If you run a venture-backed SaaS company in San Jose or Palo Alto, are paying eight or ten engineers $200K each to write code, and have never claimed the federal R&D tax credit, you are leaving real money on the table. A pre-revenue startup with $2 million in qualifying engineering salaries can typically generate $130,000 to $200,000 in federal R&D credit. Because you are pre-profit, that credit does nothing against income tax. But under the §41(h) payroll tax election, that same credit can offset up to $500,000 per year of your employer Social Security and Medicare tax, starting the quarter after you file.
The credit has been around since 1981. Two things changed recently that make it more important to understand: the §174 capitalization rule that the Tax Cuts and Jobs Act layered in starting in 2022 (which gutted startup losses for three years), and the One Big Beautiful Bill Act (OBBBA, Public Law 119-21, signed July 4, 2025), which partially unwound it for tax years beginning after December 31, 2024. If you have not revisited your R&D position since 2024, you are likely missing both retroactive recoveries and ongoing benefit.
Important: The §174A immediate-expensing provisions and §174 retroactive amended-return mechanism described below are now in force under OBBBA (Pub. L. 119-21). Treasury and IRS guidance is still developing on procedural mechanics for the 2022 through 2024 amended-return window; coordinate with your CPA before filing.
This guide walks through the §41 credit itself, the four-part qualification test as applied to software, the §174 capitalization rule and its 2025 rollback, the payroll offset mechanics, and the documentation an IRS exam will actually want to see. We work through this fact pattern constantly with our startup founder team at Silicon Valley Tax.
IRC Section 41 gives you a federal income tax credit equal to a percentage of your qualified research expenses (QREs) above a base amount. Under the Alternative Simplified Credit (ASC) method, a first-time filer with no prior-year base receives 6 percent of total QREs. As the company builds a three-year base history, the credit rises toward the statutory 14 percent of incremental QREs above 50 percent of the prior-three-year average. Most software startups land in the 6 to 10 percent range of total QREs through their first few years.
The credit is permanent (locked in by the PATH Act of 2015) and dollar-for-dollar against tax liability. For a profitable company, every $1 of credit reduces federal tax by $1. For a pre-profit startup, the credit either carries forward up to 20 years or, if you qualify as a Qualified Small Business (QSB), gets monetized against payroll tax.
The Inflation Reduction Act of 2022 amended §41(h) to double the payroll-tax offset cap from $250,000 to $500,000 per year, effective for tax years beginning after December 31, 2022. The first $250,000 still offsets the employer 6.2% Social Security portion; the second $250,000 offsets the employer 1.45% Medicare portion. Many startups still file as if the cap were $250K. It is not.
Every dollar of QRE has to map to a "business component" (a product, process, software, technique, formula, or invention) that passes a four-part test. Apply the test once per business component, not once per company. A typical SaaS startup will have several business components: the core platform, each major new feature, the internal data pipeline, the ML model, and so on.
The research must be intended to develop a new or improved business component for purposes of function, performance, reliability, or quality. Cost reduction alone counts. Aesthetic changes (rebrand, color palette, UI polish that is not performance-driven) do not.
The work must fundamentally rely on the hard sciences: computer science, engineering, biology, chemistry, physics. Software engineering qualifies. UI/UX design as such does not, although the back-end engineering supporting a new UI usually does.
You must be evaluating one or more alternatives through systematic trial and error, modeling, simulation, or other methods. "We built the obvious solution and shipped it" fails. "We benchmarked three database engines, ran load tests, and iterated on the data model until p99 latency dropped below 100ms" passes.
At the outset of the work, there must be uncertainty about capability (can this be built?), method (what approach will work?), or design (which design will achieve the requirements?). Stack already proven and documented in vendor SDK examples? Probably no uncertainty. Novel ML architecture, new distributed systems pattern, or first-of-its-kind integration? Real uncertainty.
