What is the R&D tax credit? A plain-English guide
§41, qualified expenses, and what the credit is actually worth — written for the conversation you’ll have with a client, not for a tax journal.
6–10% of qualified research expenses, back as a federal credit.
Most explanations of the R&D tax credit are written for people who already understand it. This one is written for the conversation you’re about to have with a client who asked whether they qualify, and wants an answer in the next two minutes.
It’s a credit, not a deduction
The federal research credit lives in §41 of the Internal Revenue Code. It rewards companies that develop or improve products, software, or processes in the United States, and it comes off tax owed dollar for dollar rather than off taxable income. That distinction is the first thing to say out loud: a $40,000 deduction is worth whatever the client’s marginal rate makes it worth, while a $40,000 credit is worth $40,000.
Roughly 36 states run a credit of their own on top of the federal one, computed on the same underlying expenses after apportionment. For a company concentrated in a generous state, the state credit can add materially to the federal number.
The four-part test, in the words a client uses
An activity qualifies if it passes four tests. The statutory language is dense; the practical version is not:
- Permitted purpose — the work aims to improve the function, performance, reliability, or quality of a product, process, or piece of software the company develops or uses.
- Technological in nature — it relies on engineering, computer science, physical or biological science. Not on market research, not on design taste.
- Technical uncertainty — at the start, the team did not know whether it could be done, how to do it, or how it should be designed.
- Process of experimentation — they resolved that uncertainty by trying things: prototypes, alternatives, modelling, testing, iterating.
The test says nothing about laboratories, patents, or novelty to the world. Work that is routine for the industry but genuinely uncertain for this team, resolved by experiment, qualifies. That is why software companies, manufacturers, engineering and architecture firms, hardware startups, biotech, and food producers reformulating a product all end up in scope.
What actually goes in the number
Qualified research expenses come in four categories. Wages for employees performing, directly supervising, or directly supporting qualified research. Supplies consumed in the work — prototype materials, not capital equipment. Sixty-five percent of amounts paid to US contractors doing qualified research on the company’s behalf and at its risk. And cloud computing used for development.
Wages are almost always the largest component, often by an order of magnitude, which is why a serious study starts with the payroll register rather than with a questionnaire. The allocation of each person’s time to qualified activity is the single most examined judgment in the whole file, and it is the one most often done badly.
What it’s worth
On the Alternative Simplified Credit method — the method most companies use, because it doesn’t require reconstructing gross receipts from the 1980s — the federal credit typically lands around 6–10% of qualified research expenses, before any state credit. Measured against a company’s total US engineering spend rather than against QREs alone, the recovery usually falls somewhere in the 10–15% range.
A company with $300,000 of qualifying US engineering salaries is generally looking at a federal credit in the $30,000–$40,000 range. What any particular company gets depends on facts that have to be reviewed.
And it recurs. A company that qualifies this year usually qualifies next year too, which is what makes the credit worth building into a client relationship rather than treating as a one-off filing event.
What doesn’t count
Building the product qualifies. The marketing site does not. Neither does research conducted outside the United States, work funded by a customer who bears the risk, routine data collection, quality control on existing production, market research, or adapting an existing product for one customer without technical uncertainty.
Being honest about the exclusions is not caution for its own sake. An aggressive study is a liability that surfaces three years later, when the people who did the work have left and nobody can explain the number.