Which Projects Qualify for the German R&D Tax Credit? 5 Examples from Research & Development

Dr Anne Küsters, September 22nd 2026

Germany’s R&D tax credit, the Forschungszulage, is attractive for many startups and technology companies. Yet one question often remains: Is what we are developing really research and development, or is it simply sophisticated product development?

In practice, eligibility rarely depends on whether a project sounds “innovative.” AI, robotics, cloud software, or new hardware are not automatically eligible for the German R&D tax credit. What matters is whether a company is working on a scientific or technical challenge that has not yet been solved, whether the path to a solution is genuinely uncertain at the outset, and whether the team addresses that uncertainty systematically through development, testing, and experimentation. The best way to understand this distinction is through real project examples.

The following five cases are based on real R&D projects that we have supported at DnA Ventures in connection with the German R&D tax credit. To protect confidentiality, company names and selected technical, commercial, and project-specific details have been changed. The underlying R&D logic, however, remains the same.

Example 1 - R&D Tax Credit for AI: When AI Agents Need to Make Reliable Decisions

Our first example comes from a project designed to automate complex internal decision-making processes and highly manual workflows in a critical operational environment. At first glance, this might sound like a straightforward case of integrating a Large Language Model into existing enterprise systems. But this is precisely where the distinction between routine software development and genuine R&D begins.

The technical challenge lies in orchestrating multiple specialized AI agents so that they can work with heterogeneous enterprise data, perform different tasks, and still arrive at reliable and consistent outcomes. Structured and unstructured information from multiple internal systems and documents must be processed within a single workflow. At the same time, the system should do more than summarize information. It should prepare complex decisions or, within clearly defined boundaries, make certain decisions autonomously.

The central R&D question is therefore not: “Can we build an AI agent?” It is: Can a probabilistic AI system operate reliably, reproducibly, and securely within a multi-step, business-critical process?

This is where the technical uncertainties arise. Domain-specific fine-tuning on small or inconsistent datasets can produce unstable results. Specialized agents may interpret information differently or generate conflicting outputs when objectives compete. Evaluating the performance of such systems is also significantly more complex than testing conventional software functions. On top of this come questions of robustness and security, particularly when the system has to handle unstructured, incomplete, or potentially manipulated inputs.

The project therefore goes far beyond a conventional AI integration. Its R&D core lies in developing and testing new orchestration, evaluation, and safety mechanisms for agentic systems. These are exactly the kinds of technical questions that can distinguish routine development from eligible research and development.

‍

‍

Example 2 - R&D Tax Credit for Energy Tech: Optimizing Industrial Battery Storage

Another example involves the development of software for the intelligent control of industrial battery storage systems. Here again, the use of Machine Learning or a digital twin is not what determines R&D eligibility. The decisive factor is the technical problem behind the technology. An industrial battery storage system has to account for several variables at the same time. Energy consumption and generation change continuously, market prices fluctuate, and the battery itself operates within technical and physical constraints. Effective control therefore requires forecasts, system behavior, and market information to be combined in a single decision-making process.

To address this challenge, the project develops an integrated system consisting of forecasting models, a digital twin, and mathematical optimization. The goal is to derive an optimal operating strategy automatically from different data sources and transfer that strategy to the physical storage system in real time.

The R&D question lies in the interaction between these individual components. A forecasting model may perform well under normal conditions but deteriorate significantly during extreme weather, changing production profiles, or highly volatile market conditions. A digital twin can only approximate the real behavior of a storage system. And even a mathematically optimal result has limited value if latency, communication failures, or real-world operating conditions prevent the control strategy from being implemented reliably.

The system must therefore be tested, simulated, and validated continuously. The technical uncertainty does not lie in a single algorithm, but in whether forecasting, optimization, and real-time control can function reliably as an integrated system. That is what can make this type of project relevant for the German R&D tax credit. It is not simply the implementation of a standard energy management system, but the development of a new integrated decision model under real technical uncertainty.

Example 3 - R&D Tax Credit for Deep-Tech Software: Generating Complex 3D Models in Real Time

This project focuses on developing a cloud-based solution capable of generating and collaboratively processing complex three-dimensional models from heterogeneous technical data significantly faster.

The challenge is not limited to visualization. The underlying datasets can differ substantially in quality, resolution, and structure, and may even contradict one another. At the same time, the modeling process needs to be performant enough for results to be updated almost in real time rather than after long computation periods.

The R&D core therefore lies in the question of whether heterogeneous and partially inconsistent input data can be processed in a way that produces stable, interpretable, and rapidly computable 3D models.

To achieve this, the project develops new probabilistic and Machine Learning-based modeling methods, investigates approaches for quantifying uncertainty, and optimizes the computational logic for a scalable cloud architecture. In parallel, large volumetric datasets need to be processed efficiently enough to support visualization and collaborative work across different devices.

The technical uncertainties arise primarily at the interfaces between these components. Contradictory inputs can destabilize models, large datasets can prevent real-time computation, and resource-intensive visualization can limit usability on devices with lower computing power.