You do not need every project to be groundbreaking. Most qualifying software work is incremental: shipping a new feature that requires new architecture, optimizing a slow query path through systematic profiling, building a new ML pipeline. The bar is "uncertainty at the outset that you resolved through experimentation," not "patent-grade novelty."
Two software-specific rules trip up startups regularly.
Software developed primarily for the taxpayer's own internal use (back-office systems, internal data warehouses, employee-only tools) has to clear a higher three-part test on top of the §41(d) four-part test: it must be innovative, involve significant economic risk, and not be commercially available. This is a meaningful hurdle. A startup building its own ERP-like internal admin tool will struggle to claim those engineering hours.
The good news for most SaaS founders: customer-facing software (your product) is explicitly not internal use software. Dual-function software (used both internally and by customers) is split by a safe harbor under Treasury Reg §1.41-4(c)(6)(vi): if at least 10% of the use is by third parties (customers), the third-party portion escapes the IUS restriction. For a normal SaaS company shipping a product to paying customers, IUS is rarely a blocker.
Research that is "funded" by a third party (a customer contract under which they pay you to perform research and they retain rights) generally does not qualify under §41(d)(4)(H). Two prongs: payment is not contingent on success (you get paid even if it fails) AND you do not retain substantial rights in the result. If both prongs are true, the work is funded and disqualified.
Typical SaaS pattern: ARR subscription customers who pay for product access do not "fund" your research because their payment is contingent on the product actually working, and you retain all IP. That is unfunded and qualifies. Custom dev shop contracts with milestone payments and a full IP assignment to the client? Funded, generally disqualified.
Three categories of cost roll into QREs:
Engineer wages typically make up 85 to 95 percent of total QRE for a software startup. AWS bills used for ML training can be a meaningful fourth bucket for AI companies.
For tax years beginning in 2022 through 2024, the Tax Cuts and Jobs Act required all R&D expenditures to be capitalized and amortized rather than deducted in the year incurred. Under the new IRC §174, domestic R&D had to be capitalized over 5 years (half-year convention, so you got 10% in year one) and foreign R&D over 15 years.
For startups, this was financially catastrophic. A company spending $2 million on engineering salaries used to deduct the full $2M, producing a $2M net operating loss that carried forward to offset future income. Under the post-2022 §174 rule, that same $2M generated only $200K of year-one deduction, leaving $1.8M of "phantom income" still on the books even though every dollar had been spent on payroll. Profitable-on-paper, cash-poor-in-reality startups had to pay tax on income they had already spent.
The Build Back Better Act, the EARN Act, the Tax Relief for American Families and Workers Act, and several other 2023 and 2024 bills proposed restoring immediate expensing. None made it into law in time to help the 2022, 2023, or 2024 tax years.
The One Big Beautiful Bill Act of 2025 finally addressed §174 for tax years beginning after December 31, 2024. The headline change:
If your startup paid federal income tax in 2022, 2023, or 2024 because of §174 capitalization, you very likely have a refund coming via amended returns. Run the §448(c) gross-receipts test (under $31 million average for the three prior years for 2025) and, if eligible, file the §174 retroactive election. We have seen six and seven-figure refunds on this.
Sitting on three years of phantom §174 income? We run the eligibility test and the amended returns for Bay Area software startups regularly. Schedule a free consultation and we will quantify what is recoverable in your specific case.
The payroll-tax offset is the headline benefit for pre-profit startups. Eligibility:
The mechanics, year by year:
A startup with $2M in QREs and a Section G-supported $200K credit can offset roughly $200K in employer payroll tax over the next four quarters, which on a 20-engineer team is real money against burn.
The R&D credit is one of the most-audited federal credits. In Siemer Milling Co. v. Commissioner, T.C. Memo. 2019-37, the Tax Court denied a substantial portion of claimed R&D credits because the taxpayer could not tie its claimed activities to specific business components, could not show the four-part test had been applied per component, and produced documentation that was largely reconstructed after the fact. The same pattern repeats across most R&D credit denials. Do not file an R&D claim without these:
The credit is worth claiming, but only if the documentation will survive a desk audit. Filing a paper-thin Form 6765 with no project documentation is worse than not filing at all because it puts your return into the audit pool with nothing to defend.
