Summary. The research credit rewards activity that most companies are already doing and do not think of as research: developing a product, improving a process, writing software, or engineering a solution where the outcome was uncertain at the outset. Eligibility turns on a four-part test applied to each business component, and the credit is computed on a narrow set of expenses — wages for qualified services, supplies consumed in the process, and a fraction of contract research payments. The credit is also among the most heavily examined positions on a return, and the difference between a defensible claim and a disallowed one is almost entirely documentation created while the work happened. This guide covers eligibility, the expense categories, the two computation methods, the payroll tax offset available to small companies, the filing and refund claim requirements, and how to survive an examination.

An engineering firm redesigns a fixture to hold tolerance under thermal load. A software company rewrites a data pipeline three times before finding an architecture that scales. A food manufacturer runs eleven formulations to get a shelf-stable product that does not separate.

None of these companies calls that activity research. All three may be conducting qualified research under 26 U.S.C. § 41, and the credit is meaningful — commonly six to eight percent of qualifying spend, dollar for dollar against tax, and for some small companies against payroll tax whether or not there is any income tax to offset.

The credit is also examined aggressively, disallowed frequently, and litigated constantly. The reason is consistent: companies claim it based on a conclusion that the work sounds innovative, and the statute requires a test applied to specific projects with specific documentation.

The four-part test

Section 41(d) requires that research satisfy all four parts, applied component by component rather than to the company as a whole.

1. Section 174 test — permitted purpose and expenditure character

The expenditures must be research or experimental expenditures under § 174, and the research must be undertaken to discover information that is technological in nature and useful in developing a new or improved business component.

A business component under § 41(d)(2) is a product, process, computer software, technique, formula, or invention held for sale, lease, or license, or used in the taxpayer's trade or business. Internal processes count.

The improvement must relate to function, performance, reliability, or quality. Cosmetic changes, style, taste, and seasonal design do not qualify.

2. Technological in nature

The process of experimentation must fundamentally rely on the principles of physical or biological sciences, engineering, or computer science.

This excludes research in the social sciences, arts, humanities, economics, and management. Market research does not qualify. Neither does a new pricing model, however innovative.

The taxpayer need not exceed or refine existing knowledge in the field — the "discovery test" that once applied was eliminated by regulation. It is enough that the taxpayer sought information it did not have, even if others did.

3. Elimination of uncertainty

At the outset, the taxpayer must have faced uncertainty regarding the capability of developing the component, the method of developing it, or the appropriate design.

Design uncertainty is sufficient, and this is the most commonly satisfied prong. A company that knows a thing can be built but does not know how to build it faces qualifying uncertainty.

4. Process of experimentation

The hardest prong and where claims fail. Treas. Reg. § 1.41-4(a)(5) requires a process designed to evaluate one or more alternatives to achieve a result where the capability, method, or design was uncertain — involving identification of the uncertainty, identification of alternatives, and a process of evaluating them, through modeling, simulation, systematic trial and error, or other methods.

Substantially all — eighty percent or more — of the activities must constitute elements of a process of experimentation, under the shrinking-back rule of § 1.41-4(b)(2): if the test fails at the product level, apply it to the next most significant subset of elements, and continue shrinking back until it is satisfied or until the most basic element is reached.

What this means practically. Trial and error qualifies if it is systematic and directed at evaluating alternatives. Building one design and shipping it does not. Documentation must show alternatives considered and evaluated, not merely that the project was difficult.

The exclusions

Section 41(d)(4) removes:

