Artificial intelligence (AI) no longer is confined to technology teams or innovation labs. It increasingly is embedded across an enterprise, shaping how organizations operate and make decisions, including how they support tax teams.
As AI becomes part of day-to-day operations, it can influence the underlying facts that support a research and development (R&D) tax credit claim, extending beyond how the credit is calculated and into how qualified research is identified, scoped, and substantiated.
For tax functions, this shift calls for more than technical familiarity with AI; it also requires reassessing whether existing R&D study approaches still reflect an organization’s operational reality.
Who Performs R&D When AI Is Everywhere?
For decades, identifying qualified research followed a familiar pattern of engineering, product development, and information technology (IT) teams conducting R&D. Tax professionals knew where to look.
AI invites a reassessment, not because the definitions of qualifying R&D have changed (they have not), but because the way innovation occurs is changing. As organizations deploy AI-enabled tools, experimentation might occur in places that historically were not considered R&D.
For example, an operations team builds and iteratively tests a customer-facing, AI-driven forecasting or optimization tool to resolve performance, reliability, and integration uncertainties not known at the outset. Much of this might be routine implementation, but some efforts could involve qualified research and require careful fact review.
Not all AI-related activity qualifies as research. In many cases, it will not. But a look at how AI adoption extends across the organization can raise important questions, including:
- Which teams are evaluating performance, reliability, or quality under constraints that are not fully known at the outset?
- Where does technical uncertainty arise as teams test, refine, or validate AI-driven solutions?
- Can technical uncertainty and the process of experimentation be clearly tied to principles of computer science or engineering, rather than to functional or operational decisions?
- Are existing R&D study approaches designed to identify experimentation when it emerges outside traditional R&D teams?
- Are stakeholders aware of how qualified research is defined, even if they do not view themselves as performing R&D?
Titles and team names matter less than the underlying facts of technical uncertainty and experimentation. When innovation becomes more distributed, tax professionals face a choice: they can continue to scope studies based on historical organizational boundaries, or they can make sure their qualification analyses reflect how experimentation actually occurs within the business. Both options have real risks attached. Overqualification—trying too aggressively to count costs as R&D expenses—invites audit exposure, whereas underqualification can result in credits for eligible costs going unclaimed as the claim diverges from operational reality.
Thus, as AI reshapes work, tax professionals should revisit with fresh eyes whether their current R&D tax credit methodology remains well-calibrated to the evolving nature of innovation.
How AI Might Influence R&D Tax Credit Claims
Under Internal Revenue Code Section 41, qualified research expenses (QREs) include:
- taxable wages paid to in-house employees who perform, directly supervise, or directly support qualified research;
- tangible supplies used in conducting qualified research;
- computer rental costs (including certain cloud computing costs) used in qualified research; and
- contract research expenses, generally limited to sixty-five percent of amounts paid to third parties to perform qualified research on the taxpayer’s behalf, subject to rights, risk, and US location requirements.
In practice, wage-based QREs often represent the largest component of an R&D tax credit, reflecting the number of individuals engaged in qualified research. Recent headlines underscore why this factor matters. Organizations are announcing workforce reductions and hiring pauses, alongside increased investment in automation and AI-enabled tools. Often, AI is discussed as a means to improve efficiency, speed, and scalability, even as organizations reassess staffing models and resource allocation.
Against that backdrop, tax professionals might begin to observe new trends in research activities. As machine-assisted tools support testing, iteration, and development, a natural question emerges: Will different patterns develop in how many individuals qualify for the credit? Changes in headcount, team structure, or third-party resourcing could affect wage-based QREs, even as research activity and innovation continue to expand.
As organizations adjust their resourcing models, increased reliance on foreign contractors or globally distributed teams could reduce R&D tax credit eligibility, since QREs are limited to US-based activities. Those same shifts also carry IRC Section 174 implications, requiring foreign research to be capitalized and amortized over fifteen years.
AI adoption also often coincides with significant investment in cloud computing for model training, data storage, and deployment. Although these investments can be substantial, not all related costs qualify for the R&D tax credit (for example, software licenses fall outside the definition of QREs). As a result, some companies might experience a gradual change in their QRE profile, with a greater share of costs associated with computer rental arrangements rather than with traditional wage or contract research expenses. Thus, tax professionals should consider:
- how changes in the mix of employee wages, US-based contract research, and cloud computing costs compare to prior years, and whether trends can be clearly explained, substantiated, and defended;
- whether existing R&D study approaches are designed to capture these changes and align qualification methods with how research is actually performed; and
- how to make sure descriptions of AI-enabled activities do not unintentionally suggest routine automation or implementation rather than technical uncertainty and experimentation as AI terminology becomes more common in documentation and interviews.
The next practical step is to monitor year-over-year QRE mix changes and align the documentation and the narrative so they explain both the activity and the shift in costs.
The Evolving Role of an R&D Tax Professional
Not long ago, preparing and defending an R&D tax credit meant boxes of invoices and physical stacks of documentation, with manual analysis and review—a human exercise of tracing facts to conclusions.
Since then, digital records, analytics, and collaboration tools have already changed how tax professionals work. As AI reshapes operations and the associated costs, it also marks an inflection point in how R&D tax credit claims are prepared, reviewed, and defended. This inflection point reflects a broader Industry 5.0 shift, one in which advanced automation augments, rather than replaces, human expertise. For R&D tax professionals, technology can enhance efficiency and insight, but judgment remains central, raising new questions:
- Where does AI meaningfully enhance an R&D study, and where does professional judgment remain essential?
- How do tax professionals make sure that outputs that appear complete also reflect the underlying facts and the law?
- When analysis becomes faster, how do tax professionals confirm that outputs are reviewed, challenged, and validated rather than simply accepted?
The future R&D tax professional must be human-centered and technology-enabled. Technology can support efficiency and insight, but judgment remains central. This evolution also changes the training professionals need. As AI automates tasks that once served as learning tools, organizations must be intentional about developing the ability to evaluate uncertainty, apply the regulations thoughtfully, and defend positions under scrutiny. Speed might change the process, but it cannot replace understanding and positioning.
What This Moment Calls For
For tax professionals, a key consideration is how emerging AI-driven trends can gradually influence who performs qualified research, which costs are incurred, and how claims are evaluated.
The most effective tax functions will respond deliberately by asking tailored questions, aligning R&D tax credit claims to operational reality, and using technology to enhance insight without replacing professional judgment as AI reshapes how innovation occurs.
Sophia V. Shah is a partner, tax, at Crowe.
