Research Allowance for Software, AI & SaaS
Software and AI companies are eligible on a technology-neutral basis – but only the research share, not routine development. Eligible are new algorithms, methods and prototypes with genuine technical risk; testing, debugging, porting or configuration are not. We translate your development work into audit-proof research language.
All free & non-binding – no upfront risk.
Context
The BSFZ review guidelines list ICT as a dedicated research field (including Data Science/Big Data, Industry 4.0, IoT). The decisive factor is the distinction: software development is only R&D when there is a scientific-technical uncertainty that cannot be resolved with routine, 'by-the-book' procedures.
Which R&D activities in Software, AI & SaaS are eligible?
- Development of new or significantly more efficient algorithms on an original technological basis
- New encryption or security mechanisms that go beyond implementing known standards
- Resolving hardware/software conflicts through novel system or network configurations that cannot be solved 'by the book'
- New methods for data acquisition and processing that enable insights beyond known analyses
- Prototypes or pilot applications in a laboratory or simulated environment to validate technological foundations
- Transferring a method into a new, technically uncertain context (e.g. B2C → data-scarce B2B) with genuinely new performance limits
Example: research allowance in Software, AI & SaaS
For illustration only – not a real client case.
Assumption: Assume around €480,000 of eligible R&D staff costs per year for a qualifying Software, AI & SaaS project.
Simplified example based on the eligible R&D staff costs (as of July 2026); excluding the overhead flat rate, contract research, owner work and the cap. Your actual funding depends on the individual case.
Typical non-eligible activities & reasons for rejection
This is exactly where many first applications fail. These points come from the BSFZ review guidelines:
- Applying existing methods for data security; certification of existing products or programs
- Setting up data centers, firewalls, monitoring or hosting services
- Installing and configuring existing server-client, cloud or edge solutions
- Porting software (e.g. standalone → cloud)
- Assembling software from modular building blocks without scientific-technical risk
- Routine development including requirements analysis, testing, debugging and documentation
Research language instead of development language
The BSFZ assesses by scientific criteria. An example from Software, AI & SaaS:
“The search didn't work properly.”
“Whether keyword-based retrieval methods achieve sufficient precision and recall across heterogeneous source systems – in fact they failed due to a lack of semantic disambiguation.”
FAQ: research allowance for Software, AI & SaaS
Which R&D activities in software/AI are eligible?
Eligible is the research work involving technical uncertainty: new or significantly more efficient algorithms, new data-processing methods, new security mechanisms and prototypes to validate technological foundations. The benchmark is a genuine technical-scientific risk, not the degree of novelty for your company alone.
Are testing, debugging and documentation eligible?
No. The review guidelines explicitly count routine development – including requirements analysis, testing, debugging and documentation – among the non-eligible activities.
Does migrating our software to the cloud count?
Pure porting (e.g. standalone → cloud) is not eligible. A cloud implementation can only become eligible if it involves a genuine technical uncertainty that cannot be resolved with routine procedures.
Is GDPR compliance an eligible risk?
No. According to the review guidelines, data protection concerns do not count as technical uncertainty. The risk must be scientific-technical (e.g. 'does the model converge under these conditions?').