Is software development eligible for the CIR, or for the other French R&D incentives?
· By David Jian, Partner and Certified Public Accountant, KeyPoint International
Short answer
Some of it can be. Writing software is not research, however new or difficult the software is. France's research tax credit (the CIR, or crédit d'impôt recherche) covers the part of the work that resolved a technical problem for which no solution existed in published knowledge, and only that part.
The government guide that tax inspectors work from puts it bluntly: building a piece of software, an application, or a new feature in existing software, with no explicit research approach behind it, does not qualify.
If your work falls the wrong side of that line, the CIR is not the end of it. Two other schemes are drafted in a way that often fits software better.
What the government guide says
The reference most inspectors use is the research tax credit guide published each year by the ministry for higher education and research. Its 2025 edition, the current one, devotes a section to software and IT.
The principle it sets out is the same one that governs every other field. The work has to start from a clearly identified problem with no obvious, available solution in the existing knowledge of the field, and it has to follow a rigorous, documented scientific approach that removes the uncertainty. Our six-question eligibility guide covers that general test.
What the software section adds is a list of what this looks like in practice.
The four kinds of software work that qualify
| Category | What it means |
|---|---|
| A new or improved method, algorithm or protocol | A new technical element, alone or added to an existing system, documented with the obstacles identified, the operating principles, the development stages and the results. Performance has to be compared against earlier solutions. |
| A formalised or adapted methodology | Formalising a development, test, deployment or validation process that was not documented before, or substantially adapting an existing methodology to a specific case. The output is a reproducible, assessable framework. |
| Applying recent scientific concepts | Experimental use of concepts at the edge of current knowledge, such as artificial intelligence, quantum computing or advanced cryptography, in a new or constrained setting, with rigorous validation showing a gain over existing approaches. |
| Experiments and prototypes | Prototypes and mock-ups built to test and measure a scientific or technical advance, with defined experimental protocols: datasets, test scenarios, performance criteria. |
For each one, the guide expects you to justify the scientific or technical obstacle, the objective pursued, the novelty obtained, and the experimental protocols used to validate it.
Eight traps the guide names
Most software claims are lost here rather than on the four categories above.
- Novelty: every piece of software is unique by definition, so being new proves nothing on its own. Only work that resolves a new scientific or technical problem and produces knowledge qualifies.
- Complexity: a hard project is not an eligible one. What counts is an original solution, with an approach that is generic and reusable.
- Agile: iterative development is not research in itself, and iterations that fix defects or work towards a specification are generally outside the base.
- Prototypes: building one proves nothing by itself. It has to validate a research concept.
- Skills: getting your team up to speed on a technology or a tool is training. The CIR funds the creation of usable scientific knowledge.
- Emerging technology: integrating it can qualify, but only where making it work required solving a scientific or technical problem. Describe the problem, not the technology.
- Methodology: designing or adapting a development methodology can qualify, provided the research approach behind it is made explicit.
- IT as a tool: using software to solve a problem that is not itself a software problem does not make the work software research. On multidisciplinary projects, mark the eligible part out carefully.
Not sure which side of the line your project sits on? Our CIR eligibility test walks through the criteria in two minutes, with no sign-up. → Take the CIR eligibility test
What the file has to contain
For software, the guide asks for a documented trail: the hypotheses, the development stages, the evaluation methods, the data collected, the analysis of the results and the conclusions.
Two practical points follow from that. A repository history is not a technical file, but it is good evidence of when work happened and how many attempts it took, so keep the experimental branches rather than squashing them away. And the attempts that failed are worth more than the ones that worked, because they are what shows the uncertainty was real.
If the CIR does not fit: two schemes that may
The innovation tax credit (the CII) covers the design of prototypes and pilot installations of new products, at 20% of spending capped at €400,000 a year, and it runs until 31 December 2027. The law defines a new product as a tangible or intangible good that is not yet on the market and that outperforms what exists on technical grounds, on ergonomics, on eco-design or on functionality. Software fits that definition, and the test is commercial novelty, not a scientific one, which is a far lower bar than the CIR. The catch is that the CII is reserved for small and medium-sized companies as the EU defines them, and that test counts your whole group.
The IP Box is the one software companies most often miss. Article 238 of the tax code lists copyright-protected software among the assets whose net licensing income can be taxed at 10% instead of the standard rate. There is no size condition and no group test. If you license your software rather than only selling subscriptions to a service, model the difference before you file.
Beyond those, young innovative company status (jeune entreprise innovante, or JEI) exempts employer social contributions on R&D staff, for young companies still held by their founders. An industrial research training agreement (CIFRE) pays €14,000 a year for three years towards a PhD researcher, computer science included.
Frequently asked questions
Is an artificial intelligence project eligible for the CIR? It can be. The government guide names artificial intelligence among the emerging concepts whose experimental application can qualify. Using an existing model through an API is not research; resolving a technical obstacle in making it work in a new or constrained setting can be.
We build a SaaS platform. Is that eligible? Not as such. The platform is the product. What may be eligible is a specific operation inside it that resolved an obstacle with no published solution, for a defined period, which is rarely the whole build.
Does working in Agile stop us claiming? No, but it does not help you either. The guide is explicit that iterative development is not in itself research, and that iterations aimed at fixing defects or meeting a specification are generally not eligible. What matters is the uncertainty you were resolving.
Can we claim the CII if the CIR is refused? They are different tests, so yes in principle, provided the company is a small or medium-sized one at group level and the work concerns a prototype of a new product. The two are claimed on the same form and can apply to different phases of the same project.
Next steps
→ Take the CIR eligibility test → Estimate your CIR with the calculator → How the innovation tax credit works → How the IP Box works
Have a software project you are unsure about? We help dozens of software companies a year separate the eligible operations from the engineering around them, and document them so they survive an audit. Send us a short description of what you built and what stopped working, and we will tell you what is worth claiming. → Contact us
This article reflects the law and the government guide as at 13 October 2026. It is not advice on a specific situation.