R&D Tax Credits and Section 174 for Startups

There is a federal programme that hands cash to unprofitable startups for work they were doing anyway, and a large share of eligible companies never claim it.

There is also a related rule that, for several years, generated taxable income for companies with no profit — producing surprise tax bills for startups that had lost money all year.

Both concern the same activity: research and development spending. This guide covers the credit, the capitalisation saga and where it landed, what actually qualifies, and how to document a claim that survives examination.

1. The Payroll Tax Offset: Cash for Unprofitable Companies

The federal research credit under Section 41 has existed for decades, but historically it only reduced income tax — which is useless to a company with no taxable income. Credits accumulated as carryforwards nobody could use.

That changed with a provision allowing a qualified small business to elect to apply the research credit against payroll taxes instead. The effect is genuine cash: your quarterly payroll tax liability is reduced, so real money stays in the business.

The broad shape of the rules:

• Qualified small business generally means gross receipts below a threshold in the current year, and no gross receipts more than five years before the current year. That second condition is what makes it a startup provision — a long-established company with small revenue does not qualify.
• The offset applies against the employer portion of Social Security tax, and following changes made in 2022 legislation, also against the employer share of Medicare tax — which roughly doubled the maximum annual benefit.
• The election is made on the return and then claimed on payroll filings, so the cash arrives quarterly rather than as a single refund.
• There is a limit on how many years a company can use the election.

The specific dollar caps and thresholds are adjusted over time. Confirm the current figures with your accountant rather than relying on any published number, including here.

The practical point stands regardless of the exact cap: a pre-revenue company employing engineers in the United States is very likely leaving money on the table if it has never looked at this.

2. What Actually Qualifies

Qualifying activity is tested against four criteria, all of which must be met:

1. Permitted purpose. The work is intended to create or improve a product, process, software, technique, formula or invention — improving function, performance, reliability or quality.
2. Technological in nature. It relies on principles of physical or biological science, engineering or computer science.
3. Elimination of uncertainty. At the outset there was uncertainty about capability, method, or appropriate design.
4. Process of experimentation. Substantially all the activity involved evaluating alternatives — modelling, simulation, systematic trial and error.

Founders frequently assume this means laboratories and patents. It does not. Ordinary software engineering routinely qualifies — designing a new architecture, building a novel algorithm, solving a performance problem with an uncertain solution, integrating systems in a way that required experimentation.

Qualified research expenses generally include wages for people performing, supervising or directly supporting qualified research; supplies consumed in the process; a portion of contract research paid to third parties; and certain cloud computing costs used for development.

What does not qualify: research conducted outside the United States, routine data collection, market research, quality control testing, adaptation of an existing product for a particular customer, work funded by another party where you bear no financial risk and retain no rights, and research after commercial production begins.

That last exclusion — funded research — matters for anyone taking government grants. Work funded by an SBIR award may not be creditable, depending on the risk and rights analysis. It is not automatic in either direction, and it needs specific advice.

3. The Section 174 Saga

This is the part that caused real damage, and it is worth understanding even now that the position has improved, because it affects prior years.

Before 2022: research and experimental expenditures could be deducted immediately in the year incurred. Straightforward.

From 2022: a change enacted in the 2017 tax act took effect, requiring research and experimental expenditures to be capitalised and amortised rather than deducted — over five years for domestic research and fifteen for foreign.

The consequence was counterintuitive and painful. A startup that spent, say, $4 million on engineering and lost money overall could no longer deduct that $4 million in the year it was spent. Only a fraction was deductible, so the company showed taxable income it did not economically have, and owed cash tax while burning cash. Companies with no profit received tax bills.

From 2025: legislation enacted in July 2025 restored immediate deduction for domestic research expenditures, effective for tax years beginning after 31 December 2024, under a new provision. Foreign research remains subject to long amortisation, which is a real consideration for companies with offshore engineering.

The legislation also provided routes to deal with the 2022 to 2024 period — including an ability for smaller companies to apply the change retroactively to those years, and mechanisms for others to accelerate the remaining unamortised domestic balance over a short period.

