AI and the R&D Tax Credit: Powerful Tool, but Not a Push-Button Solution

AI is changing how companies approach the R&D tax credit. For businesses with large engineering and development teams, the appeal is easy to understand. Instead of manually sorting through thousands of Jira tickets, Git commits and project records, or having employees manually complete time surveys, AI can analyze large volumes of information quickly, identify patterns and help determine which activities may qualify for the credit.

That can make an R&D credit study more efficient. But there is an important distinction between using AI to help analyze your R&D activity and relying on AI to determine your R&D credit.

The quality of the result still depends heavily on the quality of the information you have to begin with.

AI Can’t Analyze What Isn’t There

One of the biggest limitations is also one of the simplest: AI cannot create documentation that doesn’t exist.

If your engineers and developers consistently track projects, tasks and commits in systems such as Jira or Git, AI may have a substantial amount of information to work with. If your project tracking is inconsistent, however, the picture can look very different.

This becomes especially important for employees who contribute to R&D but aren’t regularly writing or committing code. Product managers, engineering supervisors and managers, dev-ops and other network support personnel and other team members may spend significant time on qualified activities without leaving the same digital trail as a developer.

Even relatively small data issues can create problems. An employee may appear under different names in different systems. Tasks may be assigned to a department or job function instead of an individual. Records may not provide enough information to connect a specific activity to the employee performing it and ultimately to the wages being included in the credit calculation.

AI can analyze the records you give it. It can’t tell you with certainty what is missing.

Don’t Assume Everyone Was Captured

Completeness is another area where human review matters.

We’ve seen situations where AI analysis of project-management data did not capture a significant portion of the engineering population. Without comparing the output back to employee and payroll records, that gap could easily have gone unnoticed.

If you’re using an AI-enabled R&D credit platform, one of the questions to ask isn’t simply, “What did the system find?” It is also, “Who and what did the system miss?”

Reconciling AI results against engineering and development headcount, payroll information and other company records can help identify employees or activities that never made it into the analysis.

Surveys May Still Be Necessary

AI is sometimes positioned as a way to eliminate R&D surveys entirely. In practice, that depends on the underlying data.

If your systems don’t contain enough information to determine how employees divided their time among business components and activities, you’ll still need another way to capture it. Surveys and interviews remain useful for filling those gaps.

This is particularly important when project records show that someone worked on a project but don’t provide enough detail about what that person actually did or how much time was spent on potentially qualified work.

The goal shouldn’t be to eliminate every manual step. It should be to use technology where it improves the process while still collecting the information needed to support the credit.

Classification Still Requires Judgment

AI is particularly good at analyzing large datasets and identifying relationships among projects, epics, issues, task types and employee roles. Based on those relationships, it can recommend whether an activity appears qualified, non-qualified or needs further review.

But those recommendations aren’t the same as tax conclusions.

Clearly defining what your organization considers qualified research and requiring the system to provide the basis for its classifications can make the output much more useful. Items identified as “Needs Review” or “Likely Non-Qualified” should receive additional attention rather than simply being accepted or discarded.

There are also areas where many AI-based R&D platforms still have limitations. State R&D credit calculations may require separate work, and situations involving short tax years, acquisitions or dispositions can add complexity that a standard automated process may not handle correctly.

Use AI to Strengthen the Process, Not Replace It

AI has the potential to make R&D credit studies faster and more efficient, particularly for companies with substantial project data. It can help your team work through information that would be difficult to analyze manually and focus attention where additional review is needed.

The strongest approach, however, combines that technology with experienced tax judgment and a clear understanding of your business.

Before relying on an AI-generated R&D credit analysis, make sure someone is asking the questions the software may not: Is the source data complete? Are all relevant employees represented? Do the classifications make sense? Are state and transaction-related issues addressed? And is there enough documentation behind the result to support the credit if it’s examined?

AI can be a very useful part of the R&D credit process. It just shouldn’t be the only part. Start a conversation with our team today.