What Is SR&ED?
The Scientific Research and Experimental Development (SR&ED) program is a federal tax incentive administered by the Canada Revenue Agency (CRA). It has existed since 1986 and is the cornerstone of Canada's R&D support system.
The program encourages Canadian businesses of all sizes and industries to conduct research and development in Canada by providing investment tax credits (ITCs) on qualifying R&D expenditures. These credits can be refundable — meaning the government sends you a cheque even if you owe no taxes.
SR&ED covers a wide range of activities including basic research, applied research, and experimental development. The program supports work in all fields of science and technology, including software development, artificial intelligence, and machine learning — as long as the work meets the CRA's three-part eligibility test.
Unlike grant programs such as NRC IRAP or CDAP (now closed), SR&ED does not require a pre-approval application. You file your claim with your corporate tax return after the work has been completed. This makes it one of the most accessible R&D incentives available to Canadian businesses.
Eligible expenditures include salaries of employees directly performing R&D, contractor costs (at 80%), materials consumed in R&D, and overhead (calculated using the proxy method or traditional method). Most software and AI companies claim primarily on salary expenditures.
SR&ED Rates: Federal and Provincial
SR&ED credits are calculated at both the federal and provincial level. The total credit rate depends on your corporation type and province. Canadian-Controlled Private Corporations (CCPCs) with taxable income under $500K receive the most generous rates.
| Jurisdiction | Rate | Type | Notes |
|---|---|---|---|
| Federal (CCPC) | 35% | Refundable | First $3M of qualifying expenditures |
| Federal (Other) | 15% | Non-refundable | No expenditure limit; carry forward 20 yrs |
| Ontario (ORDTC) | +8% | Refundable | Ontario R&D Tax Credit |
| Quebec | +14% | Refundable | R&D salary credit (up to 30% for SMEs) |
| British Columbia | +10% | Refundable | BC SR&ED Tax Credit |
| Alberta | +10% | Refundable | Alberta Innovation Employment Grant |
Ontario CCPC
43% combined
35% federal + 8% Ontario ORDTC
Quebec CCPC
49% combined
35% federal + 14% Quebec R&D credit
Provincial rates may vary based on corporation type, income level, and specific program rules. The rates above reflect the most common scenario for qualifying CCPCs. Always confirm current rates with your tax advisor or the relevant provincial authority.
What Qualifies as SR&ED?
The CRA uses a three-question test to determine whether work qualifies as SR&ED. All three criteria must be met. This test is defined in the Income Tax Act (Section 248(1)) and further clarified in CRA's Eligibility Policy document (CPS 2015-017). Understanding these criteria is essential for both identifying eligible projects and building defensible claims.
Technological Uncertainty
Was there a technological uncertainty that could not be resolved by routine engineering or standard practice? The uncertainty must be technological, not business, market, or financial in nature. It must exist at the outset of the work and must not be resolvable simply by applying existing knowledge or consulting available documentation.
AI example: “Could a transformer-based model achieve >92% accuracy on our proprietary medical imaging dataset, given the limited labelled training data and class imbalance inherent in the domain?” The answer was not knowable in advance through existing literature or standard practice.
Systematic Investigation
Was the work carried out through a systematic investigation or search using hypothesis, testing, and analysis? The work must follow a structured approach — forming hypotheses about potential solutions, designing experiments to test them, and analyzing results. Ad hoc trial-and-error without a structured methodology does not meet this criterion.
AI example: “We hypothesized that data augmentation combined with a custom loss function would address the class imbalance. We ran controlled experiments comparing five augmentation strategies, measured precision/recall across minority classes, and analyzed failure cases to iterate on the approach.”
Technological Advancement
Was the work undertaken for the purpose of achieving a technological advancement? The goal must be to extend the existing base of technology or scientific knowledge — even if that advancement is specific to your company's context. The advancement does not need to be novel to the entire world; it must be beyond the company's or industry's current technological capability.
AI example: “We developed a novel ensemble approach combining a fine-tuned vision transformer with a domain-specific post-processing pipeline that achieved 94.3% accuracy on our dataset — a result that was not achievable through any known or documented method applied to this problem domain.”
