Could You Defend Your AI Fee?

When I was at PwC, there was a line you did not want to cross on margin. Charge too much relative to the actual work, and you’d get flagged. Someone would ask questions. At minimum, you’d end up doing a free project or rolling the scope forward, because a fee that looked too good on the firm’s side did not look good for business.

That line just became a federal question.

Back in June, the IRS’s Office of Professional Responsibility put out guidance that applies to anyone using AI in tax prep: if AI is creating savings for you, you’re supposed to pass those savings to the client. The reasoning ties back to Circular 230 Section 10.27(a), the decades-old rule banning an “unconscionable fee” for matters before the IRS. That rule has been around forever, mostly aimed at abusive contingent-fee arrangements. This June was the first time the IRS applied it specifically to AI-driven time savings.

Here’s the scenario that makes it concrete. I remember this exact debate as an associate: we’d build an Alteryx workflow, someone estimated it would take four hours, and it actually took one. What do you charge? You bill the one hour you actually worked, that part’s not in question. But there’s a real question sitting in that three-hour gap. I saved that time. Does it belong to me, or does it belong to the client who was quoted four hours?

The IRS’s answer this year: it belongs to the client. If you’re not adjusting for it, you may be charging an unconscionable fee.

The fee language wasn’t the only thing in the guidance. The IRS also laid out four duties for anyone using AI in tax practice: competence, actually understanding what your tools can and can’t do; confidentiality, client data staying on secure, firm-approved platforms; due diligence, a human reviewing the output before it goes anywhere; and firm procedures, written policies instead of tribal knowledge. None of that is new. It’s the same professional responsibility CPAs have always had, spelled out for AI specifically.

The AICPA pushed back hard on the fee piece earlier this month, and their argument is worth sitting with. Their point isn’t that AI doesn’t save time. It’s that treating the saved hours as pure, costless profit ignores everything underneath it: licensing, training, the project it took to clean up and aggregate your data before AI could even use it, the years of context and judgment that make the tool useful in the first place. None of that shows up if you’re only looking at how long the task took this time.

I’ve lived this from the other side too. One of the first questions I’d get whenever I mentioned a firm had a good AI use case: “So are my fees going down?” Clients ask this half-joking. They’re not really joking.

None of this is settled. The IRS hasn’t walked anything back, and the AICPA is on record saying the guidance isn’t authoritative and is overstepping. But if you’re billing anything AI-assisted right now, waiting for a ruling isn’t a strategy. The move that works no matter who wins: stop pricing by the hour and start pricing by the value of the engagement, the complexity, the risk, the outcome. If the fee was never about hours in the first place, “AI made this faster” stops being something you have to explain.

The other half of that: keep a real accounting of what AI actually costs you. Licensing. Implementation. Training your team. The governance and review layer you built around it. If a fee ever gets questioned, “here’s what this actually costs to run responsibly” is a real answer. “That’s what I’ve always charged” is not.

Key takeaways

• Circular 230 Section 10.27(a) has banned “unconscionable fees” for decades. In June 2026, the IRS applied it to AI-driven time savings for the first time.
• The AICPA is actively disputing the IRS’s position as of this month — this is not settled guidance.
• The defensible move regardless of outcome: price on value, not hours.
• Document your real AI costs — licensing, training, governance — as evidence your fee reflects more than a faster prompt.
• This won’t stay in tax prep. Expect the same question to reach audit and advisory next.

Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: Could You Defend Your AI Fee?