  • Research after commercial production begins.
  • Adaptation of an existing component to a particular customer's requirement. This excludes a great deal of custom engineering and is the most contested exclusion in practice.
  • Duplication of an existing component from a physical examination or from plans.
  • Surveys, studies, market research, and routine data collection.
  • Routine quality control testing.
  • Software developed primarily for internal use, unless it satisfies a high threshold of innovation test: the software is innovative, its development involves significant economic risk, and it is not commercially available for use without substantial modification. The 2016 regulations narrowed what counts as internal use software, excluding software developed to be commercially sold, leased, or licensed, and software enabling third-party interaction — which brought most customer-facing web and mobile applications outside the internal-use rules entirely. This was a significant liberalization and many companies have not revisited their position since.
  • Foreign research conducted outside the United States, Puerto Rico, or a U.S. possession.
  • Social sciences, arts, and humanities.
  • Funded research — research funded by a grant, contract, or another person. Under Treas. Reg. § 1.41-4A(d), research is funded unless the taxpayer retains substantial rights in the results and bears the economic risk — meaning payment is contingent on success. A fixed-price contract where the contractor bears the risk of failure and retains rights is generally not funded; a cost-plus contract generally is. This determination decides whether the contractor or the customer may claim the credit, and both sometimes do, which is a problem.

Qualified research expenses

Only three categories count, under § 41(b), and this is narrower than most companies expect.

Wages for qualified services, § 41(b)(2)(D). The Box 1 wages of employees who are:

  • Engaging in qualified research;
  • Directly supervising it — first-line supervision, not executive oversight; or
  • Directly supporting it — a technician machining a prototype, an assistant recording results.

Not included: general administration, human resources, accounting, legal, marketing, or executives whose involvement is oversight rather than direct supervision.

The substantially all rule, § 1.41-2(d)(2): if eighty percent or more of an employee's time is spent on qualified services, one hundred percent of their wages count. Below eighty percent, only the actual proportion.

Supplies, § 41(b)(2)(C). Tangible property consumed in the conduct of qualified research, other than land and depreciable property. Prototype materials qualify. Capital equipment does not, which surprises companies that spent heavily on test equipment.

Contract research, § 41(b)(3). Sixty-five percent of amounts paid to a third party for qualified research performed on the taxpayer's behalf, where the taxpayer bears the economic risk and retains substantial rights. Increased to seventy-five percent for payments to a qualified research consortium and one hundred percent for certain energy research.

Cloud computing costs for hosting a development environment have been treated as qualifying rental of computer time under § 41(b)(2)(A)(iii) in some circumstances; the analysis is fact-specific and the position should be documented.

Computing the credit

Two methods. Compute both and elect the better one.

The regular credit

Twenty percent of qualified research expenses exceeding a base amount, § 41(a)(1).

The base amount is the fixed-base percentage multiplied by average annual gross receipts for the four preceding years, with a floor of fifty percent of current-year QREs.

The fixed-base percentage is the ratio of QREs to gross receipts for the 1984–1988 period, capped at sixteen percent. Start-up companies — those without QREs and receipts in at least three of those years — use the rules in § 41(c)(3)(B), beginning with a three percent fixed-base percentage for the first five taxable years with QREs and then a computed ratio.

The regular credit produces a larger credit for companies whose research intensity has grown substantially, and it requires historical data most companies no longer have.

The alternative simplified credit

Fourteen percent of QREs exceeding fifty percent of the average QREs for the three preceding taxable years, § 41(c)(5).

Where the taxpayer had no QREs in any of the three preceding years, the ASC is six percent of current-year QREs — a straightforward computation available to any first-time claimant.

The ASC requires only three years of history and is what most companies use. The election is made on Form 6765 and, once made for a year, is irrevocable for that year.

The § 280C reduced credit election

Section 280C(c) requires a taxpayer claiming the credit to reduce its § 174 deduction or capitalized amount by the credit, or to elect a reduced credit — the credit multiplied by one minus the maximum corporate rate — and keep the full deduction.

The election is made on a timely filed return including extensions, on Form 6765. It cannot be made on an amended return, which is a trap for refund claims: a taxpayer claiming the credit retroactively takes the full credit and the corresponding deduction reduction.

The payroll tax offset

The most valuable provision for pre-revenue companies.

A qualified small business — one with gross receipts of less than five million dollars for the taxable year and no gross receipts for any year before the five-year period ending with the current year — may elect under § 41(h) to apply up to a statutory maximum of the research credit against the employer portion of Social Security tax, and, following the Inflation Reduction Act, against the employer portion of Medicare tax as well, with the cap raised to five hundred thousand dollars for taxable years beginning after 2022.

