3 things for CPAs to look out for when using AI

Going into 2026, almost every CPA has used AI in their job, but not every CPA feels like they are getting as much value as they can from this new technology.

According to Stanford, AI is helping the accounting industry by doing the "boring stuff." That's a great start, but many CPAs are now interested in having AI do even more – such as helping with analysis, marketing and reporting. The problem is that AI comes with some risks – it can hallucinate and come to false conclusions when asked to extrapolate or reason through large amounts of information. 

This presents both a caution and an opportunity for CPAs. This coming year, CPAs can do two things: embrace AI innovations that go beyond the boring work while ensuring that the AI they use doesn't pose a risk.

As they evaluate AI features and solutions, CPAs can ask themselves three questions:

Are there more ways for AI to create efficiency?

When a CFO or controller incorporates any new technology into a finance and accounting process, their primary objective is almost always saving time or extra work.  Saved time results in many benefits, such as:

• A faster monthly close, giving management the financial data they need timely to make better decisions;
• The finance and accounting team can spend more time on higher-value work; and
• The organization can scale more efficiently.

When exploring incorporating AI into processes, CPAs often start with tasks that AI is good at such as data entry.  According to a study of AI use by accounting firms, AI initiatives resulted in a "reallocation of approximately 8.5% of accountant time from routine data entry toward high-value tasks such as a business communication and quality assurance."  In fact, those two areas would be a great place to start pilot projects to create even more efficiency during the day. For example, having AI draft standardized emails to clients and prospects or review written communications and make suggestions to make it flow better, or act as a QA "copilot" checking for errors.

On the contrary, there are other areas that seem promising uses for AI but have produced mixed results when it comes to creating efficiency.  In the case of categorizing bank transactions for an accounting firm's clients, an area with a lot of focus by AI initiatives, defined bank matching rules often win out over AI suggestions.  While AI may be able to give firms an edge in categorizing transactions, that is more than offset by the review and potential rework from variability and uncertainty across a large client base. 

Can I explain this to an auditor?

Any accountant who's been through an audit knows that documentation is critical to achieving a good result.  Throughout my career as an auditor and accountant, various versions of the phrase "if it isn't documented, it isn't done" are imprinted in my mind.  That was stressed ad nauseum from day one, and for good reason. There's often no way to validate that a control was performed, a certain number is accurate, or an estimate was logically calculated without proper documentation.

This poses a major challenge for LLMs that many AI solutions are built on.  Magically arriving at an answer that is most likely right isn't an acceptable solution for most accounting tasks. Hallucinations, errors and estimates don't translate in accounting but are prevalent when AI doesn't have guardrails.

The Secret CFO put it well in a post on X: Even if a model is 99.99% accurate, it can't be a black box. In finance and accounting, the "why" matters just as much as the "what." If we can't validate an AI-generated number or conclusion, it's a non-starter. While AI can be great at data entry, standard calculations and error checking, it's not as reliable at interpreting the complex tax code or assessing the best ways to set up an accounting structure for a business client. With black-box reasoning, anything AI does that's complicated wouldn't be explainable and could create liability.

AI still can have a place in critical finance and accounting processes, but small tests work best. For example, using AI to run through potential scenarios in order to determine what the pros and cons are of different business decisions, vs. having AI do the decision-making on its own is lower risk and helpful (assuming AI can provide an explanation for its suggestions!)

Are junior employees able to support the output?

A CPA that's solved for the first two points is probably already using AI to automate a ton of lower level work, that's a huge success — but a new problem emerges.  If CPAs are the lynchpin for any AI initiative, they actually end up with more difficult work. 

CPAs that have employed AI for every easy task find that all they have left is the hard stuff — critical reviews of AI work on top of strategic projects and client interaction. To solve for this, CPAs can delegate. For instance, if a pilot project shows that an AI tool will likely help improve accuracy in a firm's quality assurance process, the CPA has to review the output and has inadvertently added an hour of deep focused work on top of their other strategic priorities.   

Instead, the CPA should identify a member of the team who can leverage that tool to take on that additional work. Even if it takes them a little longer, this is a much more scalable approach. 

Set high standards for AI and reap the rewards

Starting with time savings is a great way to get started with AI, but in 2026, CPAs can get deeper value from the technology. Not every AI solution is created equal; CPAs need to ensure the black box doesn't take over important tasks and must look for AI solutions that have guardrails to eliminate hallucinations and errors, and support auditability. At the same time, AI has the power to up-level everyone's work and give back time in the day to think about hard problems and work on new opportunities. Starting small, delegating and focusing on low-risk projects are all great ways to get more out of AI in 2026. 

CPAs are known for having a critical eye and sweating the details. In the case of AI, these traits pay dividends.