence. Lawyers need a reasonable understanding of a tool’s capabilities and limits, and must review its output.
• Confidentiality. Before putting client information into a self-learning tool that could expose it, a lawyer generally needs the client’s informed consent, and boilerplate in an engagement letter isn’t enough.
• Communication, candor, and supervision. Lawyers stay responsible to clients and courts for what they file, and must supervise how their teams use AI.
• Fees. Charges must be reasonable. A lawyer billing by the hour can bill only the time actually spent, and generally can’t bill clients for learning a general-purpose tool.

Accountants and consultants have their own versions of these duties, around confidentiality, independence, and professional care. The practical requirements are the same: know your tools, protect client data, verify outputs, and document what you did.

The billable hour problem

If AI turns a ten-hour task into a two-hour one, a firm billing by the hour earns less for the same result. Clio’s report pointed at the obvious response, shifting more work to flat fees and other alternative pricing. Firms that capture AI’s gains will price the outcome, not the hours, and use the time saved to take on more work or to do deeper work for the same clients.

A practical roadmap for firms

1. Set a strategy, not a pilot. Thomson Reuters’ data shows the gap between firms with an AI strategy and those experimenting ad hoc. Decide which practice areas and workflows matter most, who owns the program, and how success will be measured.
2. Map tasks against the frontier. For your top workflows, list the tasks and mark which ones AI can do well with review, which it can support, and which stay entirely human.
3. Buy for common work, build for your edge. Vertical legal and tax AI products handle much generic research and drafting. Build custom where your advantage lives: your precedents, your templates, your client data, and the systems they sit in.
4. Ground everything in sources. Answers should come from your documents and approved databases, with citations shown next to every claim so checking takes seconds.
5. Engineer the controls. Matter-level permissions and ethical walls, no training on client data, single sign-on, retention rules, and an audit log of what the AI produced and who approved it.
6. Make verification a step, not a hope. Build review into the workflow, so nothing reaches a client or a court without a named professional signing off.
7. Measure against a baseline. Record time per task, error rates, and turnaround before rollout, then compare. It’s also how you’ll set fair fixed fees.

How Squarify Studio helps firms

We work with firms on both sides of the roadmap. AI consulting covers the decisions: which workflows to target, which vendors to shortlist, and what risks to control. AI integration and internal tools cover the build: source-grounded search over your precedents, document intake and review workflows, and client portals, all shipped in two-week sprints with a working version every week. See how we work, or start a conversation.