The R&D Tax Credit for Startups and Growing Businesses

The research and development tax credit lives in Section 41 of the tax code and has been permanent law since 2015. It is a credit, not a deduction. A deduction reduces the income you are taxed on; a credit reduces the tax itself, dollar for dollar. That distinction is why a mid-five-figure credit is worth chasing even when the paperwork is annoying.

The credit rewards attempting to solve technical problems, not succeeding. A failed architecture that taught you the approach does not work is still qualified research, because the test is about uncertainty and experimentation, not outcomes.

Two audiences read this page. If you are profitable, the credit offsets income tax. If you are a pre-profit startup, the next section is the one that matters: Congress built a way for you to use the credit against payroll taxes years before you owe a dime of income tax.

A credit against income tax is worthless to a company with no income tax. So Section 41(h) lets a qualified small business elect to apply the credit against the employer share of payroll taxes instead. The Inflation Reduction Act doubled the annual cap from $250,000 to $500,000 for tax years beginning after December 31, 2022.

"Qualified small business" has two tests, and both are about gross receipts, not headcount or funding:

• Under $5 million in gross receipts for the year you claim the credit.
• No gross receipts at all for any tax year before the five-year window ending with the credit year. In practice: if your company had its first revenue in 2022 or later, a 2026 claim can still qualify; a company with revenue back in 2020 cannot.

The offset applies first against the employer share of Social Security tax, up to $250,000 per quarter, and any remainder against the employer share of Medicare tax, with unused amounts carrying forward quarter to quarter. The election can be made for a maximum of five tax years, which pairs naturally with the five-year gross receipts window.

You cannot understand founder tax anxiety from 2022 to 2024 without Section 174. For decades, companies deducted research costs the year they spent them. The Tax Cuts and Jobs Act quietly changed that starting in 2022: all research and experimental costs, including software development wages, had to be capitalized and amortized over five years for domestic work and fifteen for foreign work.

The result was brutal arithmetic. A startup that spent $1 million on developers and lost money in cash terms could deduct only a fraction of that spend in year one, so it showed taxable income and owed real tax while burning runway. That is the trap thousands of founders searched their way into.

The One Big Beautiful Bill Act, signed July 4, 2025, fixed the domestic side. New Section 174A restores immediate expensing for domestic research costs, permanently, for tax years beginning after December 31, 2024. Foreign research stays on fifteen-year amortization.

What about the deductions trapped in 2022 through 2024? Two paths existed, and the distinction matters right now:

• The catch-up deduction (still available): any domestic research costs from 2022 to 2024 still sitting unamortized can be deducted in full on the first tax year beginning after December 31, 2024, generally the 2025 return, or spread evenly across 2025 and 2026.
• The retroactive amended-return option (window now closed for most): small businesses with average gross receipts of $31 million or less could elect to apply Section 174A retroactively to 2022 through 2024 by amending those returns, but Rev. Proc. 2025-28 set a deadline of July 6, 2026 for those amended returns. As of this writing, that window has closed for most calendar-year filers, which makes the catch-up deduction on the 2025 return the live path.

If your 2022 to 2024 returns carry unamortized research costs, the catch-up interacts with your burn, your NOLs, and your credit history, and the modeling is worth doing before the 2025 return is finalized. Clean books make that modeling fast; our guide to startup accounting covers how to keep development costs trackable in the first place.

Every activity you claim must pass all four parts of the Section 41(d) test. In plain English:

• Permitted purpose. You are creating or improving a product, process, software, technique, formula, or invention: better function, performance, reliability, or quality. Shipping a new feature counts; rebranding it does not.
• Technological in nature. The work relies on hard science: engineering, physics, biology, or computer science. Software development qualifies on its face. Market research and design taste do not.
• Elimination of uncertainty. At the start, you did not know whether you could build it, how to build it, or what the right design was. If the answer was already on Stack Overflow and you just implemented it, there was no uncertainty.
• Process of experimentation. You systematically evaluated alternatives: prototypes, modeling, simulations, structured trial and error. Branches you built and threw away are evidence in your favor here.

