The IRS Four-Part Test Under IRC Section 41
The federal R&D tax credit is governed by IRC Section 41, and the IRS uses a four-part test to determine whether an activity qualifies as "qualified research." First, the activity must have a permitted purpose, meaning it aims to create a new or improved business component such as a product, process, software, technique, formula, or invention. Second, the activity must involve technological uncertainty, where the capability, method, or design is not known at the outset. Third, the taxpayer must undertake a process of experimentation, including evaluating alternatives through modeling, simulation, systematic trial and error, or other methods. Fourth, the research must rely on principles of engineering, physical sciences, biological sciences, or computer science.
All four prongs must be satisfied simultaneously. The IRS has become increasingly sophisticated in auditing R&D credit claims, particularly for software companies. Generic statements like "we built new features" will not survive scrutiny. You need to document the specific technical uncertainty you faced, the alternatives you evaluated, and how you resolved the uncertainty through experimentation.
Qualified Research Expenses for Software Startups
Under IRC Section 41(b), qualified research expenses (QREs) fall into four categories. The largest for most startups is W-2 wages paid to employees who directly perform, supervise, or support qualified research. This includes software engineers, data scientists, QA engineers involved in testing experimental functionality, and first-line engineering managers who directly supervise that work. If 80% or more of an employee's services for the year are qualified services, the regulations let you treat all of that employee's wages as QREs (the "substantially all" rule).
Supply costs under Section 41(b)(2) cover tangible items consumed during research, though for software companies this category is typically small. Contract research expenses under Section 41(b)(3) include payments to third-party contractors who perform qualified research on your behalf, but only 65% of those payments count as QREs. Amounts paid for the right to use computers in qualified research, such as cloud costs for development and testing environments (not production hosting), can also count under §41(b)(2)(A)(iii).
A common mistake is including all engineering salaries without distinguishing between qualified and non-qualified activities. Bug fixes for known issues, routine data entry, quality control testing for production readiness, and management of production systems do not qualify. You must allocate each engineer's time between qualifying and non-qualifying activities.
Activities That Do Not Qualify
Section 41(d)(4) lists several exclusions that catch startups off guard. Research conducted after commercial production begins does not qualify, so once your product is live, only improvements involving new technical uncertainty count. Adaptation of existing technology to a specific customer's needs typically fails the technological uncertainty test. Internal-use software faces a higher bar under the "high threshold of innovation" test from IRS regulations, requiring that the software be innovative, that the development involve significant economic risk, and that the software not be commercially available for purchase.
Foreign research conducted outside the United States is excluded entirely. Research in the social sciences, arts, or humanities does not qualify. Funded research, where another party bears the financial risk and you are essentially a contractor, is also excluded. Finally, surveys, studies, and market research never qualify regardless of how technically complex the methodology might be.
The internal-use software exclusion under Treasury Regulation 1.41-4(c)(6) is particularly important for SaaS startups. If your software is used primarily to manage your own business operations rather than sold or licensed to customers, you face the three-part high threshold of innovation test. However, most SaaS products that are licensed to external customers are not classified as internal-use software.
Applying the four-part test to a sample project
The quickest way to test a project is to answer each prong in one or two sentences, using facts from the time the work started. A hypothetical example for a logistics software startup building a route optimization engine:
- Permitted purpose: the engine is a new business component meant to improve function (shorter delivery routes and faster calculation), not appearance.
- Technological in nature: the work relies on computer science, specifically combinatorial optimization and distributed computing.
- Technical uncertainty: at the start, the team did not know whether it could return near-optimal routes for 5,000 stops in under two seconds, or which algorithm design would get there.
- Process of experimentation: the engineers built and benchmarked three approaches (a heuristic solver, an integer programming model, and a hybrid), measured each against the target, and iterated on the hybrid.
Under the regulations, substantially all of the activities, meaning 80% or more measured on a cost or other consistently applied basis, must be elements of a process of experimentation for the component to qualify. Building the admin dashboard that displays the routes would usually fail that test and should be excluded.
