AI adoption is no longer a fringe question in accounting. In Thomson Reuters' 2026 Future of Professionals report, 74% of professionals said they use AI several times a week, and within tax firms specifically, about a third report their organization is already using generative AI. The tools exist. The harder question, and the one this post is about, is AI ROI for accounting firms: what the investment actually returns, and how you would know.
Here is the uncomfortable part. That return is rarely measured. Only 18% of professionals say their organization tracks it at all. Plenty of firms are paying for tools without a clear read on what comes back. That gap, not the technology, is the real problem, and it is fixable.
This post weighs the three things you actually need: the benefits AI delivers, the costs that rarely make the pitch deck, and how to measure payback so you are managing an investment instead of guessing.
The most-cited return is time. Respondents to Thomson Reuters' 2025 survey predicted that AI would save professionals about five hours a week, which the publisher valued at roughly $19,000 a year per person. Treat that as potential capacity rather than realized profit: it is what practitioners expected the tools to free up, not an audited result.
Reclaimed time is not the return. What you do with it is. Hours that used to go to manual research and data entry can move to review, planning, and advisory work that bills at a premium. That shift, from compliance throughput to advisory margin, is where the dollars actually show up.
Two second-order returns are easy to leave out of the math. The first is retention. In the same report, 24% of professionals who feel their organization is underdelivering on AI said they are considering leaving within two years, and access to professional-grade tools is becoming a factor in whether people take or keep a job. The second is client demand: 78% of corporate clients call AI-enabled quality improvements essential or very important, while only about 6% say most of their providers deliver. The firms that close that gap have room to win work and keep it.
The subscription is the most visible and easiest number to find. The costs that actually determine payback are the ones that never appear on a pricing page. Build your estimate across all of them:
Two are especially easy to underestimate. Training and change management decide whether the tool gets used at all, since the fastest way to waste AI spend is buying software staff never adopt. And expect a productivity dip before the lift as people learn the new workflow.
When client tax return information will be entered into a third-party AI system, the firm has to work through Internal Revenue Code Section 7216 before the tool is approved. Consent is not automatically required in every case. The regulations permit a range of disclosures and uses without it, and whether you need consent depends on the purpose, the vendor's role, where the information is accessed, and whether a return-preparation or auxiliary-services exception applies.
The FTC Safeguards Rule is a separate obligation. A vendor that receives, stores, or can access customer information generally counts as a service provider under the Rule, which means vetting its safeguards, setting security expectations by contract, monitoring the relationship, and updating your written information security plan when something material changes. Using software does not shift a practitioner's applicable Circular 230 duties. None of this blocks adoption. The mechanics deserve their own careful treatment, but for the ROI calculation the point is simpler: they take time, and that time belongs in the cost column.
The largest cost is subtler: paying for AI and getting little back. The problem is no longer adoption alone. Access and execution matter just as much. In the 2026 report, 41% of professionals said they lack tools actually built for professional work, and 34% admitted to using AI their organization has not sanctioned, a sign that risk is moving faster than oversight. The dividing line is strategy. In firms with a named AI strategy, 66% of professionals said AI is meeting or exceeding expectations; where there is no active strategy, that figure drops to 22%. A tool bought without a plan can generate exposure faster than return.
The standard equation is straightforward: ROI equals measurable benefits minus total costs, divided by total costs. For reclaimed time, count only the portion converted into additional capacity, collected revenue, reduced labor cost, or another measurable outcome. Costs include subscriptions, implementation, training, and the ongoing time spent on governance and review.
The variable that decides everything is what happens to the reclaimed hours. If you bill by the hour and simply finish sooner, you can cut your own revenue. The return only materializes if you redeploy the time: more clients at the same headcount, or a move into advisory work priced on value rather than hours.
Run the math with your own conservative inputs. Assume a five-person firm adopts a tool at $150 per user per month, treated here as an assumption rather than a market benchmark: about $9,000 a year, call it $14,000 in year one with onboarding and training. If each person reclaims three hours a week but none of it is redeployed, the return is zero, no matter how good the tool is. Redeploy half of those hours into additional work that produces $200 per hour of collected revenue, and the firm generates $19,500 in incremental revenue over a 13-week quarter, enough to cover the assumed $14,000 year-one outlay before quarter-end. For a stricter profitability calculation, use contribution margin rather than gross billings. The tool did not change. The decision about the hours did.
On timing, be skeptical of any universal payback number. A focused research tool can show gains within one busy season, while anything needing integrations and firm-wide process change takes longer. Baseline your metrics before rollout, re-measure at 90 days, and reassess at six and twelve months.
Because only 18% of surveyed professionals say their organizations collect ROI metrics, the practical fix is to pick your metrics before you buy. Choose two or three you can baseline now and re-measure later:
Track adoption separately. It explains why expected ROI did or did not materialize, but usage alone is not a return.
One caution that protects the whole calculation: AI that makes review harder destroys ROI. If a partner cannot verify a citation quickly, the rework eats the savings. Weight citation quality and auditability heavily when you choose a tool.
Research is one workflow where the return can be measured directly: time to a usable answer, time to verify the authorities behind it, and time to turn it into a review-ready deliverable. Marble's Intelligence agent is built around that research-to-draft loop. Ask a federal or state tax question in plain English, get a citation-backed answer that links directly to the relevant federal or state authority, and turn it into a memo or client letter ready for your review. Join the Marble waitlist.
There is no universal figure. The timeline depends less on the tool than on two things: how fast staff actually adopt it, and whether you redeploy the freed-up hours into billable or advisory work. A narrow research tool may pay back within a filing cycle; a firm-wide rollout takes longer.
Two, usually. Not baselining before rollout, so there is nothing to measure against, and assuming faster work automatically means more profit. If you bill hourly and simply finish sooner without redeploying the freed-up time, AI can shrink revenue rather than grow it.
Small firms can absolutely see ROI. They often have fewer approval layers and can concentrate adoption on a few repeatable workflows, which helps. They also have less room to absorb a failed rollout, so the tool should solve a clearly identified problem without heavy customization.
Implementation and integration, training and change management, and the productivity dip while staff learn the new workflow. Governance is the other blind spot: a Section 7216 analysis, reviewing and updating the firm's written information security plan where necessary, and ongoing vendor oversight under the FTC Safeguards Rule all take time that belongs in the calculation.
AI ROI does not require cutting headcount. A firm can capture value by adding capacity, easing seasonal overload, improving turnaround, or shifting professional time into planning and advisory work. Whether staffing changes follow is a separate management decision, not an automatic result of adopting the technology.
This article is a general discussion of certain accounting and tax developments and related topics of interest and should not be relied upon as accounting or tax advice. If you require accounting or tax advice you should consult a qualified practitioner.
