From Experiments to Embedded Practice: AI Adoption in Accounting Firms

As AI continues to dominate conferences and marketing decks, the reality on the ground in most accounting firms looks quite different.

At one end of the spectrum, the Big 4 are investing heavily in AI. At the other end, a lot of small and mid‑tier firms are still experimenting at the edges (often with tools like ChatGPT), without yet changing the core of how their practices run.

The latest 2025 GenAI in Professional Services report from Thomson Reuters puts some numbers around this. Only 21% of tax firms say they are actively using generative AI today, although more than half report that they are planning to use it or are considering it. Among the firms that are already using GenAI, just over half (52%) are relying on general open tools like ChatGPT, while only 17% are using industry‑specific AI tools that have been built with tax and accounting processes in mind.

Source: Thomson Reuters

The Big 4 Are Redesigning Processes

When you look at what the large global firms are doing, it is clear they have moved well beyond casual experimentation. Deloitte, EY, PwC and KPMG are embedding GenAI agents into their audit platforms so that documentation can be reviewed, patterns can be identified and suggested improvements can be generated before a human looks at the file. They are building unified AI platforms that sit across tax, audit, advisory and internal IT, rather than treating AI as a standalone tool. Some are reporting productivity gains in the range of 20–50% in their internal software development teams as they use GenAI to assist with coding, testing and troubleshooting.

Source: Thomson Reuters

This is not simply about "playing with prompts". It is about quietly redesigning how work flows through their organisations.

The Reality for Small and Mid-Tier Firms

By contrast, the picture in many small and mid‑tier firms looks different, and understandably so. Often there are one or two partners who are personally interested in AI and are trying things out. A handful of team members might use ChatGPT or Copilot to draft emails, summarise documents or do a first pass at research. These experiments can be genuinely helpful, but they usually sit on top of existing processes rather than changing them. The core work of the firm (tax preparation, compliance, workpapers, audits, management reports, advisory conversations) often still follows the same patterns it did a few years ago.

None of this means those firms are "behind". Most are navigating real constraints: client demands, capacity pressure, legacy systems, risk and quality concerns, and sheer lack of time to step back and rethink processes. But it does mean that there is a growing gap between "we use AI here and there" and "AI is part of how our firm operates".

A Different Way to Think About AI Maturity

When we sit down with firms, it can be useful to discuss AI in terms of a simple maturity ladder and determine where the firm sits on that ladder today.

At the most basic level, there is AI as a toy. In this stage, people are experimenting with prompts and tools in an ad hoc way. There may not be any clear policy, guidance or governance. The main benefit is curiosity and a few isolated time savers.

A step up from that is AI as an assistant. Here, AI is used more consistently to help with specific tasks such as research, drafting content or doing a first‑pass analysis. Some informal guidance or training might exist, but AI is still layered on top of existing processes rather than built into them. If the AI tools vanished tomorrow, the firm could carry on broadly as before.

The next level is AI as part of the process. In this stage, AI is deliberately embedded into particular processes: tax preparation, workpaper review, bookkeeping routines, year‑end packages, or proposal creation. Triggers, templates and review steps are designed with the expectation that an AI tool will play a role and that a human will review and sign off on its output. The process itself has been reshaped around that collaboration.

At the most integrated end of the scale is AI as infrastructure. Here, the AI layer connects with practice management, document management, tax software and other core systems. Rather than being something individual staff "go and use", AI sits behind the scenes, helping route work, flag potential risks, surface relevant insights and suggest next actions. In this world, AI is part of the firm's operating system, not just another application.

When we ask firms where they would like to be, most point towards the later stages of this ladder. When they reflect on where they are right now, many conclude that they are somewhere in the early to middle stages. That gap between aspiration and reality is where the interesting work begins.

Questions Worth Some Thought

For leadership teams who are trying to work out a practical way forward, a few simple questions can open up useful discussion.

One question is: if every AI tool currently in use at your firm suddenly disappeared, how much would your day‑to‑day operations really change? Would you lose some convenience and a few efficiencies, or would key processes genuinely be disrupted?

Another question is: where, specifically, does AI currently touch a client‑facing deliverable? That might be a tax return, an advice paper, a board pack, an audit file or a management report. If AI use is limited to internal admin, email drafts and marketing content, there may be more scope to explore how it can support the core work clients actually pay for.

A third question is about ownership. Who is responsible for AI in your firm? Is it clearly part of someone's role, with the time and mandate to experiment and implement, or is it something that sits vaguely on everyone's "we should look at that" list?

If the answers to these questions feel a little uncomfortable, that is not a sign of failure. It is simply a sign that the topic is still emerging and that there is space to make deliberate choices before AI‑native firms become the default benchmark.

Moving From Experiments to Embedded Practice

At AI Dojo, we spend most of our time with Australian accounting firms working through exactly this transition: from one‑off experiments and individual enthusiasm, to AI that is quietly embedded in everyday processes.

That usually does not start with a grand transformation program. It starts with a realistic conversation about where the firm sits on that maturity ladder, what constraints it faces, and which specific processes are best suited for a first round of redesign with AI in mind.

If you are a partner, CFO or practice manager thinking about these questions, the important thing is not to chase every new tool, but to be clear about the role you want AI to play in your firm over the next few years – and to take the first practical steps towards that, one process at a time.