Closing the AI skills gap in your accountancy practice

Key takeaways

• A lack of skilled personnel is the second biggest AI implementation challenge in the profession, cited by 48% of accountants.
• Around 38% describe their practice as having limited readiness or no readiness at all, and fewer than a third feel well or highly prepared.
• Roughly one in five accountants name a people concern as their single biggest worry, split between the erosion of entry-level jobs and the undermining of foundational skills.
• Improved staff wellbeing and job satisfaction was reported as a benefit by around 17%, so the upside for your team is real without being automatic.
• Membership organisations offer free e-learning on generative AI fundamentals, and CCAB has published a hub of guidance, checklists, and podcasts.

The most expensive AI mistake a practice can make has nothing to do with choosing the wrong tool. It is buying the right one and rolling it out in a way that makes your team quietly decide to work around it.

The numbers say this is the live risk, as recent research from AccountingWEB in association with Sage reveals.

A lack of skilled personnel is the second most cited implementation challenge in the profession at 48%, behind only data security. Around 38% of practices describe themselves as having limited readiness or not being ready at all.

Skills gaps close with time and training. Readiness gaps close with something harder to buy, which is the sense among your people that this is being done with them rather than to them. This guide covers both, and here’s what we discuss:

The skills gap in numbers

The AccountingWEB research reveals that 48% of accountants name a lack of skilled personnel as an implementation challenge, putting it second only to data security concerns, at 62%.

The readiness picture underneath this is more encouraging than the headline suggests.

Around 8% describe their practice as highly prepared and around 22% as well prepared. Roughly 31%, the largest single group, describe themselves as moderately ready with some skills and further development needed.

Around 26% report limited readiness, and around 12% say they are not ready at all.

Focus on that middle group.

Nearly a third of the profession already has some capability and knows precisely what it is missing. Those practices are not starting from zero, and they tend to close the gap quickly once somebody gives the work a name and an owner.

Introduce AI gradually so change does not unsettle people

Large simultaneous rollouts fail for a predictable reason.

When a tool arrives across the whole practice at once, with training scheduled and targets attached, people draw the obvious conclusion about why it is there.

A sequence that works better:

• Start with a task everyone already finds tedious, so the first experience of AI is relief rather than threat.
• Make early participation voluntary, and let the volunteers report back to their colleagues.
• Publish what you learn internally, including the things that did not work.
• Expand only when there is internal demand, which there usually is once one team saves real time.

Then say out loud what happens to the time you save.

Ambiguity is what makes people anxious, not automation.

If the answer is more advisory work and fewer late nights in January, say so. If you genuinely do not know yet, say that instead, and say when you will decide.

People handle uncertainty far better than they handle silence.

Building CPD when sector training is still catching up

Formal sector training is behind the technology, so most practices are assembling their own. That is more achievable than it sounds, and a good deal of it is free.

External sources worth using now:

• ICAEW members can access up to eight hours of free e-learning covering the fundamentals of generative AI and its application in finance.
• CCAB’s AI hub gathers guidance, checklists, podcasts, and resources on ethical AI use across five professional bodies.
• The PCRT topical guidance on AI, with its accompanying webinar and slides, is a straightforward way to cover the 2026 changes.
• Your software vendors run product training that is usually included in what you already pay them.

Internally, three habits do most of the work.

Run a thirty-minute monthly session where one person demonstrates something they tried, successful or otherwise. Keep a shared prompt library so useful discoveries do not stay on one person’s laptop. Write a short internal guide covering approved tools, prohibited data, and who to ask.

Record all of it as CPD, because that is what it is.

Structured learning relevant to your role counts whether or not somebody sold it to you.

The people doing a task know where the time actually goes, which is often not where a partner assumes it goes.

A tool chosen at partner level and handed down tends to meet one of two fates. It gets used badly, or it gets politely avoided while the old process continues in the background.

Practical mechanisms that avoid this:

• Ask the team, in a short survey, which tasks they would most like help with.
• Put at least one sceptic in every trial group, deliberately.
• Let the team ask the questions in the vendor demo rather than the partners.
• Give the trial group a genuine veto if the tool adds work rather than removing it.

The sceptic point deserves emphasis.

Someone actively looking for problems will find them during your trial rather than after your rollout, which is exactly when you want to find them. Professional scepticism is an asset here in the same way it is in an audit.

Protecting junior development and foundational skills

This concern is real and widely held. In the AccountingWEB research, around 12% of accountants named the erosion of entry-level jobs for new starters as their single biggest AI worry, and around 11% named the undermining of foundational skills.

