Tax Practice Automation Adoption Challenges

Most tax practices have not rejected automation. They have climbed partway up the adoption curve and stopped there, and the stall looks like reluctance from the outside when it is really a structural ceiling. Thomson Reuters' 2026 State of Tax Professionals Report found that only a small fraction of firms report using no automation at all, so the profession has largely stepped onto the curve already. Yet 27% of firms automate only up to half of their workflow, so broad entry into automation has not turned into deep or complete implementation anywhere close to universally. The space between a firm that uses some automation and a firm whose practice runs on it is where most of the profession actually sits, and that space is what this piece examines.

The four structural barriers that explain the stall

The reasons firms stop climbing are concrete, and none of them need you to assume anyone is dragging their feet on principle. 51% of respondents in the Thomson Reuters report cite lack of time and resources as a top barrier to further automation, with cost of implementation close behind; workflow complexity and internal resistance to change follow after that. These four sit in a rough order of weight, but they are not independent problems stacked side by side. A firm short on time during busy season cannot free up staff to evaluate a new tool, let alone implement one, so the time shortage raises its costs the moment a firm tries to buy its way past it with outside help. Progress Software's State of Accounting 2026 found that a majority of accountants report being slowed by switching between too many tools, and a large share cite a lack of automation for routine tasks, so the tool sprawl firms already carry is itself adding friction to any further automation effort. Workflow complexity then compounds the cost problem in a specific way: before AI can run on a tax workflow at all, the underlying data usually needs to be cleaned, consolidated, and reconciled, and that work has to happen before any return on the automation investment appears. None of this is a story about firms doubting the value of automation. It is a story about firms running current-season client work with no slack left over to build the next layer of infrastructure on top of it.

The data infrastructure problem that precedes everything else

A prerequisite most firms do not plan for is data infrastructure readiness, and discovering the need for it mid-implementation is what turns a budgeted project into an open-ended one. Thomson Reuters reports that many tax departments find they have to address their data infrastructure before they can implement any AI strategy at all, because the data a firm needs often lives across multiple ERP systems, requires manual transfers between them, and takes hours of reconciliation before anyone can act on it. Accounting Seed's State of AI in Accounting 2026 found that only 16% of finance teams have implemented AI in day-to-day workflows, and the distance between that number and the widespread interest in AI is a readiness gap, not a sign that firms doubt the technology. The real cost of automation includes this upstream data work, and firms typically find out how much of it they need only after they have already committed budget and staff time to an implementation.

The corporate direct tax world makes this concrete rather than abstract. For the GloBE Information Return due 30 June 2026, you do not get stuck on the rules. It is the workflow: reviewing data at the jurisdiction and entity level, collecting inputs from systems that were never built to talk to each other, validating the XML output, and making sure every judgment call made along the way can hold up under audit. A firm can understand the GIR requirements perfectly and still miss the deadline because its data was never in a state that any tool, automated or not, could act on.

How firm size widens the competitive gap between firms

These barriers do not land on every firm with equal force, and the unevenness is turning a shared structural problem into a competitive one. Thomson Reuters found that among large firms, 63% describe themselves as being in the proactive or optimized stage of technology development, but among the smallest firms that figure drops to roughly a third. Large firms have already done the infrastructure work that smaller firms are still stuck on, and they have moved on to using automation for workflow enhancement and advisory expansion, which puts them in a different kind of competition altogether. Thomson Reuters also points to intensifying competition from private equity-backed firms, mergers, and tech-enabled entrants that can deliver compliance work at substantially lower cost, and that pressure falls hardest on the smaller practices that are least equipped to automate in response. Midsize firms face a related but distinct squeeze: they often cannot match the salaries and career paths that larger firms offer, so they feel the talent pressure more acutely than either their larger or smaller counterparts.

The mechanism that makes this gap durable rather than temporary is simple to state. If a firm cannot automate its routine work, it cannot free up staff time to plan or build the next stage of automation, so the ceiling it hit last year comes back around this year. Firms that got their data ready for automation early are compounding gains while firms still readying their data are standing still, and the distance between the two groups grows every cycle that passes without a change in course.

Clients notice the gap before firms do

The stall is not only an internal operating problem. Client expectations for AI-enabled service have moved faster than most firms' implementations, and that gap is now producing real attrition risk with quantifiable stakes. Clients have already formed opinions about it, often before the firm serving them has noticed anything is wrong. Thomson Reuters' Future of Professionals Report 2026 found that a large majority of corporate clients consider it very important or essential that their providers deliver AI-enabled quality improvements, yet only a small fraction of those clients say most or all of their current providers are delivering on that expectation. That mismatch between what clients expect and what firms are providing is already producing decisions, not just dissatisfaction. Thomson Reuters found that 32% of corporate clients have already reconsidered, or plan to reconsider within the coming year, their relationship with a firm they believe is falling behind on AI, and a third of that group estimates more than a million dollars in annual work is at stake.