The election is made on Form 6765 with a timely filed return, and the offset is claimed on Form 8974 filed with the employment tax return, beginning in the quarter after the income tax return is filed.

For a startup with substantial engineering payroll and no taxable income, this converts a credit that would otherwise carry forward for years into cash within months. It is regularly missed, and it cannot be elected on an amended return.

Section 174: the change that reframed everything

The Tax Cuts and Jobs Act amended § 174 to require capitalization and amortization of specified research or experimental expenditures for taxable years beginning after 2021 — over five years for domestic research and fifteen years for foreign research — rather than immediate deduction.

The consequences were substantial and largely unintended by taxpayers: companies with heavy research spending and modest margins found themselves with taxable income they did not have cash to pay, because the expenses they incurred could be deducted only twenty percent in the first year (subject to the mid-year convention).

Section 174 and § 41 are related but distinct. Section 174 expenditures are broader than § 41 qualified research expenses — software development costs are treated as § 174 expenditures by statute, and § 174 has no four-part test. A company can have substantial § 174 amortization and no § 41 credit.

Legislative attention to this provision has been continuous since it took effect, with proposals to restore immediate expensing, to apply relief retroactively, or to modify the amortization period. Confirm the current state of § 174 before advising, because the answer has changed and may change again, and a return position taken on last year's understanding may be wrong.

Filing, and the refund claim requirements

Form 6765, filed with the return. The IRS has revised the form to require substantially more detail, including business component information, the qualified expenses by component, and a description of the information sought and the activities performed — phased in over recent filing seasons. Complete it from underlying records rather than from a summary.

Amended returns and refund claims face a heightened standard. Under IRS guidance following Chief Counsel Memorandum 20214101F, a valid research credit refund claim must identify:

  • All business components to which the claim relates;
  • For each component, all research activities performed, all individuals who performed each activity, and all information each individual sought to discover; and
  • The total qualified employee wage expenses, supply expenses, and contract research expenses for the claim year.

The IRS has provided transition relief permitting perfection of a deficient claim within a stated period, and the requirements have been the subject of continuing controversy and litigation. A refund claim that omits this information may be rejected as deficient, and the limitations period may run in the meantime.

Limitations. A refund claim must be filed within three years of filing the return or two years of payment, whichever is later, § 6511. Credits generally may be carried back one year and forward twenty under § 39.

State credits. A majority of states offer their own research credit, several of them refundable, and many define qualified research by reference to § 41 while differing on apportionment, carryforward, and refundability. A company claiming the federal credit should check every state where it has research activity.

Documentation

The credit is won or lost here, and nearly every disallowance traces to records that were not created while the work happened.

What examiners want to see:

  • Project-level records identifying each business component, the uncertainty faced at the outset, the alternatives considered, and how they were evaluated.
  • Contemporaneous technical documentation: design documents, specifications, test plans and results, engineering change orders, lab notebooks, source control history, ticket and issue tracking systems, and meeting notes.
  • Time records allocating employee hours to projects and, within projects, to qualified activities. Time tracking by project is the single most valuable system a claiming company can implement.
  • Payroll records supporting the wage calculation, and an explanation of the substantially-all determination for any employee whose full wages are claimed.
  • Supply records showing consumption in the research rather than in production.
  • Contracts for contract research, showing who bore the risk and who retained rights.
  • A nexus between the claimed expenses and the identified activities. An examiner asked to accept a departmental total without project allocation will disallow it.

What does not work: an estimate prepared by a consultant two years later, based on interviews and a percentage assumption, unsupported by contemporaneous records. This is the fact pattern in a substantial share of disallowed claims, and courts have been unsympathetic — though Cohan-style estimation has occasionally been permitted where the taxpayer established that qualified research occurred and the records were merely imperfect.

Build the system prospectively. A company claiming the credit annually should have: a project register identifying components and uncertainties, time tracking allocated to projects, a technical narrative updated as the work proceeds, and an annual file assembled contemporaneously rather than at filing.

Surviving an examination

The research credit is on the IRS's list of frequently examined positions, and an examination follows a predictable pattern.

