The Karbon State of AI in Accounting 2026 report lands at a critical inflection point. Yes, 98% of accounting firms now use AI daily or multiple times daily. But here is what matters to UK regulated firms: the productivity and talent wins are not distributed evenly. They cluster around firms that have written AI strategy, mandatory training programmes, and documented governance. This is not a nice-to-have. Under the FRC's ISA UK requirements and the ICO's framework for AI and data protection, documented AI governance is becoming a regulatory expectation, not optional innovation theatre. Mid-market practices that treat AI as a productivity bolt-on rather than a governed capability will find themselves at both a compliance and a competitive disadvantage within 18 months.

This story is part of a larger pattern. The industry is moving past the hype cycle — past 'we bought ChatGPT for the team' — and into the messy reality that AI only works inside a disciplined operating model. The same pattern played out in legal services with discovery and contract review tools (Harvey, Luminance, Legora), and in insurance with claims assessment. Tools alone do not unlock value. Strategy, training, and accountability do. The firms that are pulling away are not those with the fanciest AI; they are those that have spent money on change management, role design, and clear policies about what tasks AI can handle without human review. The Karbon finding about graduate recruitment — 91% of professionals say AI usage makes firms more attractive to early career talent — is the canary in the coal mine. If your competitors are running AI-enabled onboarding and using AI to reduce grunt work, you will lose the people you need.

Trovix's position on this is unambiguous: AI implementation without governance is liability creation dressed up as transformation. We have seen firms buy powerful generative AI tools and then spend six months in quiet panic when the FCA or ICO asks how they validate the outputs. The honest truth is that most commercial off-the-shelf AI products (including Microsoft Copilot in financial services contexts) assume the firm has already solved governance, training, and change management. They have not. That is why we built Trovix Watch to surface regulatory change and emerging AI governance obligations in real time, and why we embed governance into our other products rather than bolt it on afterwards. The Karbon data shows that firms with 'policies and strategy' win. But creating those policies in a vacuum, without real-time visibility into regulatory change and without systematic audit trails, is insufficient. You need both the policy framework and the tooling that enforces it.

For a mid-market accountancy practice, law firm, or financial services outfit, the action is now: First, commission a 90-day AI audit. Map what you are already using (ChatGPT, Copilot, Claude, whatever), understand where it touches regulated activity or client data, and document the gaps in your governance. Second, write a simple AI policy — not a 40-page manifesto, just a clear statement of what AI can and cannot do in your firm, who owns sign-off, and how you audit it. Third, make AI training mandatory and measure completion. Karbon's data shows these three things separate the winners from the laggards. Fourth, if you are using AI for knowledge work (fee-earner support, document analysis, intake), you need systematic oversight. That is where tools like Trovix Aria and Trovix Sift become essential — not because they are slick, but because they create audit trails and embed governance into the workflow itself rather than leaving it to hope.

Source: Karbon