Can behavioural science AI improve professional judgement?

Since its release in late 2022, artificial intelligence has rapidly become embedded in the daily workflows of accountants and finance teams. But while much of the early conversation centred on automation and efficiency, a more consequential advancement is now underway: AI is moving into the realm of professional judgement. Now, as 2026 has already been set in motion, the sector is no longer asking whether AI can process invoices or reconcile accounts more efficiently, but rather whether it can (or should) shape how accountants assess risk, interpret regulation and make complex decisions under uncertainty.

When automation turns into assumption

Recent evidence suggests that, when used carelessly, it can just as easily undermine judgment as enhance it. Research conducted by Dext among 500 UK accounting professionals paints a sobering picture, as half of respondents said they were aware of businesses suffering direct financial losses after acting on incorrect or misleading guidance from general-purpose AI tools such as ChatGPT. Those losses included overpayments, missed allowances, penalties, fines and wider compliance failures.

Nearly a third of those accountants reported encountering client mistakes caused by incorrect AI-generated advice on a weekly basis, while 7% see such errors daily. VAT miscalculations, payroll errors, flawed tax planning and misinterpretation of business expenses are among the most common problems.

The impact was not limited to clients; among accountants who regularly encounter AI-related errors, 93% said they spend up to 10 hours per month correcting mistakes linked to AI-generated advice. Almost four in 10 spend between four and 10 hours a month on remediation work.

“AI has a powerful role to play in finance,” Paul Lodder, vice president of accounting product strategy at Dext, says, “but there is a fundamental difference between specialist tools built for accounting and bookkeeping, and general-purpose chatbots that do not know a business’ true financial context.”

Dext is warning on the growing risk that businesses will get overly comfortable with treating automated outputs as a substitute for professional advice. And the tech firm is not alone: nearly half of accountants expect more businesses to make decisions based on misplaced confidence in inaccurate AI-generated advice in 2026, further compounding financial and compliance risks. A third warned that continued reliance without oversight could even increase the change of insolvencies.

“When someone drops a file into a public AI tool, they’ve effectively shared that document with the world.” – Ben Crow, Iplicit

In short, the problem lies with unstructured AI use being left to steadily replace professional judgement. Another side of this conversation is whether accountants, and other skilled finance professionals, should be the only ones to use such AI tools given their knowledge and ability to spot errors in its answers. 

A separate survey of 250 UK mid-market finance leaders, commissioned by Iplicit, touches upon this worry of widening governance gaps and how to make sure the AI is serving its purpose within firms. While 83% of respondents claim to have adopted AI within their finance function, only 53% have a formal policy or framework in place governing its use. Among partial adopters – the largest group – fewer than half have clear safeguards.

The research also reveals widespread “shadow AI” usage, which describes finance professionals that turn to public AI tools outside official systems. Almost half of organisations that have not formally rolled out AI report that staff are already using tools such as ChatGPT. Even among adopters, patterns suggest workarounds are common.

Ben Crow, vice president of partnerships at Iplicit, explains how this disconnect reflects a deeper issue. “When someone drops a file into a public AI tool, they’ve effectively shared that document with the world. If that file contains sensitive employee data, customer information, or confidential financial details, you’ve just opened up a serious data privacy risk.”

Despite the industry’s eagerness to embrace AI tools in the workplace, only 32% of respondents said they have received formal training. This means that most are self-teaching, which fuels a tension between the rapid experimentation at play and firms’ uneven digital maturity. 

Overconfidence in inaccurate outputs is a risk in this environment, but so is its insufficient critical interrogation. Which brings us to a fundamental question: what happens when AI moves beyond automation and into judgement?

The judgement gap

Wendy Jephson, co-founder and chief executive of behavioural science AI start-up Let’s Think, believes the profession is at a pivotal moment.

“The biggest gap sits between using AI to increase speed or productivity and using it to improve judgement quality,” she says.

Accountancy has long relied on technology for automation, where outputs are relatively easy to verify. But newer forms of AI operate in interpretive territory: assessing risk, drawing inferences, summarising complex regulations and proposing courses of action.

“The goal shouldn’t be to outsource thinking to AI. It’s to use technology to support better thinking – capturing expertise, making it shareable, and helping the next generation learn how good judgement is actually formed.” – Wendy Jephson, Let’s Think

“AI is extremely effective in the hands of experts,” Jephson explains. “Experienced professionals know how to question outputs, spot what’s missing, and recognise when something doesn’t quite fit. But where AI falls short is that it lacks access to tacit, experiential knowledge – the things professionals learn only by doing the work over many years.”

That tacit knowledge, which is often unwritten and informal, as Jephson points out, is under threat. With large numbers of senior professionals approaching retirement and fewer juniors entering the pipeline, firms face a double risk: loss of experiential insight and reduced opportunities for learning-by-doing as routine tasks are automated.