It is precisely this combination of algorithmic development, scalability, uncertainty modeling, and experimental validation that forms the R&D core of the project.

‍

‍

Example 4 - R&D Tax Credit for Climate Tech: Automating Complex Heat Networks

This project focuses on developing an AI-based optimization solution for complex heat networks. Different heat sources, storage systems, volatile electricity prices, weather data, and changing demand profiles all need to be considered simultaneously. The technical challenge is not simply to collect these variables, but to turn them into reliable control decisions within a short period of time. Conventional optimization methods can model such systems in principle, but often reach their limits as complexity increases.

The project therefore develops a hybrid approach combining Machine Learning, constraint programming, and mathematical optimization. The objective is to connect forecasts, technical system constraints, and economic parameters in a way that allows the heat network to be controlled automatically and adaptively. The central R&D question is whether these different methods can be combined in a way that makes the overall system fast, precise, and robust enough for real-world operation.

Several technical uncertainties remain. Computation may be too slow to react to short-term market changes. The models may not represent actual system behavior with sufficient accuracy. And combining different optimization methods may produce unstable or unreliable results.

The development work therefore involves designing, combining, simulating, and validating new optimization approaches under real operating and market conditions. This methodological and technical uncertainty is what forms the R&D core of the project.

Example 5 - R&D Tax Credit for Robotics and Advanced Manufacturing: When a Robot Develops Its Own Repair Strategy

This project involves the development of a mobile robotic system for the automated repair of large industrial components. The goal is to automate a process that is currently highly manual, enabling damaged surfaces to be detected, analyzed, and subsequently repaired using robotics. The geometry of the component is first captured using sensors and converted into a digital model. Based on this model, the software is intended to derive an appropriate repair strategy and machining path automatically before transferring the instructions to the robotic system.

The R&D core does not lie in using a robot or individual sensors. It lies in the question of whether sensors, real-time analysis, path planning, and robot control can operate reliably as one integrated system under real industrial conditions.

The technical uncertainties are significant. Dust, changing lighting conditions, and different surface properties can affect geometry capture. Sensor data must be precise enough to generate reliable repair paths. At the same time, the software has to process complex component geometries quickly and generate safe movement strategies for the robot. The integration of the individual components adds another layer of uncertainty. Even if the sensors, software, and robot control work independently, this does not guarantee that the complete system will operate robustly, safely, and in real time.

The development work therefore includes experimental testing of geometry capture, the development of automated analysis and path-planning methods, and validation of the hardware-software integration under real operating conditions. This technically uncertain system integration forms the R&D core of the project.

What These Five R&D Tax Credit Examples Have in Common

The projects are technologically very different. They range from AI agents and Energy Tech to 3D modeling, Climate Tech, and robotics. Yet their potential eligibility for the German R&D tax credit follows the same fundamental logic. Each project begins with a concrete limitation of existing solutions.

The relevant question is not whether a company is developing something “innovative,” but rather: What technical capability cannot existing technologies reliably deliver?

The second element is uncertainty. In an eligible R&D project, the path to a solution is not fully known at the outset. The team cannot be certain that the chosen approach will work, that the required performance will be achieved, or that several technical components can be combined reliably.

Finally, the project must follow a systematic development process. Teams develop models, build prototypes, run simulations, train algorithms, compare different approaches, and validate their results under real-world conditions.

This is also why the German R&D tax credit should not be confused with ordinary product development. A new software feature may be technically challenging and still qualify as routine development. At the same time, one specific technical component of a larger product may constitute genuine R&D even if other parts of the product clearly do not.

Assess your R&D project's eligibility together with DnA

Technology-driven startups often underestimate how much genuine R&D is already embedded in their product development. Companies working on new software architectures, AI models, hardware, robotics, energy optimization, industrial automation, or new manufacturing processes should therefore ask more than whether their product is innovative.

A better question is: What technical problem did we have to solve where, at the beginning, it was unclear whether or how it could be solved at all?

If there is a clear answer to that question and the team has addressed the problem systematically through development and testing, it may be worth assessing the project for eligibility under the German R&D tax credit.

DnA combines technical and project expertise with a startup and financing perspective. Together, we define the R&D project, identify potentially eligible expenses and determine the next steps. Contact us to check your project’s eligibility now or schedule a call.

About DnA Ventures

From industry, for industry: DnA Ventures combines industrial and startup experience with venture capital and non-dilutive R&D financing. Anna Saari and Maite Pazmino are your contacts for an initial eligibility assessment.

Please note: This article provides general information and does not constitute tax or legal advice. The examples shown are based on real project types from our work. Names and selected technical, commercial, and project-specific details have been changed to protect confidential information. The examples do not constitute an automatic indication of eligibility or approval for comparable projects. Actual eligibility and the amount of funding available depend on the individual circumstances. Information correct as of September 22nd, 2026.

‍

‍