The information document requests will ask for the project list, the methodology, the time allocation, the wage detail, and the technical documentation for a sample of components.

The interviews. Examiners interview engineers, not accountants. Prepare them: they should describe the technical uncertainty in their own words, the alternatives considered, and how they evaluated them — and they should not characterize the work as "routine," which engineers do reflexively about work they found difficult.

The sample. Examinations frequently proceed by sampling components and extrapolating. Negotiate the sample and understand that a weak component in the sample is extrapolated across the claim.

The exclusions will be raised: adaptation to customer requirements, funded research, internal use software, and research after commercial production. Have the analysis ready for each.

Section 6662 penalties may be asserted. Reasonable cause based on a professional's substantive advice, and § 6751(b) supervisory approval, are the defenses.

Appeals is the settlement forum, and research credit cases settle there routinely because the issues are factual and the hazards are genuine on both sides.

Practical counsel

Ask every operating client whether they develop or improve products, processes, or software. Most companies that qualify do not know they do, and the ones that most often qualify — manufacturers, engineering firms, software companies, food and beverage producers, and construction firms doing design-build — rarely think of themselves as research companies.

Do the analysis before the money is spent, so the documentation can be built.

Compute both methods, and check the payroll offset eligibility for any company under five million dollars in receipts and within five years of first revenue.

Do not let a contingent-fee provider drive the position. Studies prepared on a percentage of the credit have a structural incentive toward aggressive inclusion, and the taxpayer bears the penalty. Review the methodology, confirm the four-part test was applied component by component, and confirm the documentation exists.

Coordinate with § 174. Every dollar of § 41 QREs is a § 174 expenditure subject to amortization; the credit and the deduction timing must be modeled together.

Watch the elections. The § 280C reduced credit election and the § 41(h) payroll offset election both require a timely filed return and are unavailable on amendment.

Primary authority

  • 26 U.S.C. § 41 — the credit, including § 41(a) (amount), § 41(b) (qualified research expenses), § 41(c) (base amount and the alternative simplified credit), § 41(d) (qualified research and the four-part test), § 41(d)(4) (exclusions), § 41(f) (aggregation and allocation), and § 41(h) (payroll tax credit for qualified small businesses).
  • 26 U.S.C. § 174 — specified research or experimental expenditures and the post-2021 capitalization and amortization requirement.
  • 26 U.S.C. § 280C(c) — the deduction reduction and the reduced credit election; § 38 and § 39 — general business credit ordering and carryovers; § 6511 — refund claim deadlines.
  • Treas. Reg. §§ 1.41-2, 1.41-3, 1.41-4, 1.41-4A, 1.41-6, and 1.41-9 — qualified expenses, base amount, qualified research and the process of experimentation, funded research, controlled groups, and the alternative simplified credit; Treas. Reg. § 1.174-2.
  • T.D. 9786 (2016) — the internal use software regulations and the third-party interaction exclusion.
  • Forms 6765, 8974, 3800, and 8974 instructions — claiming the credit and the payroll offset.
  • Chief Counsel Memorandum 20214101F and subsequent IRS guidance — the specificity requirements for research credit refund claims and transition relief.
  • Union Carbide Corp. v. Commissioner, T.C. Memo. 2009-50, Trinity Industries, Inc. v. United States, 757 F.3d 400 (5th Cir. 2014), Populous Holdings, Inc. v. Commissioner (funded research), and Little Sandy Coal Co. v. Commissioner, 62 F.4th 287 (7th Cir. 2023) — the substantially-all and process-of-experimentation requirements and the burden of substantiation.
  • State research credit statutes, several of which are refundable and most of which define qualified research by reference to § 41.

A worked claim

A forty-person industrial automation company designs custom material-handling systems. Revenue $14 million. Engineering payroll $2.9 million across eighteen engineers and technicians. It has never claimed the credit.

Step one: identify the business components. Not "our engineering department." Each system design, each control architecture, each software module. The company identifies fourteen projects in the year, of which nine involved genuine design uncertainty and five were repeats of prior designs with dimensional changes.

Step two: apply the four-part test to each of the nine.

