Published in partnership with Sage. Based on Accountancy Age’s Leading Voice Broadcast with Chris Downing and Jack Choppin from Sage.
There is a question that Chris Downing, Director for Accountants and Bookkeepers at Sage, raised during our recent broadcast that most partners in UK accountancy firms will recognise the moment they hear it. When your firm uses AI to complete a piece of work faster, do your clients think you are cheating the system?
It sounds like a question about client perception. At a deeper level, it is a question about whether your firm has the confidence to charge for a service that AI is making faster, and whether the commercial model sitting underneath that service is built to support that charge. For most firms right now, the answer to both parts of that question is no, and the two problems are more connected than they appear.
The confidence gap is broader than most firms realise
Chris identified three distinct layers to the confidence gap during the broadcast, and understanding all three matters because firms that address one without addressing the others tend to find themselves stuck at the same commercial ceiling.
The first layer is capability, which is the question of whether your people know how to use the tools available well enough to trust the outputs they produce. This is the layer most firms are actively working on, through training, experimentation, and the gradual accumulation of experience with specific workflows.
The second layer is trust, which sits alongside capability but is not the same thing. A firm can have capable people using AI tools effectively and still carry a deep institutional nervousness about whether the outputs are reliable enough to act on without significant senior review. That nervousness is not irrational, it reflects the genuine professional standards that accountants and bookkeepers hold themselves to, and Jack Choppin, Senior Accountant Success Manager at Sage, made the point during the broadcast that those standards will ultimately make the profession’s approach to AI more robust than most. In the short term, though, the additional review layer that professional diligence requires is absorbing a significant portion of the efficiency gains that AI is supposed to be generating.
The third layer is the one most firms have not yet addressed directly. It is the question of data security, and it goes beyond the general concern about whether AI tools are safe to use with client information. It encompasses specific, practical questions that every firm using AI should be able to answer clearly. Where is the client data being processed and stored? Does the firm have the right to delete that data, with the correct audit trail to demonstrate deletion? What happens to the data if the tool is discontinued or goes offline? And critically, are the AI services the firm is using learning from the client information being passed to them, building patterns from repeated use of the same client data in ways that may not be visible or controllable?
Chris was direct about this last point during the broadcast. Passing the same type of client information into an LLM repeatedly means the model starts to learn those patterns, and most firms have not yet built the governance framework to manage that risk deliberately rather than inadvertently.
Data security as a governance question, not a barrier to adoption
One of the most useful reframes in the broadcast came from Chris’s observation that data security concerns, while legitimate, should not function as a reason to avoid AI adoption altogether. The PCRT guidance published by the professional institutes in January sets out clearly what firms are required to do, and reading it carefully reveals that many of the practical data concerns are manageable if firms approach them with the same rigour they apply to their compliance obligations.
Most UK accountancy firms are Microsoft houses, and the paid version of Microsoft Copilot operates within a data environment that has already passed the firm’s existing data governance standards. That is a starting point that removes the most significant data security barrier for the majority of firms without requiring a complex procurement or implementation process.
The broader governance requirement, updating letters of engagement to reflect AI use, building a policy for reviewing AI tools on a regular basis as they update and change, establishing a fallback position for when tools go offline, is an extension of the compliance and risk management infrastructure that firms already maintain for AML, KYC, and professional indemnity purposes. Jack made this connection explicitly in the broadcast. The firms that treat AI governance as a new and separate administrative burden tend to find it overwhelming. The firms that treat it as an extension of their existing compliance framework find it manageable, and they find it creates a credible basis for the client disclosure conversations that the PCRT guidance requires.
Why solving confidence without changing the model is not enough
This is the point that Jack made in the broadcast that deserves the most attention from senior leaders in UK accounting firms, because it is the one that explains why so many firms that have invested seriously in AI capability and governance are still not seeing the commercial improvement they expected.
Once a firm has addressed the capability gap and built the governance framework to manage data security appropriately, the commercial model underneath the practice has not changed. The firm is doing the existing model faster, with more confidence, and none of those improvements translate into improved margin if the pricing model was built for a different version of the work.
Jack framed this with a precision that is worth repeating directly. Solving the confidence and capability problem while leaving the commercial model unchanged means the firm is doing what it has always done in a faster way, rather than changing what it does in a way that reflects what AI makes possible. The efficiency improvement is real. The commercial improvement requires a separate decision.
This is why the confidence gap and the commercial model problem are not sequential challenges where one must be solved before the other is addressed. They are parallel problems that need to be worked on simultaneously, because firms that wait until they feel fully confident before beginning the commercial model conversation will find that by the time the confidence arrives, the window for proactive commercial repositioning has narrowed considerably.
What early adopters are doing differently
Chris made an observation during the broadcast about the firms seeing the strongest early results from AI adoption that is instructive for practices at any stage of the journey. The early adopters seeing genuine commercial improvement are not necessarily the firms with the most sophisticated AI implementations. They are the firms that already had scalable systems and workflows in place before they introduced AI tools, which meant that when they added the right LLM or AI capability, the gains were immediate and visible rather than being absorbed into the friction of inconsistent underlying processes. Building those systems and governance infrastructure now, even before the AI capability is fully developed, is the work that determines whether the AI investment produces commercial returns when it is ready to scale.