R&D credit planning rarely lives alone. It interacts with your entity structure (C-corp vs. LLC vs. S-corp), QSBS planning, and how you treat founder compensation. A few overlap points worth knowing:
The most common pattern we see at intake: a startup that ran a self-prepared "R&D credit calculator," took 8 to 10 percent of total engineering payroll as QRE, and filed a Form 6765 with no project-level documentation. The 2024 Form 6765 redesign forces every filer to disclose business components in Section G. A skinny disclosure paired with a generic percentage attestation puts the return into the audit pool, and the typical outcome is full disallowance plus a 20 percent accuracy-related penalty. The bill for a $200K disallowed credit including penalty and interest commonly tops $260K.
DIY filing software does not run a study, does not isolate IUS exposure on internal admin tools, does not flag funded-research contracts that disqualify portions of payroll, and does not file the §174 retroactive amended returns. Each of those items can swing the result six figures in either direction.
Almost certainly yes if you are building real software with engineers in the United States. The credit does not require revenue. The most common reason a pre-revenue startup does not claim it is that nobody told them about the payroll offset. If you have W-2 engineers writing code that meets the four-part test, you qualify. The credit is calculated on your QREs regardless of whether you have any income.
Most software startups using the Alternative Simplified Credit get 6 to 10 percent of QREs in year one. On $2M of engineer salaries, that is roughly $120K to $200K of credit. You can convert up to $500K of credit per year to a payroll-tax offset as long as you qualify as a QSB.
Open tax years (generally the last three) can be amended to claim previously-missed credit. You can also claim retroactive §174 expensing for 2022 through 2024 if you meet the small-business gross-receipts test. We have seen six-figure refunds come back from this combination.
For the §41 credit itself, no. Only US-based research (W-2 employees and 65% of US contractor pay) generates QRE. For the §174 expensing rule under OBBBA 2025, foreign R&D is still capitalized over 15 years. If most of your engineering is offshore, the credit is small and the §174 capitalization continues to bite. Onshoring the engineering team has real tax consequences either way.
It does not automatically trigger an exam, but the R&D credit is on the IRS's frequent-issue list. The 2024 Form 6765 redesign (Section G business-component reporting) increased disclosure substantially. If your claim is well-documented and methodologically sound, audits are defensible. If your claim is a percentage applied to total payroll with no project-level support, expect to lose at exam.
No. The credit and the deduction are separate mechanisms. For tax years after 2024, you can both expense your domestic R&D under §174A and claim the §41 credit on the qualifying portion. You do have to reduce your §174 deduction by the credit amount under §280C, or alternatively make a reduced-credit election under §280C(c). Most startups elect the reduced credit to keep deductions clean.
R&D credit work is one of the highest-ROI tax projects available to a software startup. Done right, a pre-revenue company can recover six figures of payroll tax per year and a profitable company can shave double digits off the federal effective rate. Done wrong, it puts the return in the audit pool with no defense.
At Silicon Valley Tax, we run R&D credit studies for Bay Area software startups across the funding spectrum, including the §174 retroactive amended returns for the 2022 through 2024 capitalization period. If you have never claimed the credit, have claimed it without a study, or have not revisited your position since OBBBA 2025 passed, the marginal review is almost always worth it. See our broader services list for adjacent founder work.
Schedule a free consultation and we will walk through your engineering spend, your QSB eligibility for the payroll offset, and whether the retroactive §174 amended returns are worth filing for your situation.
The §41 R&D credit plus the §41(h) payroll offset can return six figures a year against your burn. The §174 retroactive election can recover prior-year tax. Talk to us before you file.