Jephson argues that AI can produce answers that look persuasive but miss crucial context or risk signals. Without prior experience, users may not know how to challenge what they are seeing. In her view, the danger is not that AI replaces accountants, but that it gently hollows out the development of professional scepticism.

If general-purpose AI risks eroding judgment, could a different approach strengthen it?

Jephson’s work draws on behavioural science AI to “codify” expertise – not just what decision was made, but how and why it was reached. Traditional AI tools tend to capture outcomes. Behavioural science AI aims to surface reasoning: the cues noticed, the uncertainties weighed, the alternatives considered and the risks prioritised.

In audit, for example, documentation typically records the final conclusion. Rarely does it capture the mental model behind it: why certain red flags were escalated, why one risk was prioritised over another, or what contextual signals influenced judgement.

By structuring and analysing that reasoning, behavioural science-led systems aim to create a living evidence base of expertise. Jephson says this type of AI can help juniors see not just what decision was taken but how it was formed. Equally, seniors can revisit past decisions with full context, reducing hindsight bias and supporting learning.

“The goal shouldn’t be to outsource thinking to AI,” Jephson says. “It’s to use technology to support better thinking – capturing expertise, making it shareable, and helping the next generation learn how good judgement is actually formed.”

From inputting data to validating the engine

At a European policy level, similar themes are emerging. Eelco van der Enden, chief executive of Brussels-based professional association Accountancy Europe, argues that AI is not eliminating professional judgment but shifting its focus.

“Instead of concentrating on manual data input or spreadsheet manipulation, professionals increasingly assess whether the output makes sense and whether the underlying system is robust,” he explains. “In other words, the judgement moves from data input towards validating the validation engine.”

This reframing is significant, according to van der Enden, and validates Jephson’s idea of what behavioural science AI can do for finance. As AI systems generate insights, accountants must evaluate not only outputs but the reliability, bias and integrity of the algorithms producing them.

The introduction of the EU’s Artificial Intelligence Act, which was introduced in August 2024, reflects this governance shift that places emphasis on risk management and safeguards. Accountability, van der Enden suggests, is being redefined rather than reduced: greater scrutiny of AI governance, stronger expectations around controls, and continued human oversight.

Ethics remains central. “You can automate whether something falls within the letter of the law,” he notes. “It is much harder to automate whether something respects the spirit of the law.”

The implementation of the Corporate Sustainability Reporting Directive (CSRD) is currently being phased in for the period covering 2025 until 2029, and  has further underlined the importance of professional judgment in interpreting evolving standards. Technology can streamline data collection, but interpretation, challenge and contextual application remain human responsibilities, according to van der Enden. The opportunity up for grabs now is less manual bureaucracy and more analytical, advisory work. The risk lies in poor governance and insufficient training.

Fears of deskilling accompany every technological shift. Yet, as van der Enden points out, the profession has repeatedly adapted to spreadsheets, analytics and digitalisation without losing its core expertise. Having seen how the industry developed over the past 35 years, van der Enden says the greater danger may be under-preparedness. Without structured education in how to interrogate AI outputs and assess tool quality, misuse becomes more likely.

Jephson shares that concern. If automation reduces exposure to foundational tasks and firms fail to capture experiential knowledge, future professionals may inherit powerful tools but lack the intuition to use them wisely.

Therefore, Jephson’s and van der Enden’s strategy for leaders suggests adopting AI at a natural, intuitive pace, while ensuring it strengthens – rather than substitutes for – human judgement. 

Jephson’s advice hinges on asking the right questions during this process, such as: are investments improving the quality of risk assessment; are they supporting the development of expertise; and are governance frameworks robust enough to withstand scrutiny?

AI is often described as either a threat or a panacea, but reality is often much less sensational. General-purpose tools used without oversight can amplify errors, create misplaced confidence and introduce new compliance risks, whereas specialist systems embedded within strong governance frameworks can enhance insight and reduce manual burden.

“You can automate whether something falls within the letter of the law. It is much harder to automate whether something respects the spirit of the law.” – Eelco van der Enden, Accountancy Europe

Behavioural science-informed approaches suggest a third path: using AI not merely to automate decisions, but to make reasoning visible and transferable.

For accountants and finance professionals, the differentiator will not be access to technology. It will be the quality of judgment applied to it because, as experts point out, professional scepticism cannot be automated. Ethical interpretation cannot be delegated. Validating the engine – and ensuring it aligns with public interest obligations – remains a human task.

AI may reshape how decisions are formed, documented and scrutinised, but the core responsibility of the profession endures: to exercise informed, independent judgement in the face of uncertainty. The firms that thrive will not be those that adopt AI fastest, but those that use it to think better.