Taken together, roughly one in five put a people issue above every other concern including data security.

The mechanism behind the worry is sound.

Juniors have traditionally built judgement by doing the work that AI now does quickly. Bookkeeping, reconciliations, and basic preparation are how a trainee learns what normal looks like, which is the only reliable way to recognise abnormal later on.

But there are four protections that work:

1. Have juniors complete a task manually before they are permitted to use a tool on it, so they know what the tool is doing.
2. Teach review as an explicit skill with a checklist, rather than assuming people can evaluate output they were never taught to produce.
3. Keep a deliberate proportion of work manual for training purposes, and be open that this is why.
4. Give juniors the checking, verifying, and questioning work, which is higher-order than data entry ever was.

Reviewing AI output well requires knowing what right looks like.

That is a teachable skill, it is more valuable than the task it replaced, and a trainee who has it will be a considerably better accountant in their third decade of life than one who learned only to key in invoices.

Reframing AI as capacity rather than replacement

The benefit that appears most consistently in the research is reduced time spent on manual tasks, reported by around 42% of respondents—and well ahead of every other outcome the survey asked about.

That is capacity, and capacity is the honest way to describe what AI currently offers a practice.

So have the capacity conversation before you roll anything out. What would your practice do with a day a week? More clients on the same headcount, more advisory work at better margins, faster turnaround, or shorter hours in busy season? There is no wrong answer, and there is a great deal of harm in not having one.

Around 17% of respondents reported improved staff wellbeing and job satisfaction as a benefit of AI. That is a real outcome but a minority one, which tells you it depends on how the change is handled rather than on the tools themselves.

A team that knows what the saved time is for will help you find it. A team that suspects the saved time is for reducing headcount will help you find nothing at all.

Final thoughts

The skills gap is the easier half of this problem.

Training is available, much of it free, and the largest group in the profession already has some capability to build on.

The readiness gap is the half that takes leadership.

It closes when people can see why a tool was chosen, who chose it, what happens to the time it saves, and how their own development is being protected.

None of that requires a budget. All of it requires somebody to decide it matters and say so in public.

Your takeaway: Ask your team which tasks they want help with before you look at any tool. Start with something tedious and voluntary. Put a sceptic in the trial group on purpose. Book a monthly half hour for shared learning and log it as CPD. Answer the capacity question out loud, in front of everyone, before the first rollout.

Do that and the 48% skills barrier stops being a reason to wait. It becomes the specific piece of work you are getting on with.

Read below—State of the nation: AI in accountancy and bookkeeping, produced by AccountingWEB in association with Sage

Frequently asked questions

Why do accountants say they lack the skills to adopt AI?

A lack of skilled personnel is cited by 48% of accountants as an AI implementation challenge, second only to data security. The gap is partly technical and largely structural, because formal sector training has not kept pace with how quickly the tools have changed. Around 38% of practices describe themselves as having limited readiness or no readiness at all, while roughly 31% report some skills with further development needed, which makes the middle group the largest in the profession.

How can a small practice train its team in AI without a training budget?

Most of what a small practice needs is free. ICAEW members can access up to eight hours of free e-learning on generative AI fundamentals, CCAB has published a hub of guidance, checklists, and podcasts on ethical AI use, and the PCRT AI guidance comes with a webinar and slides. Internally, a monthly thirty-minute session where one person demonstrates something they tried, plus a shared prompt library, covers most of the rest. All of it counts as CPD.

Will AI replace junior accountants?

The research shows this is a genuine concern, with around 12% of accountants naming the erosion of entry-level jobs as their single biggest worry about AI. The more likely outcome is that junior roles change rather than disappear, moving from producing routine work towards checking, verifying, and questioning AI output. That is higher-order work, and it requires trainees to understand the underlying task, which is why many practices now have juniors complete work manually before allowing them to use a tool on it.

How should a practice introduce AI without unsettling staff?

Start with a task everyone finds tedious, make early participation voluntary, and let the volunteers report back to their colleagues. Involve the team in tool selection, including at least one deliberate sceptic in any trial group. Most importantly, state clearly what will happen to the time the tool saves. Ambiguity about headcount causes far more resistance than the technology itself, and silence is generally read as bad news.

Does AI training count towards CPD?

Yes. Structured learning relevant to your professional role counts towards CPD whether or not it was purchased from a training provider. That includes free e-learning from your professional body, vendor product training, reading professional body guidance such as the PCRT topical guidance on AI, and internal sessions where colleagues share what they have learned. Record what you did, how long it took, and what you took from it, in the same way as any other CPD activity.

Browse more topics from this article