The same commoditization pressure that makes automation urgent for firms is also what clients use to judge whether a firm is worth keeping. A firm still handling routine compliance work by hand is the firm most exposed to exactly the clients most likely to walk. None of this means the relationship is already lost. It means the client has drawn a conclusion the firm may not yet know about, and the firm still has time to close the gap before that conclusion becomes a decision.

The governance deficit that partial adoption creates

Stalling in the middle of implementation is not a safe place to sit, because the absence of policy does not mean the absence of AI use inside a firm. It means AI use is happening without anyone tracking it. Accounting Today found that a substantial share of tax and audit professionals already use AI tools their firm has not authorized for client work, and that kind of unauthorized use, often called shadow AI, leaves no record of what the tool was given, what it produced, or whether anyone reviewed the output before it reached a client. Financial Cents' 2026 State of AI in Accounting & Bookkeeping found that the vast majority of firms using AI have no formal written AI policy at all, so most firms running client work through these tools have no documented rules about what information can or cannot go into them.

Shadow AI is a direct consequence of the stall. Staff who can see the time savings available to them, but whose firms have not given them a sanctioned tool or a clear policy, will find their own workaround, and every month a firm delays its official adoption decision is a month in which unofficial adoption keeps expanding. The regulatory exposure this creates is not hypothetical. IRS Circular 230 guidance released by the IRS Office of Professional Responsibility made clear that existing professional standards apply fully to AI use: practitioners remain responsible for accuracy, for client data confidentiality, and for the reasonableness of fees, and they must review every AI-generated document thoroughly before it goes to the IRS or to a client. The same guidance also requires firms to vet third-party AI tools, set up secure data-handling protocols, and train staff on the risks involved. Bloomberg Tax has noted separately that when AI tools fail to provide clear audit trails or cite their sources, they undermine confidence in the output and create compliance risk on their own, independent of whatever policy a firm has written down, so tool selection becomes part of governance.

The skill erosion risk that runs alongside the governance risk

A second risk runs beside the governance one, and it concerns not whether a firm automates but what it automates first. Automating judgment-forming work before routine work risks weakening the analytical skill a firm depends on to catch an AI system's errors, and that dependency is rarely planned for in advance. Former IRS Commissioner Danny Werfel, writing in The Tax Adviser in August 2026, warned of exactly this pattern: staff build expertise in operating the AI tool rather than in the tax law the tool is applying, and firms gradually lose the independent analytical capability they need either to catch the tool's errors or to handle complex matters without it. Thomson Reuters' Future of Professionals survey found that a large majority of professionals believe human judgment matters more, not less, as AI spreads, but the profession has not settled on consistent practices for preserving the judgment it claims to value.

Some of the resistance to automating judgment work has a plainer economic explanation. Peer-reviewed research published in ScienceDirect found that at Big Four firms, the reliance on hours billed as a primary revenue source is itself a substantial structural obstacle to automation, since automating billable judgment work cuts directly against how those firms get paid. The choice, in other words, is not whether to automate judgment-intensive work eventually but which workflows a firm automates first, and in what order, so that staff development and automation build on each other instead of working against each other.

What moving past partial implementation requires

A firm needs more than a single decision to get past partial automation. It requires working through the structural barriers in a deliberate sequence: data readiness first, then tool selection matched to the specific workflow causing the most friction, then governance policy that covers how the tool gets used. Thomson Reuters' own analysis backs this ordering: the strongest return on AI investment comes from automating the highest-friction workflows first and measuring the results against a clear framework, rather than adopting tools piecemeal and hoping the gains appear on their own.

In practice, that means starting with the workflows that eat the most time relative to how little judgment they actually require, client intake, document collection, and routine compliance checks are the clearest examples, where automation cuts the burden without touching the work staff need to develop their own tax judgment. Thomson Reuters ONESOURCE describes one version of this approach as "touchless compliance," a model built around being compliant by default and manual only by exception, in which agentic AI handles data import, validation, and mapping while direct tax professionals keep oversight and spend their time on the judgment-intensive parts of the work. That framing captures what separates firms that move past the stall from firms that remain in it: the data has to be ready, the tool has to match the actual point of friction, and the policy governing both has to exist before, not after, the staff start relying on them.