Can AI improve my CPA firm's realization rate?

Yes. AI improves a CPA firm's realization rate by getting jobs finished inside the hours they were budgeted for, so the chargeable time that used to be written off when a job ran over gets billed and recovered. It only moves the number once you know which jobs run over, where and why, and finding that out means going through the whole firm before any AI goes in.

Where does a CPA firm's realization rate leak?

The way it works is that you budget the hours for each job and set the fee from that budget. When the job runs over, because the client sent documents late or the review notes went back and forth, the extra chargeable time sits in WIP until a partner writes it down or writes it off. Your realization rate is what's left. The National MAP Survey measures it as net client fees divided by gross client fees (Journal of Accountancy, Oct. 2023), and UK firms call it the recovery rate, which Xero describes as invoiced revenue divided by potential revenue at standard rates (Xero).

The number on your report can look healthier than the firm really is. In the 2023 survey, realization for all firms reached 99% on fiscal 2022 results while firmwide utilization dipped to 59.6% from 62.3%, and the Journal of Accountancy suggested that staff at firms that stress high realization "may be disposed to under-record billable time" (JofA, Oct. 2023). The CPA Journal made the same point in 2021.

So the write-off you can see is only part of the overrun, because hours nobody put on a timesheet never reach the realization report. We find them by asking each person in their intake how long a task really takes and setting that next to the time records.

If you're wondering where your own firm's hours are leaking, the quiz below answers that in ten questions.

How does AI get a job back inside its budget?

Overruns usually come from the same kinds of work every time, such as chasing clients for documents, matching bank and card transactions, tying out workpapers and drafting the client letter. AI can do that work on a repeat basis while the person who owns the job checks what it produces, so the senior's hours go on the judgment calls and the job lands inside its budget. That way the firm recovers every hour it bills instead of writing chargeable time off, and revenue and gross margin both go up with the same headcount. The hours that free up go to the next client, so the firm takes on work it would have turned down, and it doesn't have to hire to do it.

Tax work needs more care. The IRS Office of Professional Responsibility's June 2026 guidance says billing clients for time that was not actually spent may breach section 10.27, the Circular 230 rule against unconscionable fees, depending on the facts (IRS OPR, Issue 2026-19). So on returns, the gain we count is the write-offs that stop and the extra clients the team has room for, and never hours nobody worked. The OPR director said in November that more guidance on AI and fees is coming (The Tax Adviser, Nov. 9, 2026), and nobody knows yet where it will land, so every solution keeps a record of what AI did on each job.

What does a firm need sorted before AI can move realization?

We look at each of the firm's AI layers first, meaning the business, each of its departments, the data and tools those departments rely on, and the people in each team, and we put right whatever we find wrong with its software, its data or how securely that data is held. We always say you need those solid foundations before you think about AI, and most of the time the tech stack and the data turn out not to be prepared for it.

For realization, the data that matters most is time. If fixed-fee work isn't timed, or the time codes don't say which part of the job the hours went on, nobody can see where a job ran over, and AI pointed at the wrong part of it moves nothing. The IRS guidance also says client data belongs only in secure AI tools the firm has approved (IRS OPR), and section 7216 has its own rules, which we cover in our page on client consent before AI touches a return.

How do you find the jobs that keep running over?

We map the work. In a short intake, every person in the firm tells us every task they do, how long it takes, what their hourly rate is and how they already use AI on it, and we put that next to each job's budget and how often that job runs over. That shows you whether the overrun on the monthly close is in the bank reconciliations or in the review. The map also picks up new work AI makes possible for clients, such as quicker answers in busy season, which brings in revenue nobody has budgeted for yet. We explain why the map comes before any tool in our guide to where a firm's AI work starts.

How do you tell which solutions will lift realization the most?

Every solution gets costed against those job budgets before anything is implemented, so the partners know, before a dollar is spent, how many hours it will free, what implementing it will cost and what it returns in year one. We take the hours a task takes today and multiply them by somewhere between 0.7 and 0.85, which leaves the job's owner time to check and correct the AI's work. You can trust what comes out, since every solution keeps the check its owner makes today.

To show the sum, here's an example with invented numbers, so it isn't a client result. A senior who bills at $160 an hour spends 5 hours a week on the reconciliations behind the firm's monthly client accounting jobs, and those jobs run over their budget most months. Multiplied by 0.7 to 0.85, that frees 3.5 to 4.25 hours of the senior's week, which over 48 working weeks is 168 to 204 hours, or $26,880 to $32,640 a year of chargeable time that stops being written off or goes to new clients. We do that sum for each task each person does, across the whole firm, and the results rank the solutions.

Each solution is also tested for compliance, governance and data security under the rules of whoever regulates the firm, which usually means the IRS, your state board, your peer review and the FTC's Safeguards Rule. The result decides where the work runs, whether on your own machine, in your own cloud account or, if it isn't sensitive at all, on Claude or ChatGPT's own servers.

