Thomson Reuters found organisation-wide AI use in professional services jumped from 22% to 40% in a year, while just 18% of professionals could say their firm tracks the return. Adoption is no longer the constraint.
Thomson Reuters published its 2026 AI in Professional Services Report last month, drawing on more than 1,500 respondents across 27 countries. Organisation-wide AI use almost doubled in a year, from 22% to 40%. Fifteen per cent have deployed agentic tools and another 53% are planning or considering it. On the adoption question, the argument is over.
One figure in the same report is doing more work than the rest. Only 18% of respondents said they knew their organisation was tracking the return on its AI tools. Among the firms that do measure, most are counting internal operational metrics — usage, seats, documents processed — rather than anything that connects to client outcomes or revenue.
That is a firm buying a capability and never establishing what it bought.
The Irish position is a size problem, not an appetite problem
CSO data published in February 2026 puts AI use among Irish enterprises with ten or more staff at 20.2% for 2025, and the split by size is stark: 17.2% of small firms, 28.6% of medium, 57.7% of large. Accounting and financial management is one of the two most common use cases, which is exactly where a five-partner accountancy practice feels the pressure.
The large-firm number is not evidence of better judgement. It is evidence of somebody whose job it is to run the pilot, chase the licences and write the policy. A twelve-person solicitors' practice in Galway has the same obligations and the same opportunity, with the work landing on a partner who already has a full matter list. That is the real gap between 17.2% and 57.7%, and it is why small firms who do adopt tend to do it well — the partner running it is the partner using it.
Time saved is not money earned
Here is where most professional services firms lose the return. AI compresses the input side of the work: first drafts, document review, extraction from filings, chronologies, research memos, the long tail of administrative correspondence. On a fixed-fee engagement that improves margin immediately. On time-recorded work it removes billable hours, and unless the practice has decided in advance what fills the recovered hour, the honest answer at year end is that AI reduced fee income.
We have seen a firm run a genuinely successful pilot — real time saved, verified, on a defined process — and report no financial benefit whatsoever, because the recovered capacity went into work that was already being written off. The tool performed. The economics were never redesigned around it.
The recovered hour has to have a destination decided before deployment: more matters through the same team, work brought back in-house from counsel or subcontractors, a repriced fixed-fee product, or a genuine reduction in write-offs. Those are different commercial decisions with different consequences, and none of them happen because a licence was purchased.
Review time is a real cost and it belongs in the calculation
The professional signs off. Chartered Accountants Ireland, ACCA and the Law Society have all been consistent on that — competence, due care and confidentiality apply regardless of which tool produced the draft. So every hour AI saves in production adds some amount of review time back, and in regulated work the reviewer is usually the most expensive person in the building.
Firms that measure this properly often find the net gain is smaller than the pilot suggested, and concentrated in a narrower set of tasks. That is a more useful result than a headline percentage, because it tells you where to expand and where to stop. It is also the number that stops a pilot from quietly failing on its way to production.
Since 2 August the AI Act's transparency and deployer obligations are live, which adds a second reason to know what your tools are doing on client work. A firm that cannot say which systems touch which matters cannot meet Article 50 disclosure either, and that inventory problem is the same one that surfaces shadow AI.
What to count instead
Four measures, per matter type rather than firm-wide. Cycle time from instruction to delivery. Review hours added at senior grade. Write-offs and recoverability on the matters where the tool is used. And where the recovered hours actually went — because if nobody can answer that, they went nowhere.
The Acuity AI position
We diagnose before we prescribe, and professional services is where that discipline pays fastest. The firms getting a return are not the ones with the best tool. They are the ones who decided what the freed capacity was for before they bought anything, and who measure the same handful of numbers every quarter.
If your practice has AI in use and no agreed measure of what it has returned, that is a two-week piece of work, not a transformation programme. Our AI consulting engagements start with what the firm is actually doing and what it is actually getting, and our AI policy work for professional services firms covers the obligations that come with it.
Forty per cent adoption and eighteen per cent measurement is not an AI problem. It is a management one, and it is fixable this quarter.