You Owe Clients Credit for AI Savings You Cannot Measure

Two pieces of the accounting profession's AI picture came into focus this month, and they do not fit together comfortably.

The adoption argument is over

Accounting Today reported on August 13 on findings from Thomson Reuters' AI in Professional Services Report. Comparing the 2025 and 2026 surveys, the share of firms saying they had no plans to use generative AI fell from 28% to 19%. The share saying they were already using it climbed from 22% to 40%.

Those two lines moving in opposite directions inside twelve months settles a question a lot of firms were still treating as open. Whether to use these tools is no longer the live debate in the profession.

What makes the reporting useful is that Thomson Reuters then argues the obvious read is wrong. The meaningful divide is not between firms that use AI and firms that do not, because the large majority of professionals now use it at least weekly. The divide the data actually surfaces is what the report calls an AI value gap, and it is reported by more than nine in ten professionals: organizations claim benefits from AI that individual practitioners do not experience in their own work.

A separate study from KPMG, Baker Tilly, and Fieldguide referenced in the same coverage splits practitioners close to evenly, with 51% active deployers and 49% casual users.

So the picture is not early adopters versus holdouts. It is a profession where nearly everyone has the tools open and a large share cannot point to what changed.

The part that turns this into a compliance question

On its own, a value gap is a management problem. You would work on it the way you work on any underperforming investment, on your own schedule.

Except the IRS has already spoken about what happens to AI efficiency in a tax practice.

The Office of Professional Responsibility issued its first formal AI guidance in June, and the Journal of Accountancy's coverage lays out the fee piece directly. Practitioners are expected to "fairly credit to the client's account any cost reductions" produced by AI efficiencies. Billing for time not actually spent, because AI absorbed it, raises an unconscionable fee question under Circular 230.

Nothing in that is a new rule. Circular 230 has always governed fee reasonableness, and OPR was explicit that existing standards apply unchanged. Its summary framing is that technology is a tool and not a substitute for professional judgment.

Put the two findings side by side and the problem states itself. You have an affirmative duty to credit clients for AI-driven cost reductions. More than nine in ten practitioners report they cannot identify the value AI is producing in their own work. A duty to pass along savings you have no method of measuring is not a comfortable position to occupy during an examination of your billing practices.

Why the measurement is genuinely hard

This is not firms being lazy about tracking. The savings are real but diffuse, and they land in places timekeeping was never built to see.

A research question that used to take forty minutes now takes eight, but it happens six times across a week rather than once on a billable matter. A first draft of a client memo arrives in a usable state instead of a rough one, so the partner review shortens without the drafting line changing. A staff accountant stops escalating a class of question entirely. None of that surfaces cleanly in hours by client by matter.

Firms billing hourly feel this first, because the efficiency shows up as a smaller number on the invoice with no corresponding line explaining it. Firms on fixed fees have the reverse exposure. The savings are invisible to the client by design, which is exactly the condition the fee guidance is pointed at.

What we would actually do

The instinct is to launch a firmwide tracking initiative. We would not. Those collapse within a quarter and produce data nobody trusts.

Pick three recurring work types and time them honestly. Not the whole practice. Three things the firm does repeatedly enough that a before-and-after is meaningful: a specific return type, a monthly close, a standard research memo. Measure the actual elapsed time now, against what it took eighteen months ago. Two or three data points beat a survey of impressions.

Decide the fee posture before a client asks. There are defensible answers. Credit the reduction. Hold the fee and expand the scope. Move that work to a fixed fee reflecting the new cost. Any of those can be explained. What cannot be explained is not having considered the question when a client raises it, or when OPR does.

Write down which tools are approved for client data. The same guidance requires client data be handled using only secure, enterprise-approved AI, and puts firm leadership on the hook for training, accuracy monitoring, third-party tool vetting, and documented evidence of compliance. A staff member pasting client figures into a consumer account is the confidentiality problem, and it is a larger one than the billing question.

Name the gap where it is real. If a tool has been in the firm for a year and nobody can describe what it changed, that is a finding. The honest options are to fix the workflow around it or stop paying for it. Continuing to carry it while telling clients the firm is more efficient is the version that creates the exposure.

The shape of it

The profession spent two years arguing about adoption and won that argument. The next two years are going to be about whether anyone can demonstrate what adoption produced, and the standard is not going to be set entirely by firms. Part of it is already written into Circular 230.

Firms that can answer what changed, in hours, on specific work, are in good shape. Firms carrying a general sense that things are faster are going to find that a difficult sentence to say to an examiner.

If your firm is working through where AI actually sits in the practice, and what to tell clients about it, that is a conversation we are glad to have.