A SaaS company spent four months building a real-time sync feature, decided it could not scale, and scrapped it. The engineers moved on. The founder wrote it off as a lesson. It was also a qualified research expense, and the credit on those four months of payroll was real money.
Software companies are the clearest fit for the R&D credit in the entire tax code, and they under-claim it for a specific reason: the credit rewards uncertainty, and founders are trained to talk about certainty.
What qualifies
The federal credit under Section 41 uses a four-part test. For software, it reads like a description of the engineering process.
You were trying to build something new or improve function, performance, or reliability.
The work relied on computer science.
At the outset, you did not know if it would work, how to build it, or what the right design was.
You evaluated alternatives through testing, prototyping, or iteration.
Common qualifying work includes new product development, new features that required architecture decisions, performance and scalability work, integrating systems that were not designed to work together, security and encryption implementations, algorithm development, and internal tools built to solve a problem off-the-shelf software did not.
Work does not have to succeed. Abandoned features count. Rewrites count. The proof of concept that convinced you not to build something counts.
What does not
Routine maintenance and bug fixes on stable code. Configuration of purchased software without technical uncertainty. UI polish that did not involve engineering problems. Work done by developers outside the United States. Adapting an existing product for a specific customer without new technical challenges.
What counts as qualified spend
Wages for engineers, product managers who directly supervise or support development, QA engineers, and DevOps staff, allocated to qualifying projects.
Cloud and hosting costs used in development and testing environments, which have become a meaningful category.
Sixty-five percent of what you pay U.S.-based contract developers.
The credit is generally 6 to 8 percent of that qualified spend at the federal level. Many states add their own credit.
The payroll tax offset for startups
This is the piece most early-stage founders miss. A qualified small business can elect to apply up to $500,000 of the R&D credit per year against the employer portion of payroll taxes rather than income tax. To qualify, the company must have under $5 million in gross receipts for the year and no gross receipts more than five years back.
For a pre-revenue or pre-profit startup, this turns a credit that would otherwise sit unused into cash reduction on quarterly payroll deposits. A seed-stage company with $1.5 million in engineering payroll can see a five-figure reduction in payroll taxes starting the quarter after the election.
The election is made on the original return. It cannot be added later on an amended return, which makes timing important.
2026 context
Domestic development costs are fully deductible again under Section 174A. The amortization rule that hurt software companies from 2022 through 2024 no longer applies to U.S. work. That change is separate from the credit, but it simplifies the analysis. If you have unamortized balances from those years, you have a transition decision on your 2025 return. If you never claimed the credit for open prior years, amended returns can generally reach back three years.
Foreign development is still amortized over fifteen years and does not qualify for the credit. Companies with offshore teams should separate the two cleanly.
Documentation that survives review
Ticketing systems and version control are your best friends here. Jira epics, GitHub commit history, sprint planning documents, architecture decision records, and design docs establish what problem you were solving and what alternatives you tried. Time tracking by project, even if approximate and reconstructed from sprints, ties payroll to work.
The studies that get challenged are the ones that claim a flat percentage of all engineering payroll with no project detail. The studies that hold up name the projects, describe the technical uncertainty, and show the iteration.