What to do about it: if your company capitalised research costs in 2022, 2023 or 2024, ask your accountant explicitly whether you are eligible for retroactive relief or acceleration, and whether amended returns are worth filing. This is a live question with real cash attached, and it is not something a general preparer will necessarily raise unprompted.

4. Credit Versus Deduction — They Are Not the Same Thing

These two are constantly conflated. They are separate provisions doing different jobs:

• Section 174 (and its successor) governs when you deduct research spending. A timing question.
• Section 41 is the research credit — a dollar-for-dollar reduction in tax, and for qualified small businesses, a payroll tax offset that produces cash.

The definitions of qualifying activity overlap but are not identical, and the interaction affects your deduction. You can claim both, and you should understand which one someone is talking about.

5. Documentation: Where Claims Are Won and Lost

The credit is not difficult to claim. It is difficult to defend if you did not document contemporaneously.

What examiners look for:

• Project-level records. What was the technical uncertainty at the outset? What alternatives were evaluated? What was the process of experimentation? A brief written statement per project, written at the time, is worth enormously more than a reconstruction.
• Time allocation by person by project. Which engineers spent what proportion of their time on which qualifying work. This is the single most contested element, and reconstructing it two years later from memory is where claims collapse.
• The nexus between wages and activity. Payroll data tied to the allocation.
• Supporting artefacts. Design documents, commit history, test results, issue trackers, meeting notes. Engineering teams generate this naturally — the failure is not capturing it in a form that maps to projects.

Federal reporting requirements for the credit have been tightened in recent years, with more project-level detail required on the return itself. That makes contemporaneous records more important, not less.

The practical recommendation: set up a lightweight quarterly process. A short memo per qualifying project, and a time allocation captured while people remember. It takes an hour a quarter and it is the difference between a claim that holds and one that does not.

6. Choosing a Provider

Specialist R&D credit firms exist alongside general accounting firms. Both can work; the questions are the same:

• How do you charge? Contingent fees as a percentage of the credit are common in this market. They align incentives toward larger claims, which is not always toward defensible ones. Fixed fees remove that pressure.
• What documentation do you produce? A defensible claim comes with a study you could hand to an examiner. Ask to see a redacted sample.
• What happens on examination? Is support included, and at what cost?
• Who does the technical interviews? Someone who can talk to your engineers about the actual work will produce a better and more defensible claim than someone working from a questionnaire.

Be wary of anyone promising a very large credit before examining your work. Aggressive claiming in this area has attracted enforcement attention, and an over-claimed credit with penalties is worse than a smaller claim you can stand behind.

7. State Programmes

Many states operate their own research credits, and several are more generous than the federal one — some are refundable, meaning cash regardless of tax position. Rules, definitions and application processes vary substantially by state.

If you have engineering staff in a state with a strong programme, this is worth examining alongside the federal claim. It is frequently overlooked because the federal credit dominates the conversation.

8. Timing: When the Cash Actually Arrives

Founders who hear "payroll tax offset" often model it as a rebate landing shortly after the spending. It does not work that way, and the lag matters if you are building a runway model around it.

1. You incur the qualifying expense through the tax year — engineering wages, mostly, month by month.
2. The credit is computed and the election made on the income tax return for that year, filed after year end. The election generally must be made on a timely filed return including extensions — a late return can forfeit the payroll offset even where the underlying credit survives.
3. The offset is then claimed on payroll filings, beginning with the first quarter that starts after the return is filed — not the quarters the return covers.
4. Cash accrues quarter by quarter as payroll tax that would otherwise have been remitted simply stays in the business, until the credit is exhausted.

The practical consequence: spending in one year begins converting to cash roughly a year later, and then takes several further quarters to be fully absorbed. That is still excellent money — non-dilutive, for work you already did — but it is not bridge financing and should never be modelled as though it arrives on demand.