Here is how that plays out against a real engineering backlog:

Once an activity qualifies, four categories of spending attached to it become qualified research expenses, or QREs:

• Wages. W-2 wages for employees performing, directly supervising, or directly supporting qualified research. For a startup this is the big bucket: engineers, plus the technical share of a hands-on CTO's time. If an employee spends at least 80 percent of their time on qualified services, all of their wages count, not just 80 percent.
• Supplies. Tangible property used in the research: prototype materials, test hardware. Not capital equipment, not general office supplies.
• Computer rental, which is where cloud costs live. Amounts paid for the right to use computers in qualified research. Cloud charges for development and testing environments can qualify under this bucket; production hosting that serves customers does not.
• Contract research at 65 percent. If you pay an outside firm or contractor to perform qualified research on your behalf, 65 percent of the payment counts, provided you bear the financial risk and retain rights to the results. Offshore development does not count regardless of contract terms, because research must be performed in the United States.

Notice what the wage bucket implies: the credit is mostly a payroll exercise. If your payroll and project records are clean, the study is cheap. If contractors, founders' sweat equity, and payroll are tangled together, the study gets expensive and the claim gets weaker. This is one of the quiet payoffs of getting bookkeeping right early, which is most of what our CPA for startups engagements fix first.

There are two ways to compute the credit. The regular method pays 20 percent of QREs above a base amount tied to your historical gross receipts and a fixed-base percentage, math that reaches back decades and rarely favors young companies. Almost every startup uses the Alternative Simplified Credit (ASC) instead:

• With a spending history: 14 percent of the amount by which this year's QREs exceed 50 percent of your average QREs for the prior three years.
• Without one: if you had no QREs in any of the three prior years, the credit is a flat 6 percent of this year's QREs. This is where most first-time startup claims land.

The 6 percent years do not last. Once three years of QRE history exist, the 14 percent incremental formula takes over: if Nova later averages $500,000 of QREs over three years and then spends $800,000, the credit is 14 percent of ($800,000 minus $250,000), which is $77,000. Growing research spend is rewarded at the margin.

The credit is computed and claimed on Form 6765, filed with your business income tax return. The form got a major overhaul: new Sections E and F collect summary detail from everyone, and a new Section G requires qualitative reporting on the business components behind the credit. Section G is optional for tax years beginning in 2024 and 2025 and becomes mandatory for tax years beginning in 2026, with two carve-outs that matter here: qualified small businesses electing the payroll offset, and filers with QREs of $1.5 million or less and gross receipts of $50 million or less, can keep skipping it.

The payroll offset election happens on that same Form 6765, and the timing rule is unforgiving: the election generally must be made on a timely filed original return, including extensions. Miss it, and you generally cannot amend your way into the payroll offset for that year. Then Form 8974 attaches to each quarterly Form 941 and actually applies the credit against deposits, starting the first quarter after the quarter in which you filed the income tax return. If a payroll provider runs your 941s, they need to know about the election, because they are the ones who file the 8974.

Documentation standards, in priority order

• Time allocation by person and project. The wage bucket lives or dies on showing who worked on what. Real time tracking beats year-end estimates; even lightweight sprint-level allocation is far better than a memo written in March.
• Project records showing uncertainty and experimentation. Design docs, architecture decision records, git history, tickets, and postmortems that show alternatives considered and rejected.
• Payroll and vendor records that tie out. W-2s, contractor agreements showing risk and IP rights, and cloud invoices separated between development and production use.

The R&D credit has a cottage industry of percentage-fee shops that promise large credits from a one-hour questionnaire. The pattern is consistent: aggressive qualification of routine work, thin or template documentation, a fee taken off the top, and no one standing behind the claim when the IRS asks questions years later. The credit itself is not an audit trigger; an outsized credit with no contemporaneous support is. And it is your company's return, your signature, and your penalty exposure, not the mill's.

Just as honestly, the credit is not always worth pursuing. Skip it, or at least defer it, when:

• The QRE base is tiny. At 6 percent, $50,000 of qualifying wages yields a $3,000 credit. If a proper study costs more than the credit, wait until the engineering payroll justifies it.
• The work is genuinely routine. Agencies configuring platforms, integrations that follow vendor documentation, and maintenance-heavy codebases mostly fail the four-part test, and forcing them through it is how bad claims are born.
• Development is offshore or client-funded. Foreign research does not count, and neither does work where your client pays and keeps the rights. If that describes most of the engineering spend, there is little left to claim.

The middle path is a claim sized to the evidence: qualify the work that clearly passes the test, document as you go, and let the credit grow with the engineering team. That posture survives scrutiny, and it compounds, because the payroll offset can repeat for up to five years. If you want a second opinion on a study someone sold you, or a first opinion on whether your backlog qualifies at all, that is squarely inside what our startup CPA team does.