The shrinking-back rule when a whole product does not qualify
If an entire product does not meet the four-part test, you do not lose everything: Treasury Regulation §1.41-4(b)(2) applies the test to the next most significant subset of elements, and keeps shrinking until it reaches a subset that qualifies or runs out of elements.
For a startup, the shrink-back is usually the difference between a defensible claim and an overreach. A mobile app as a whole may be mostly routine, but its offline sync module or its on-device image model may qualify on its own. Define business components at the level where the uncertainty actually sat, and allocate time to those components rather than to the product as a whole.
Documentation Requirements and Audit Defense
The IRS expects contemporaneous documentation, meaning records created during or near the time the research was performed, not reconstructed years later during an audit. At minimum, you should maintain project descriptions that identify the technical uncertainty, time tracking records showing hours spent on qualified vs. non-qualified activities, technical design documents and architecture reviews, Git commit histories and pull request descriptions that reference experimental approaches, and financial records tying wages and expenses to specific research projects.
Courts have accepted reasonable estimates of qualified time in some cases, but the burden of proof is on the taxpayer, and estimates built after the fact are where claims most often fail. We recommend implementing a lightweight time tracking system that captures at least weekly estimates of time spent on R&D-qualifying activities versus other work.
How SpryTax Evaluates Your R&D Credit Eligibility
Our process begins with a technical interview covering your product roadmap, engineering processes, and development methodology. We apply the four-part test to each project and activity, identify which team members are performing qualified research, and estimate the credit using both the regular credit method (20% of QREs exceeding a base amount) and the alternative simplified credit method (14% of QREs exceeding 50% of the average QREs for the prior three years). If the company had no QREs in any one of the three prior years, the ASC is a flat 6% of current QREs.
A labelled example: a startup spending $800,000 a year on engineering salaries where 60% of the work qualifies has $480,000 of wage QREs. With no QRE history, the ASC credit is $480,000 x 6% = $28,800. If it is a qualified small business (gross receipts under $5 million and no receipts before the last five years), it can apply that credit against employer payroll taxes under §41(h) instead of waiting for taxable income.
Since 2025, domestic research costs are also deductible right away under §174A, and the §280C(c) rule reduces that deduction by the credit unless you elect a reduced credit. This is general information as of October 2026, not tax advice for your specific facts.
Frequently asked questions
How does a startup qualify for the R&D tax credit?
A startup qualifies when its work passes the four-part test in IRC §41(d): the work aims to create or improve a business component, relies on engineering or hard science, faces technical uncertainty at the outset, and resolves it through a process of experimentation. It then claims the credit on qualified research expenses such as wages, supplies, cloud computing for research and 65% of US contract research.
What is the four-part test for the R&D credit?
The four-part test requires that research has a permitted purpose (new or improved function, performance, reliability or quality), is technological in nature (relies on physical or biological sciences, engineering or computer science), addresses uncertainty about capability, method or design, and involves a process of experimentation for substantially all of the activities. All four must be met for each business component.
Do pre-revenue startups qualify for the R&D tax credit?
Yes. Revenue is not a requirement. A pre-revenue startup can claim the credit on its qualified research expenses and, if it has gross receipts under $5 million and no receipts before the last five years, elect to apply up to $500,000 a year against employer payroll taxes. Without that election, the credit carries forward up to 20 years.
Does internal-use software qualify for the R&D credit?
Internal-use software can qualify, but it must also pass the high threshold of innovation test in Treas. Reg. §1.41-4(c)(6): the software must be innovative, its development must involve significant economic risk, and it must not be commercially available. Software sold, licensed or offered to customers, including most SaaS products, is not internal-use software and does not face this extra test.
What activities are excluded from the R&D credit?
Section 41(d)(4) excludes research after commercial production begins, adapting an existing product for a particular customer, duplicating an existing product, surveys and market research, social sciences and humanities, research performed outside the United States, and research funded by another party. Routine bug fixes, cosmetic changes and production support also fail the four-part test.