Two timing traps worth naming:

• Extending the return delays the cash. If the offset is material to your runway, filing early rather than on extension moves the first offset quarter forward by months.
• A small payroll caps the absorption rate. The offset is limited by the payroll tax you actually owe, so a company with a large credit and a modest US headcount cannot use it quickly. The unused portion carries forward, but the cash trickles.

9. Why Eligible Startups Still Do Not Claim

The same handful of reasons come up repeatedly, and each is worth testing against your own assumptions:

• "We're pre-revenue, so there's no tax benefit." The most expensive misconception in this article. The payroll offset exists precisely for companies with no taxable income.
• "Our work isn't really research." The statutory test is technical uncertainty resolved through experimentation — not novelty to the world, and certainly not a patent. Solving a hard distributed-systems problem qualifies.
• "Our accountant would have raised it." Many perfectly competent general preparers do not run research credit studies and do not mention them. Ask directly rather than reading silence as ineligibility.
• "We'll do it once we're bigger." The qualified small business definition has a five-year gross receipts test built into it. Waiting can move you out of eligibility for the payroll offset permanently.
• "The fee eats the benefit." Sometimes true for a very small claim in year one. It is rarely true across three or four years of engineering payroll, and the documentation habit costs almost nothing once established.

Frequently Asked Questions

Can we claim for prior years?

Generally yes, by amending returns within the applicable statute of limitations, though the payroll tax offset election has its own timing rules that are less forgiving than the income tax credit. If you have never claimed, ask specifically about prior years — there is often meaningful money there.

Do founder and executive wages count?

Yes, to the extent the person genuinely performs, directly supervises or directly supports qualified research — and only for that proportion of their time. A technical co-founder writing code has a strong case; the same person's hours on fundraising, hiring and board management do not. The allocation has to be honest and supportable, because founder wages are frequently a large share of early-stage qualified expenses and therefore a natural place for an examiner to begin.

Does equity compensation count toward qualified expenses?

Wages for this purpose generally means taxable compensation reported on Form W-2, so the treatment of equity depends on whether and when it produces W-2 wages rather than on the accounting expense you book. Companies paying engineers heavily in equity sometimes find their creditable wage base is far smaller than their reported R&D line. Ask your adviser how your specific plan interacts before assuming the two figures match.

Does taking the credit increase audit risk?

A well-documented claim within normal parameters is routine. Aggressive claims — large credits relative to payroll, weak documentation, or promoters pushing marginal positions — attract attention. Claim what you can support.

Does offshore engineering qualify?

No for the federal credit, which requires research performed within the United States. And under the current capitalisation rules, foreign research is still subject to long amortisation while domestic research is immediately deductible. This is a genuine and often overlooked cost input into where you locate engineering.

What if we are funded by grants?

Work that is funded research — where you bear no financial risk and do not retain substantial rights — is excluded from the credit. Grant-funded work needs a specific analysis of the contract terms; the answer varies by award structure and is not automatic either way.

How does this look at an exit?

Unused credits are an asset a buyer will diligence and may pay for; an over-claimed credit with thin support is a contingent liability that gets indemnified, escrowed or deducted from the price — see our guide to selling your startup. Neither position affects QSBS eligibility, which turns on the company's assets and activities rather than on its credit history.

Do investors care about this?

Yes, in two ways. It extends runway at no dilution, which every investor likes. And it appears in diligence — a company that has claimed properly looks well run; one that over-claimed with thin documentation creates a contingent liability a buyer will price.

The Bottom Line

If you employ engineers in the United States and have not looked at the research credit, look now — the payroll tax offset turns it into cash even with no profit.

If you capitalised research costs in 2022 to 2024, ask your accountant directly about retroactive relief, because that is real money sitting in prior returns.

And start documenting contemporaneously. That single habit is what separates a claim you can defend from one you merely filed.

Global Capital Network connects founders with investors and with the accounting and advisory firms that support them. See upcoming events or get in touch.

This article is general information, not tax advice. Section 174 and Section 41 rules changed materially in 2025 and thresholds are periodically adjusted. Work with a qualified tax adviser on your own position.