AI is working in tax. Scaling it is the hard part.

Most tax leaders can point to real progress with AI. Pilots have delivered measurable results. Planning cycles are compressing. Compliance workflows are getting faster.

And yet, extending that value consistently across the tax function has proved more difficult than the early momentum suggested.

According to the RSM Middle Market AI Survey 2026, 51% of respondents that ran AI pilots in the past 24 months describe their success as only moderate or limited. The most frequently cited reasons for AI pilots failing to scale among these respondents? Data quality issues (53%) and integration challenges (47%).

"Tax leaders’ day-to-day experiences reflect this in‑between phase,” says Brad Collins, RSM US principal and tax digital services go-to-market leader. “AI is delivering value, but scaling it cleanly across the function is harder than expected.”

For tax leaders, AI has reached a stage in which three unresolved signals of maturity largely determine whether adoption translates into durable value:

Defining AI maturity for tax functions

Maturity, in this context, is not about how widely AI is used or how sophisticated the tools are. It’s about the degree to which a tax function can embed AI into core tax workflows in ways that are trustworthy, governable, repeatable and defensible.

That distinction matters because broad adoption can mask uneven foundations.

Across the full survey sample, 86% of respondents report some level of AI integration, and 97% express satisfaction with the business value delivered so far.

But among respondents whose organizations currently use AI in tax and are pursuing or planning tax-related AI use cases, the picture is more nuanced: Only 8% place themselves at the leading level.

Across the full survey sample, 45% of all respondents describe their overall AI strategy as pragmatic—implementing AI where it delivers clear value today, rather than pursuing enterprise‑wide transformation.

So although the ambition is evident, the maturity required to act on it is still forming.

Tax data readiness: The foundation of AI maturity

Of the three signals, data is the most foundational, and the other two depend on it. Governance cannot function without reliable inputs. People cannot exercise judgment confidently if the data behind AI outputs is inconsistent or unexplainable.

The survey reinforces this clearly. Beyond the data quality issues already cited as a leading cause of stalled pilots, “data quality/availability/lineage issues” ranks as the single greatest inhibitor to AI deployment overall, cited by 34% of respondents.

When asked in an open‑ended question what one thing would most accelerate AI impact, respondents generally returned to the same theme: clean, well‑governed, accessible data.

For tax functions specifically, the challenge is structural. Tax data typically spans multiple systems—provision, compliance, payroll, fixed assets, legal entity structures—and often varies in format, ownership and reliability across jurisdictions and processes.

"Most tax teams aren't starting from zero on data, but the work of standardizing, cleaning and connecting it across systems is where progress stalls," says Matt Bradvica, RSM US tax digital strategy leader. "That's the work that doesn't feel like AI, but it's what makes AI usable."

Until that foundational work is treated as a strategic priority, data will continue to constrain how far AI can reach inside the tax function. Data that cannot be explained or reconciled creates a ceiling not just for AI performance, but also for the governance and review structures that depend on it.

AI governance in tax: From policy to practice

Most organizations are not ignoring AI governance. According to the survey, 98% have risk controls in place to prevent or reduce AI‑related harms. What varies is how deeply those controls reach into the way tax professionals actually work.

When asked at what stage governance controls are applied, 48% of respondents said before production. However, 16% apply controls only after issues arise, and another 7% apply them inconsistently. Only 36% cite model documentation and lineage tracking as a current practice.

In tax, auditability and sign‑off are embedded in the culture of the work itself. So the key question is: To what degree does governance operate inside the daily rhythm of tax execution?

When governance lives outside provision, compliance, reporting and review workflows, it tends to function as oversight after the fact rather than discipline during the work. The more closely governance aligns with how tax professionals already work, the more naturally AI outputs become defensible.

Tax AI and the workforce: Defining where judgment matters

The survey leaves little doubt that AI will reshape tax work. Among respondents who qualified to answer tax-specific questions, 98% expect AI to impact the nature of work within their tax function in the next two to three years.

And the direction of that impact is consistent:

Less clear is whether those evolving roles have been defined well enough for the broader workforce to step into them confidently.

Across the full survey sample, 85% of respondents agree executive leadership is more enthusiastic about AI than employees are. That gap is not necessarily a sign of resistance. More often, it reflects uncertainty about what is expected, where human judgment remains essential, and how accountability will work in practice.

“One theme that’s resonating with tax leaders is about capacity and skill gaps—all this modernization and technology is meant to scale their existing people because they’re not able to get more,” says Courtney Brown, Partner and RSM Canada Tax Applications and Transformation Co-Leader. “AI is about extending judgment and capacity, not taking it away.”

For tax functions, maturity on this dimension is less about training programs and more about organizational clarity. Specifically, defining where AI accelerates work, where humans review it, and where responsibility ultimately rests.

Why tax AI maturity depends on all three signals

For many tax leaders, all three of these signals are developing at different speeds. Progress in one often reveals strain in another. The gaps create friction that can be difficult to diagnose.

Data may be improving, but governance has not yet adapted to the pace of AI‑generated outputs.

Roles may be evolving, but without clear boundaries around judgment and accountability, confidence lags behind capability.

Controls may exist on paper, but if they do not align with how tax work actually moves through review and approval, they offer structure without traction.

None of this means the investment in AI has been misplaced. It means the organizational foundations are still catching up. Maturity does not arrive evenly across data, governance and people. Expecting it to can lead to frustration that the underlying progress does not deserve.

In fact, recognizing that unevenness is itself a step toward resolving it.

Advancing tax AI maturity: What comes next

Tax AI maturity advances when organizations invest in data foundations that earn trust, embed governance into the workflows where tax professionals already operate, and define roles and accountability structures clearly enough that the workforce can move forward without ambiguity.

Much of this work will look more like process discipline than innovation. But it is precisely this work that will determine whether AI becomes a reliable part of the tax function or remains a collection of useful but isolated capabilities.

Tax AI has reached the point where progress depends on seeing these signals clearly. The opportunity to advance is real. So is the work required to get there.