Does software development qualify for the R&D tax credit, and what records prove it?
Software development can qualify for the federal research credit under Internal Revenue Code section 41 when the work aims to improve a product's function or performance, relies on computer science, starts with technical uncertainty and resolves it through experimentation. Internal-use software faces a stricter test. Tickets, pull requests, design documents and commit histories are the usual evidence.
Does software development qualify for the R&D credit?
Often, for part of the work. The federal research credit in Internal Revenue Code section 41 is tested separately for each business component (a product, process, piece of software, technique, formula or invention), not for a company or a team as a whole. A team building a new matching engine, re-architecting for performance it was not sure it could reach, or solving an integration problem with no known method may have qualifying work. The same team's routine bug fixes, cosmetic changes and customer-specific configuration usually do not.
What decides most claims is evidence. Advisers preparing credit studies rely on the records engineers create while they work: tickets that state the problem, pull requests where reviewers argue about the approach, commit histories that show what was tried and reverted. Those are also the engineering histories AI labs and data buyers look for, which makes the credit study a good vantage point for noticing clients with deep records.
How the four-part test maps onto software work
Section 41 and the Treasury regulations ask four questions of each business component. In summary form:
| Test | What it asks | Software example | Evidence that usually shows it |
|---|---|---|---|
| Permitted purpose | Is the work meant to create or improve a component's function, performance, reliability or quality? | Rebuilding a sync service to handle far higher volume | Epic descriptions, product requirement documents |
| Technological in nature | Does it rely on principles of computer science, engineering or the physical sciences? | Designing a distributed cache with consistency guarantees | Architecture decision records, design docs |
| Elimination of uncertainty | Was the capability, method or appropriate design uncertain at the start? | Unclear whether any approach can meet a latency target | Spike tickets, open questions in design reviews |
| Process of experimentation | Were alternatives evaluated through modeling, prototyping, testing or trial and error? | Three prototype branches benchmarked, two abandoned | Pull request threads, benchmark logs, reverted commits |
The statute also excludes certain activities, including research after commercial production begins, adapting an existing product to a particular customer's requirements, duplicating an existing product, and routine testing or quality control. Read section 41 and its regulations for the exact wording rather than relying on any summary, this one included.
What changes for internal-use software
Software built mainly for a company's own back-office functions, such as accounting, HR or internal reporting, is treated as internal-use software and must clear an additional, higher threshold under the regulations. In broad terms, the work must be innovative, involve significant economic risk and produce something not commercially available. Software the company sells or licenses, or uses so customers can interact with it, generally falls outside the internal-use category.
For agile teams, the practical challenge is mapping sprints to business components. Many advisers treat an epic or a named feature as the component and roll up the tickets beneath it. Form 6765's business component reporting pushes the same way; the guide to Form 6765 Section G covers what it asks for. Expensing the underlying research costs is a separate question, covered in Section 174A explained, and the financial reporting side of internal-use software has its own documentation demands under ASU 2025-06.
Which records support a software R&D credit claim
Records made at the time carry more weight than recollections assembled at year end.
| Record | What it shows | Where it usually lives |
|---|---|---|
| Tickets and epics with acceptance criteria | The goal and the uncertainty at the outset | Jira, Linear, Azure DevOps |
| Pull request discussions | Alternatives considered and why one won | GitHub, GitLab, Bitbucket |
| Commit history and branches | Iteration, including abandoned approaches | The same repositories |
| Design documents and decision records | The technical reasoning behind the architecture | Confluence, Notion, shared drives |
| Test and benchmark results | How each alternative performed | CI logs, test reports |
| Time and payroll allocation by project | Which wages relate to which component | Payroll, time tracking, project tools |
A client that can produce all of these for several years, linked to one another, is unusual. Most can produce some of them for the current year. The difference is worth noticing.
Why the same engineering history matters to AI data buyers
AI developers are moving from models that answer questions to agents that write, review and fix code. Training and evaluating those agents takes records of real engineering work: the problem as stated, the approaches tried, the reviewer who pushed back, the fix that finally merged. Public repositories mostly show finished code; they rarely show the private reasoning around it.
Supply matters too. Researchers at Epoch AI project that, if current trends continue, language models could fully use the stock of public human-written text between 2026 and 2032. It is a forecast with wide uncertainty, but it explains the growing interest in non-public sources such as licensed, rights-cleared engineering histories from established companies.
Rights get the same scrutiny in both exercises. Work employees create within the scope of their jobs is generally owned by the employer, while work from outside contractors may not be unless the parties agree in a signed writing, as the Copyright Office explains in its circular on works made for hire. A credit study already asks who funded the research and who keeps substantial rights; a licensing review asks who owns the records. Clients whose engineering was done largely by outside agencies, or for customers who own the output, need a careful look before any introduction.
What the credit study means for a CPA firm as a referral partner
The study shows the adviser what few outsiders see: how many years of repositories exist, how consistently tickets link to code, and how many systems the engineering team uses. Treat that view as a signal only.
- Finish the credit engagement on its own terms, and keep licensing questions out of technical interviews.
