Artificial intelligence · Case study

Research allowance for AI: millions, in-house only.

Practical example based on real projects · anonymised, figures rounded · Last updated: 24 September 2026

An AI company funded explainable, robust, privacy-compliant document analysis on an eligible base over €5 million, entirely from in-house work, with no contract research at all.

€1.75M

research allowance granted

€5M

eligible base

Up to 35%

SME funding rate

Get a free funding assessment Read the story

At a glance

The mandate in figures

IndustryArtificial intelligenceExplainable document analysis

Research allowance granted€1.75M100% in-house, €0 contract research

Eligible base€5M

ScenarioCertified, in-house only

Projects1

SME funding rateUp to 35%

About the company

The company, and the R&D underneath

The company develops AI methods that not only read documents but do so in an explainable, robust and privacy-compliant way. That is more demanding than it sounds. A model whose decisions are traceable, that stays stable even with difficult or incomplete inputs and that meets high data-protection requirements cannot be assembled from standard parts. Research happens on several open fronts at once.

The entire development was in-house. There was no significant contract research, the eligible base is fed entirely by the company's own personnel work.

The challenge

Where the funding really sat

With AI projects the risk lies less in the eligibility of the technology, which is usually given, than in the precise presentation. Three things must be right.

First, the delineation from the state of the art at the right point in time. AI moves fast. What is standard today may have been open at the project start. What matters is the state at project start, not today's.

Second, the clean capture of in-house work. If the entire funding hangs on personnel costs, the eligible personnel work must be captured correctly across several years and a large team and assigned to the fiscal years.

Third, the language. An AI application must not drown in marketing terms or buzzwords. The review wants to see methods, parameters, metrics and the concrete technical uncertainty.

Novelty Technical risk / uncertainty Systematic approach

Our approach

How we built the case

We built the project as experimental development with clearly named open research questions: explainability, robustness and privacy compliance each as its own technical challenge with an open outcome. We researched the state of the art at the time of the project start and named the established methods against which the in-house development sets itself apart.

We captured the in-house work cleanly across the full term and assigned it to the fiscal years, so every euro takes effect in the right year and at the right rate. We deliberately kept the application text grounded, methods and metrics instead of slogans. The project was cleanly certified, without dispute over eligibility.

How the case moved

↻

Scoped

Open research questions named

↻

Documented

In-house work, year by year

✓

Certified

No dispute over eligibility

What made up the eligible base

In-house personnel work 100%

In-house personnel work

Eligible base over €5 million, entirely in-house work, €0 contract research. Figure shown is illustrative.

The result

A €1.75M research allowance.

The eligible base is around €5.1 million across two fiscal years (2024 and 2025), entirely from in-house work. At 25% on the share before 28 March 2024 and the 35% SME rate after that, this yields around €1.75 million of research allowance, without a single euro of contract research. For an AI company with high personnel costs, this is a central financing building block.

Eligible base €5M · SME rate Up to 35%

Certified, in-house onlyThe outcome of the mandate.

Eligible base €5MRecognised cost the allowance is calculated on.

SME rate Up to 35%Applied to the eligible base.

Key takeaways

What other companies can learn

AI development is a strong funding case, with the right presentation. Three points are decisive:

01

In-house work is the engine

With personnel-intensive development the largest lever sits in the company's own personnel costs. Clean, year-by-year capture is half the battle.

02

State of the art at project start

Especially with fast-moving technology, the state at that time decides novelty. Arguing with today's state weakens the application.

03

Write grounded

Methods, parameters, metrics and the concrete technical risk convince the review. Buzzwords do not.

More on Research Allowance: The guide Research Allowance: requirements, rates and application explains the programme step by step. How we support your project is covered under Research Allowance consulting.

“

Artificial intelligence is one of the most rewarding fields for the research allowance, the work is personnel-intensive, technically deep and often clearly delineable from the state of the art.

BeFunded On the research allowance

FAQ

Research Allowance: your questions

The most common questions on this kind of case. Short answer first, detail after.

Yes. AI and machine-learning development is regularly eligible if it goes beyond the state of the art and carries a scientific-technical uncertainty. Merely applying finished models without own development is not.

No. The company's own development work is usually the largest lever. Even a project without any contract research can reach an eligible base in the millions.

The state at the time of the project start. Because AI develops quickly, this temporal anchoring is particularly important for novelty.

The eligible personnel costs of in-house development flow into the eligible base. The funding rate is applied, for SMEs up to 35% for costs incurred after 27 March 2024.

Keep reading

More case studies

Biotechnology

Mechanics, control and biology funded as one interdisciplinary R&D project.