Three fail on the adaptation exclusion — the customer specified the outcome and the work was fitting a known design to a known space.

Six survive: in each, the engineers did not know at the outset whether a design would hold tolerance under load, whether a control approach would achieve cycle time, or how to sequence a process that had not been done at that scale. Alternatives were modeled, prototyped, and tested, with failures documented in the engineering change records and the ticket system.

Step three: the funded research question. All six were performed under customer contracts. The analysis turns on the specific agreements: three are fixed-price contracts under which the company bore the risk of failure and retained rights to the underlying design approach — not funded, and the credit belongs to the company. Two are cost-plus — funded, excluded. One assigns all intellectual property to the customer — the company did not retain substantial rights, excluded.

This single analysis moves the claim by roughly forty percent, and it is answered by reading contracts, not by estimating.

Step four: the expenses.

Wages. Time records show six engineers spent more than eighty percent of their hours on the three qualifying projects — one hundred percent of their wages count under the substantially-all rule. Seven others spent between fifteen and sixty percent; only the actual proportion counts. Two technicians directly supported the prototyping. The engineering manager's direct first-line supervision qualifies; the VP of Operations' oversight does not.

Qualified wages: approximately $940,000.

Supplies. Prototype steel, sensors consumed in testing, and materials scrapped in trials: $86,000. The test rig purchased for $140,000 is depreciable property and does not qualify, which the company had assumed would.

Contract research. $210,000 paid to an outside controls firm on projects where the company bore the risk — sixty-five percent, or $136,500.

Total QREs: roughly $1,162,500.

Step five: compute. No QREs claimed in the three prior years, so the ASC is six percent of current-year QREs: about $69,750.

Step six: the elections. Gross receipts exceed five million, so the payroll offset is unavailable. The § 280C reduced credit election is modeled against the deduction reduction and elected on the timely filed return.

Step seven: build the file now. Project register, time allocation, technical narratives written by the engineers, and the contract analysis — assembled this year, not reconstructed in an examination three years from now.

Who qualifies, by industry

The credit's reputation as a large-company benefit is wrong, and the industries that most often qualify are frequently the ones least likely to ask.

Manufacturing. Tooling and fixture design, process improvement to reduce scrap or cycle time, new material qualification, automation of a manual operation, and prototype development. Process improvements are business components under § 41(d)(2) and qualify on the same terms as products.

Software. Architecture where scalability, latency, or reliability is uncertain; algorithm development; integration where the approach is not known; and performance optimization. The 2016 regulations removed most customer-facing and third-party-interaction software from the internal use restrictions, which liberalized this area substantially.

Engineering and architecture. Structural and systems design where the approach is uncertain. Note the adaptation exclusion carefully — routine application of known methods to a specific site does not qualify, while developing a novel structural approach does.

Food and beverage. Formulation for shelf stability, texture, allergen removal, and reformulation to remove an ingredient while preserving properties. Recipe development for taste alone does not qualify; developing a process to achieve a property does.

Construction. Design-build contractors developing means and methods, temporary structures, and constructability solutions. The funded research analysis is decisive here and turns on contract type.

Pharmaceutical, biotech, medical device. The paradigm case, and the one where documentation practices are already good because regulatory submissions require them.

Agriculture. Cultivar development, irrigation and soil management techniques, and processing improvements.

Breweries, wineries, and distilleries. Process and formulation development, which qualifies more often than these businesses assume.

Tool and die, machine shops, and job shops. Where each job involves genuine design uncertainty, and where the funded research analysis under fixed-price contracts frequently favors the shop.

Aerospace and defense contractors. Substantial qualifying activity, complicated by funded research analysis under government contracts where cost-plus arrangements predominate.

Where it usually does not apply: professional services, retail, distribution, financial services product design, marketing, and management consulting — because the technological-in-nature prong excludes work grounded in economics, management, or the social sciences however sophisticated it is.

The screening question to ask any client: did your technical people face a problem this year where they did not know at the outset whether or how it could be solved, and did they try more than one approach to find out? An honest yes is worth investigating.