The quiz below takes ten questions to show which of your teams' work would come out on top once it's costed this way.

What happens once every solution has a number on it?

The solutions are then put in order on a roadmap that says which go in first and how, with the most valuable at the top, all under the firm's own AI policy. Before anything goes live, the firm also gets an operating model showing where AI is working, for which people, and the hours it frees up for each team. That's where realization gets watched by service line, because a good firm-wide figure can hide one service line that writes off far more than the rest.

Implementation comes last, one solution at a time in roadmap order. Every solution comes with an enablement pack written for the people using it, and a named owner signs off its work. When a model is updated or a regulation changes, we update the solution too, so a job it brought inside budget in the spring stays there in busy season. Your team hears one clear message, that AI is there to support them and give them more time for better work, so they get behind it. The firm also owns the intelligence, so if a vendor puts its prices up the solution moves to another model, as we explain in why your firm should own the system and rent the model.

How much work does it take to lift realization across a whole firm?

Done properly, it's a big job, and we'd rather you knew that up front. Every person in the firm, partners and admin staff included, needs interviewing about each task they do, how long it really takes, their rate and how AI is used on it today. Each type of job needs its budget lined up against its actual hours, with every overrun traced to where it happens, and each department's data and tools need checking, with patchy time records cleaned up.

Then every solution needs costing against those budgets and checking against every rule the firm works under, including Circular 230, section 7216, the FTC and the state board. The roadmap, the AI policy and the operating model all have to be in place before the first solution goes live. The partners who could do all of that usually have the highest billing rate in the building, so every hour they give it is an hour of client work they don't do. It takes the weight off your shoulders, because everything is compliant, secure and planned before anything is implemented.

If you want us to do it for you, book a free call and we'll find you at least $50,000 a year in new business that AI can unlock. It's compliant, data-secure and has governance built in.

What do firms get wrong about AI and realization?

The mistake we see most is reading a high realization rate as proof there's nothing to fix, which misses all the time that never made it onto a timesheet. Another is buying a tool and pointing it at whichever job the partners complain about most, without knowing which part of that job runs over, so the tool speeds up something that was already on budget.

Some firms also free up the hours and never decide what they're for, and capacity that doesn't turn into anything for the business moves neither realization nor revenue. So the partners agree up front what each team's freed hours will go to, whether that's new clients, advisory work or getting through busy season without having to hire.

What can a managing partner check this week?

• Pull last year's write-downs and write-offs by service line and by client, and see where most of them sit.
• Compare budgeted hours with actual hours on your recurring jobs from last busy season.
• Ask which work isn't timed at all, because that's where overruns stay hidden.
• Ask your team which ChatGPT, Claude or Copilot accounts touch client data, including free and personal ones.

If you'd rather have the answers for your own firm before pulling any reports, the quiz below gets you there in ten questions.

Common questions

What is a good realization rate for a CPA firm?

It depends on how your firm measures it. The 2023 National MAP Survey put realization for all firms at 99%, and the Journal of Accountancy said a rate that close to the top may mean billing rates are too low (JofA, Oct. 2023). That figure only counts the time people recorded, so check how any benchmark was measured before you set a target against it.

How does this work on tax returns, given the IRS guidance on fees?

The gain on returns is the write-offs that stop and the extra clients the same team can take on, because the IRS guidance warns against billing for time that wasn't actually spent. When the OPR's promised guidance on AI and fees arrives, we bring every solution in line with it.

Does AI help realization on fixed-fee work?

Yes. A fixed fee is set from a budget of hours, so when AI keeps the job inside that budget, the job makes the margin the partner priced it to make. The work does need timing for a few weeks first, or nobody can see where it ran over.

This page is general information and none of it is legal or tax advice, so have your own attorney or tax adviser confirm how it fits your firm before you act on it.

Work with us to get every job finished inside its budget

At Golden Sky AI, we help companies in regulated industries implement AI that frees up their team's capacity, so they can take on more business without having to hire. Every solution is checked for compliance, governance and data security before we implement it, and you own the intelligence, meaning any AI agent could plug into the solutions we implement, as they're robust and still deliver the same quality of work. That means you're never held hostage by one AI provider.

We'll find you $50,000 a year in new business AI can unlock in 72 hours, or you don't pay.

Sources

• IRS Office of Professional Responsibility, Issue 2026-19, Introductory Guidelines for Responsible AI Use in Federal Tax Practice (June 24, 2026)
• 26 U.S. Code section 7216, Disclosure or use of information by preparers of returns (Cornell LII)
• FTC, Standards for Safeguarding Customer Information (Safeguards Rule)
• The Tax Adviser, IRS official signals additional guidance on AI, practitioner fees (Nov. 9, 2026)
• Journal of Accountancy, MAP Survey shows firms on upward trajectory (Oct. 1, 2023)
• The CPA Journal, How Realization Negatively Impacts CPA Firms (May 2021)
• Xero, Billable hours: how to stop letting revenue slip through the cracks