Claiming Artificial Intelligence or Machine Learning centric projects are slightly more complex compared to other technologies. This is due to:
- Lack of proper understanding of the technology involved, given it is fairly new and developing at a fast pace
- Overgeneralization that AI models are only about training, and therefore require no experimentation
This can be addressed through the following approaches:
Focus on real technical improvements:
Focus on the specific (overlaid) design modifications that improved the capabilities of the known algorithms and models (that served as the core of the overall design) above and beyond published literature, and not claiming the whole AI enabled solution as SR&ED.
Technical challenges associated with real-world scenarios:
Although new algorithms and techniques are published at a very high frequency; most are still applicable in a theoretically ideal domain and therefore require experimental development and prototyping to identify effective enhancements to existing model designs that would not only be functional but practical in the real world with real-world data limitations
Unavailability of sufficient training data:
It is well understood that AI models can be trained to be highly accurate if sufficient data is available; however, in most practical scenarios lack of training data is a very serious limitation which is hard to overcome in a short period of time, which can force claimants to design and develop new training methods in-house
Choosing support activities wisely:
Data collection and data entry and subsequent model training efforts may be claimed as supporting activities rather than claiming them as experimental development.
Our SR&ED Technical Experts walk through the specifics of your projects thoroughly to ensure all SR&ED eligible are accurately identified and included in the technical reports submitted to the CRA, ensuring we maximize your benefits and reduce uncertainties with your claim process.