Pass-through entities and owner-level limitations

For an S corporation or a partnership, the credit is computed at the entity level and passes through to owners, where a second set of limitations applies and where the benefit is frequently lost.

The general business credit limitation. Section 38(c) limits the credit an owner may use to their net income tax reduced by the greater of the tentative minimum tax or twenty-five percent of net regular tax liability above twenty-five thousand dollars. An owner with modest tax liability cannot absorb a large credit in the year it arises.

Carryforward and carryback. Section 39 permits a one-year carryback and a twenty-year carryforward of the unused general business credit. The credit is not lost, but its present value declines and a company whose owners cannot use it for a decade should model whether the compliance cost is justified.

Passive activity limitations. Section 469 restricts credits from passive activities to the owner's passive tax liability. An owner who does not materially participate in the business — a common situation for a minority investor — may be unable to use the credit at all until the activity is disposed of.

At-risk and basis limitations apply before the credit limitations and can defer it further.

Allocation among owners. In a partnership, the credit follows the allocation of the underlying expenditures, and the partnership agreement should address it expressly. Special allocations of the credit must have substantial economic effect or be consistent with the partners' interests.

Section 41(f) aggregation. All members of a controlled group or of a group under common control are treated as a single taxpayer for computing the credit, with the group credit then allocated among members based on their proportionate share of QREs. This catches structures with an operating company and a separate research entity, and it must be run before the computation rather than after.

The § 280C election is made at the entity level and affects the deduction flowing to all owners.

The payroll offset is available to a qualified small business including a partnership or S corporation, applied against the entity's own employer payroll taxes — which for a pre-revenue startup with engineering payroll is frequently the only way the credit produces any benefit at all.

Practical counsel. Before investing in a study, model the owner-level result. A $90,000 credit that no owner can use for eight years is worth substantially less than the study costs, and the honest advice may be to build the documentation now and claim in a year when the income arrives.

Choosing and managing a provider

Most research credit claims are prepared by specialist firms rather than by the company's regular accountant, and the selection matters more than the fee.

Fee structures. Contingent fees — a percentage of the credit — are common and create an obvious incentive toward inclusion. Circular 230 § 10.27 restricts contingent fees in tax practice, permitting them for services rendered in connection with an examination or a claim for refund filed within 120 days of receiving a notice of examination, but not generally for preparing an original return position. Confirm the arrangement complies, and understand that the penalty exposure sits with the taxpayer regardless of who prepared the study.

Questions to ask a provider:

  • Will you apply the four-part test component by component, or to the department as a whole?
  • How will you handle the funded research analysis, and will you read the contracts?
  • What documentation will exist at the end, and will it be contemporaneous or reconstructed?
  • Will engineers be interviewed and quoted, or will percentages be estimated?
  • What is your position on the adaptation exclusion for our contract work?
  • Will you defend the claim in an examination, at what cost, and is that in writing?
  • What is your disallowance history, and on what issues?

The red flags. A provider who quotes a credit amount before reviewing anything. A methodology based on a survey asking managers what percentage of time was "innovative." Inclusion of every engineer at a uniform percentage. Refusal to exclude any project. A study delivered without underlying workpapers. And an assurance that the credit is "not audited much," which is untrue.

What the taxpayer owes the process. Access to engineers, honest answers about what was uncertain and what was routine, the contracts, and time records. A study is only as good as the company's willingness to say that five of fourteen projects do not qualify.

Coordinate with the return preparer. The credit interacts with § 174 amortization, the § 280C election, the general business credit limitation, and state filings. A study delivered to a preparer two days before the deadline produces errors in all four.

And keep the workpapers. They are the taxpayer's, not the provider's, and a company that changes providers without obtaining the prior years' files has lost the basis for its own carryforwards.

Related articles

This guide is provided for general informational purposes and does not constitute legal or tax advice. Section 174 capitalization and the research credit's refund claim requirements have been the subject of continuing legislative and administrative change; confirm the current rules before taking a position. The payroll tax offset and the § 280C reduced credit election require a timely filed return and cannot be made on an amendment. Consult qualified tax counsel and an adviser experienced in research credit substantiation before claiming the credit.