AI tools are becoming a prerequisite for accountants, just as Excel fluency was for the previous generation, as the opportunities to automate routine activities, analyse vast datasets, and support a radical reinvention of the finance business partnering concept strike a chord.

But a mismatch between perceived opportunity and operational readiness is cause for concern. Research from Chartered Accountants Worldwide found that 85% of those surveyed were willing to use AI technology, but more than half felt that insufficient skills and training was the biggest barrier to adoption.

Inadequate AI skills capabilities are not only preventing the sector from unlocking growth opportunities, but without AI skills teams face increased risks from manual errors, poor data interpretation, and a reduced ability to manage AI-related compliance or ethical concerns.

Getting started

Given the high stakes for finance teams, the question is, how should they approach increasing their AI skills capabilities? Joshua Killalea is the Head of AI at IT consulting firm Panoram Digital. “The first step in effective AI upskilling is analysing internal workflows to identify where time is being lost, where repetitive effort sits, and where AI can be applied in a way that is both useful and controllable,” he says.

The biggest opportunities to generate return on investment are usually where skilled people are spending too much time on repetitive review, summarisation, and commentary. Killalea says: “AI can compress large amounts of low-value effort without removing accountable decision-making.”

One of the biggest knowledge gaps is understanding what’s possible with these tools, says Peter Beard ACA BFP, Director of GenFinance.AI, whose conversations with over 850 CFOs last year have given him a great vantage point on the issues they face.

Although finance professionals struggle to justify time spent on AI experimentation because it doesn’t fit on a timesheet, safe experimentation is key. The ICAEW GenAI Accelerator programme is a practical starting point for structured learning.

Consider allocating resources to a champion within your firm or business to play with the tools and devise use cases, Beard says. “Investing in a business or enterprise licence for one of the big six tools - ChatGPT, Microsoft Copilot, Gemini, Perplexity, Claude and Grok – will cost around £20-£30 a person per month. Before you roll it out to the finance team, use dummy data to experiment and get used to the look and feel of the tool, before you let them have at it with your real financial data.”

Professional scepticism and other vital skills

The propensity for AI systems to hallucinate – producing incomplete or incorrect answers based on biased or flawed data – highlights the importance of critical thinking. That matters even more in finance where an answer that sounds plausible but is poorly grounded can create risk very quickly.

As use of AI tools shifts the onus from preparation to review, professional scepticism must take centre stage. Teams must have the ability to interrogate outputs, question the legitimacy of answers and understand the assumptions behind them.

“In practical terms that means going through the formulas the AI produces in an Excel and validating them. Coming up with that checklist is a new skill. The way you do it is to have those conversations with your existing preparer and transfer that knowledge to the AI,” Beard says.

“Mark formula-driven cells in green text, hardcoded values in blue, and add a final validation tab with pass/fail checks, for example do subtotals equal the total? Or has the row count changed from the source data?” Beard suggests.

Data literacy is fundamental; understanding where the data comes from, how reliable it is and how it should be interpreted. Clear governance structures are key. Everyone should be very clear on how models are validated, decisions are documented and how outputs can be explained to auditors, regulators or customers. If AI contributes to commentary, analysis, or anything that feeds a decision, there should be clarity on what was reviewed, what was changed, and where a human needed to intervene.

Caroline Carruthers, Chief Executive of data consultancy Carruthers and Jackson, admits that it takes curiosity, imagination, and bravery to interrogate the answers you’re seeing. She says: “Deploying it in pioneering ways takes creativity and a willingness to accept occasional failure – or missteps – as part of being at the cutting edge.”

Small steps lead to skilled team

A strong AI-skilled finance team is not one where every individual can build machine learning models, but one where the team collectively understands how AI supports decision making, operations and knows how to challenge it when necessary. That requires an environment and culture that encourages people to explore AI and be inquisitive about it, without fear that it will replace them.

To upskill your team, the advice is start small and stay close to live work. “Show them what good use looks like, where the system is likely to struggle, what checks are required, and which decisions remain human-owned. That tends to create far better adoption than broad training programmes that explain what AI is but never connect it to the actual work people are doing,” Killalea says.

Replicate early use cases using approved tools, defined review steps, example prompts and practical guidance on what to do when the output is weak or uncertain. “The point is not to get everyone experimenting endlessly,” Killalea says. “It is to help teams use AI in ways that are repeatable, measurable, and safe enough to become part of normal delivery.”