A machine shop in Colorado spent most of 2024 fighting a tolerance problem on a new aerospace part. They tried three fixtures, changed the tooling twice, and scrapped a lot of material before the process held. The owner described it as “a bad year.” A specialist described it as qualified research.
That gap is the whole story of the R&D credit for manufacturers. The best estimates say two out of three eligible businesses never claim it, and manufacturers are the largest group of non-claimants because they assume research means lab coats. It does not.
The four-part test, in plain English
The federal credit under Section 41 applies to activities that meet four conditions.
Permitted purpose: you were trying to make a product or process new or better in function, performance, reliability, or quality.
Technological in nature: the work relied on engineering, physical science, computer science, or similar disciplines.
Elimination of uncertainty: at the start, you did not know whether you could do it, how you would do it, or what the design should be.
Process of experimentation: you evaluated alternatives. Trial runs, prototypes, simulations, iterative testing.
Notice what is missing. The work does not need to succeed. Failed and abandoned projects count. The work does not need to be new to the world, only new to your company. And nobody needs a PhD.
What typically qualifies in a plant
Developing a new part or product, including first-article work and prototype iterations.
Process improvement: new fixtures, tooling, jigs, or work-holding designed to hit a spec you could not hit before.
Automation and robotics integration, including the programming and debugging time.
Material substitutions and testing, such as switching alloys or coatings to solve a performance problem.
CNC programming for complex geometries where the path had to be developed and proven.
Quality and yield improvements that required experimentation rather than a settings tweak.
Software you built for your own operations, such as custom scheduling or inspection tools.
What usually does not
Routine production once the process is proven. Cosmetic changes. Market research. Reverse engineering a competitor’s part without technical uncertainty. Work performed outside the United States. Quality control inspection that is simply checking against a known standard.
How the credit is calculated
The credit is generally worth 6 to 8 percent of qualified spend, taken dollar for dollar against federal income tax. Qualified spend includes wages for people doing, supervising, or directly supporting the research, supplies consumed in the process (including scrapped material), and 65 percent of contract research paid to outside engineers or testing labs.
Many states add their own credit on top. Combined federal and state benefit can run higher than the federal figure alone, depending on where you operate.
The 2026 landscape
Two changes make this year different. First, domestic research costs are fully deductible again under Section 174A for tax years beginning after December 31, 2024. The amortization rule that punished research spending from 2022 through 2024 is gone for U.S. work. Second, the credit is now cleanly separated from that deduction question, so the decision to claim is simpler than it was a year ago.
Missed years are still recoverable. Amended returns can generally reach back three tax years, so a first-time claimant is usually looking at multiple years of credit, not one.
Who is claiming it
This is not a big-company credit. IRS data shows that about 45 percent of businesses claiming the federal research credit have revenue under $5 million. The reason larger manufacturers claim it more consistently is not that they do more research. It is that they have someone whose job is to look.
Documentation that holds up
The credit is examined on evidence. Contemporaneous records matter: project lists, engineering change orders, time allocations, test logs, scrap reports, and emails that show the problem you were solving. A good study interviews the people who did the work and ties their hours to specific projects. A weak study estimates a percentage of payroll and hopes. Ask which one you are getting.