The Hidden Tradeoff of AI in Accounting: Efficiency vs. Instinct

Goldman Sachs partner Chris Churchman recently warned about “cognitive atrophy,“ the risk that professionals who rely too heavily on AI for reasoning stop building the instinct to recognize a wrong answer. He was talking about bankers and traders,  but the concern applies just as directly to accounting.

Most accounting teams have moved past AI pilots. It’s often embedded in daily close, billing, and revenue workflows, which means these tradeoffs stop being theoretical the moment a junior accountant starts leaning on a tool instead of building a habit.

The conversation around AI in accounting almost always starts with automation, which is a reasonable place to start. It’s also an incomplete one. The tasks we’re most eager to eliminate, reconciling an account that doesn’t tie, chasing a variance until you find the contract change that explains it, are often the ones that teach accountants how numbers actually behave. If AI gets good enough early enough, those accountants get faster without ever building the judgment required to review the output, challenge the result, or explain what happens when something doesn’t hold up.

The Blind Spot in the Efficiency Case

The case for automating accounting’s manual work is usually built on the talent pipeline. Fewer people are choosing this profession, and reducing the manual work is one way to make entry-level roles more attractive. I’ve made that case myself, and it’s still true. But it has a blind spot. Much of the entry-level work we’re eager to automate is exactly where accountants develop the instinct to spot anomalies and apply context. Optimize for throughput alone, and you can solve a recruiting problem while quietly weakening the leadership bench behind it.

Ernst & Young recently committed $100 million in bonuses for employees who demonstrate human skills like judgment. I am torn on this approach. While it rewards essential traits, bonuses pay for existing judgment rather than building it. Understanding is earned through the experience of doing the work, not a playbook. If we automate the learning out of entry-level roles, we risk losing the people capable of recognizing when a perfectly efficient output is fundamentally wrong. Rewarding judgment is a start, but it only matters if we ensure early-career accountants have the space to develop it.

What We’re Doing About It

AI has cut reporting time on billing exceptions by roughly 70%, and it’s shortened how long our revenue accounting team spends tracing a variance back to the contract change that caused it, but we still do reviews in a team setting instead of moving everything to async sign-off. That matters because of the kind of questions that come up. Most of what gets asked in review has less to do with whether the numbers tie out and more to do with whether the result makes sense. Those questions are easier to ask from the outside, since it’s easy for the person who prepared the work to just take the result and run with it. That fresh-eyes habit is how our team continues to sharpen our judgment.

A Few Things Worth Doing Now

• Separate transactional tasks from training work: Not every manual task exists for the same reason. Some produce an output. Others build the person doing the work, and it’s worth knowing which is which before you hand it to a model.
• Keep junior accountants close to the exceptions: Judgment gets built on oddities and edge cases, not on the routine work AI clears fastest. That means resisting the urge to route every exception to the fastest resolution path available.
• Redesign career paths on purpose: If AI absorbs the transactional repetition that used to fill a staff accountant’s first two years, the development plan for that role needs to be rebuilt, not just shortened. A role built around review and judgment from day one looks different than one built around processing volume.

The Real Risk Is the Talent Pipeline

Churchman compared cognitive atrophy due to AI to what GPS did to navigation, a skill that quietly eroded because the tool got good enough that practicing it stopped feeling necessary. Accounting has more at stake than a sense of direction. We’re responsible for numbers that regulators, auditors, and boards rely on to be defensible, not just fast.

Get it right, and AI doesn’t just make accounting a strategic, modern function, it could even help us attract the next generation of talent, all without trading away the judgment that makes the numbers worth trusting. So here’s the question worth sitting with: what are you doing to make sure the accountants joining your team now get the reps that taught the rest of us how to reason through a number that doesn’t add up?

Keep Exploring

October 20: Women Leading the Next Era of Finance 

As AI changes how finance work gets done, leaders are also rethinking how to develop the judgment, curiosity, and cross-functional perspective their teams need. On October 20, Zuora will bring together finance leaders from Zuora, PwC, Anthropic, and Amazon Web Services at The Commonwealth Club in San Francisco for a conversation about thoughtful AI adoption, sound judgment, and resilient team building.

User Guide: Getting Started With Zuora AI for Revenue Accounting Operations

AI can help revenue accounting teams investigate exceptions faster and reduce manual reconciliation work, creating more room for analysis and the judgment that keeps financial outputs trustworthy. This practical guide shows how to move from scattered quote-to-cash context to a clearer next step inside the platform.