- If the client looks like a fit, raise it separately with the CEO or CFO and ask whether they want an introduction.
- With their agreement, register and share your referral link, or submit the company through the referral form.
- SourceX confirms the baseline: 50+ full-time employees at peak (contractors excluded), several years of documented operations, rights to license, and an authorized sponsor.
- The company describes its own systems and history in a data inventory, agrees price and terms, and decides whether to sign. The firm never copies, uploads or summarizes code or tickets.
Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. Rewards become payable only after the buyer pays and SourceX receives its fee; a lead, meeting or signed agreement alone does not trigger payment, and no reward is guaranteed. The reward is a share of SourceX's fee and is never deducted from what the client receives.
Check fee rules before you register. Credit studies are sometimes priced on a contingent basis, and under the AICPA Code a contingent fee is one whose amount depends on attaining a specific result; members may not perform services for a contingent fee for a client for whom the firm performs an audit, review, certain compilations or an examination of prospective financial information, as the NYSSCPA explains. Referral fees and commissions fall under a separate AICPA rule, ET 1.520, which bars commissions where the firm performs attest work for the client and requires permitted ones to be disclosed; state boards can be stricter. If several people in the firm worked with the client, settle origination credit first; the accountants partner page summarizes the program for firms.
Limits and open questions
- Credit rules and the related expensing rules have changed recently, and more changes are possible. Check current law and IRS guidance for each tax year you prepare.
- AI-assisted coding raises open questions about how to document uncertainty and experimentation when a tool drafts much of the code. Treat confident answers in either direction with caution until the IRS addresses it.
- A strong credit file does not mean a client fits a licensing review. Repositories may be mostly open-source code, owned by clients under services contracts, or thin outside one product.
- A licensing conversation should never influence the credit position, and the credit study should never be used to extract records.
This is general information, not legal, tax or financial advice. Confirm with your own counsel, tax adviser or professional body before acting.
Next step
If a software client in your book has years of linked tickets, pull requests and design records, run a preliminary read in the company fit checker and compare it with who qualifies. Then register as a partner so the introduction is ready once the client agrees, or send the CEO your referral link, which opens sourcex.si/apply with your referral code attached.
- Step 1Share your linkSend your personal link to a company you know.
- Step 2Company appliesThe company applies itself at /apply.
- Step 3Buyer selects and paysThe buyer selects and pays for the data and SourceX receives its fee.
- Step 4You get your rewardYour share of SourceX fees becomes payable.
Common questions
Do bug fixes count toward the software R&D credit?
Routine bug fixes after a product is released generally do not, because work after commercial production and routine quality control are excluded. A defect that forces a genuine redesign, where the team is uncertain how to solve it and tests alternatives, may be different. The answer turns on the facts and the records, so document the uncertainty and the alternatives at the time rather than reconstructing them later.
Can a SaaS company claim the credit for its customer-facing platform?
Customer-facing software that the company sells, or uses so customers can interact with it, is generally not treated as internal-use software, so it avoids the higher threshold that applies to back-office tools. It still has to meet the four-part test component by component. Ask the client to map epics or features to components and to keep the related tickets and code reviews.
How far back do engineering records need to go?
For the credit, the focus is the tax years being claimed, plus whatever supports amounts carried forward. For a licensing review, longer is better: several years of linked tickets, pull requests and design records show how the product and the team's decisions evolved. Clients should not purge older repositories or archived projects just because the credit years have closed.
Does flagging a client for a licensing introduction affect independence?
It can, depending on the services the firm performs. Rules on commissions, referral fees and contingent fees are strictest where the firm provides attest services, such as audits and reviews, for that client, and state boards can add requirements. Check with your ethics partner before you raise it, and disclose any referral fee to the client.
Will SourceX ask the firm for the client's code or tickets?
No. The firm's part is the introduction. The company works with SourceX directly to inventory its systems, set redaction and de-identification requirements, agree price and terms and decide whether to sign. Delivery happens only under an executed agreement with the company's authorization, and the adviser never handles the repositories or ticket data.
Related pages
- Form 6765 Section G: what it asks for and the records behind each business component
- Section 174A explained: domestic R&E expensing, elections and the records behind them
- ASU 2025-06 explained: new internal-use software rules and the evidence they require
- How origination credit works at accounting firms, and where outside introductions fit
- Referral opportunities for accountants and bookkeeping firms
- Check Company Fit for Data Licensing
Free resources
- Working capital calculator — Net working capital, current ratio and quick ratio.
- Due diligence checklist generator — A tailored document request list by deal type.
- Cash flow calculator — A 12-month cash forecast with shortfalls highlighted.
- All free tools · MCP resource center
By SourceX Partnerships Team · Published 2026-10-09 · Updated 2026-10-09
Know a US company with valuable proprietary data?
Become a referral partner from anywhere we support, get your link and introduce an owner or authorized decision-maker.
Refer a company →I own a business
Explore licensing your company's data to AI developers worldwide. Start a short assessment; no uploads needed.
Start an assessment