SR&ED for AI and Machine Learning Companies in Canada

What AI/ML work qualifies for SR&ED

AI and ML companies frequently conduct systematic investigation of technical uncertainty — the core of SR&ED eligibility. The key is separating genuine research from routine application of established techniques.

• Novel model architecture development for domains without established benchmarks
• Systematic investigation of training methodologies beyond standard hyperparameter tuning
• Custom data preprocessing and feature engineering for new problem domains
• Inference optimization requiring systematic investigation of novel approaches
• Evaluation framework development for capabilities not covered by standard metrics

What doesn't qualify

Routine AI work generally does not qualify for SR&ED. The distinction between research and application matters for eligibility.

• Standard fine-tuning using documented procedures and established parameters
• Routine API integration with GPT, Claude, or other LLM services
• Applying published models to new data without systematic investigation
• Standard prompt engineering for well-documented use cases

More guides

Ready to check if your work qualifies? Take the free eligibility assessment.