It is important to note that SR&ED eligibility is assessed at the project level, not the company level. A single company can have both qualifying and non-qualifying projects running simultaneously. The key is identifying which specific activities within each project meet all three criteria.
AI-Specific SR&ED: When AI Development Qualifies
AI and machine learning work is one of the fastest-growing categories of SR&ED claims in Canada. However, not all AI work qualifies. The distinction hinges on whether you are pushing technological boundaries or applying established techniques to a well-understood problem. Below are concrete examples to help you assess your own projects.
Activities That Typically Qualify
Training custom ML models to solve novel problems
Building a proprietary model for a domain-specific task where off-the-shelf solutions do not achieve acceptable performance (e.g., custom NLP models for Canadian legal documents, specialized computer vision for agricultural inspection). The uncertainty about whether the model can achieve the required accuracy on your specific data makes this eligible.
Developing new data processing pipelines with technological uncertainty
Creating novel ETL architectures or feature engineering approaches where the feasibility of processing data at the required scale, latency, or quality level is uncertain. For example, building a real-time streaming pipeline that must process and classify 100K events/second with sub-200ms latency using ML inference.
Designing novel model architectures or training approaches
Modifying transformer architectures, developing custom attention mechanisms, creating new loss functions, or experimenting with training regimes (e.g., curriculum learning, novel fine-tuning strategies) to address domain-specific challenges that standard approaches cannot solve.
Building ML infrastructure with uncertain performance characteristics
Developing model serving infrastructure where it is technologically uncertain whether the system can meet latency, throughput, or cost targets. Creating automated retraining pipelines that must maintain model quality within defined thresholds over time with shifting data distributions.
Activities That Typically Do NOT Qualify
Using ChatGPT or other LLM APIs with no modification
Integrating a third-party AI API (OpenAI, Anthropic, Google) into your application using standard prompting techniques is not SR&ED-eligible. There is no technological uncertainty when the API provider has already solved the underlying technology challenge. Prompt engineering with documented best practices is considered routine application of known technology.
Routine software development with known solutions
Building a web application, mobile app, or dashboard using established frameworks and patterns does not qualify, even if the application uses AI features. If the path to the solution is well-documented and predictable, there is no technological uncertainty.
Applying standard ML techniques to a well-understood problem
Using a pre-trained model (e.g., ResNet for image classification, BERT for sentiment analysis) on a standard dataset with known performance benchmarks is routine application of existing technology, not SR&ED-eligible work.
Market research, business analysis, or data collection
Gathering training data, conducting market research to identify AI use cases, or performing business analysis to prioritize AI initiatives are not technological activities and do not qualify, even though they may support an R&D project.
What Does NOT Qualify: Common Misconceptions
Many companies either over-claim (leading to CRA reviews and repayments) or under-claim (missing legitimate credits) because they misunderstand what SR&ED covers. Here are the most common misconceptions:
Myth: “Any software development qualifies for SR&ED”
Reality: Only work involving genuine technological uncertainty qualifies. Building features, fixing bugs, or implementing well-documented patterns is routine development, not R&D.
Myth: “Using AI or machine learning automatically makes it SR&ED”
Reality: The technology used is irrelevant. What matters is whether the work addresses a technological uncertainty through systematic investigation. Using AI to solve a routine business problem with known solutions is not SR&ED.
Myth: “Business uncertainty equals technological uncertainty”
Reality: "Will customers use this feature?" or "Can we build this within budget?" are business uncertainties. SR&ED requires technological uncertainty: "Can this be achieved given the current state of technology?"
Myth: “Failed projects don't qualify”
Reality: Failure is actually a strong indicator of genuine technological uncertainty. If you attempted to solve a technological problem through systematic investigation and failed, the work can still qualify. The advancement does not need to succeed; the attempt itself qualifies.
Myth: “You need a dedicated R&D lab or department”
Reality: Any company of any size can claim SR&ED. Many successful claims come from developers who spend part of their time on qualifying R&D activities alongside routine development work. You claim only the portion of their time spent on eligible activities.
Myth: “SR&ED is only for hardware or manufacturing”
Reality: Software development, AI, and data science are among the largest categories of SR&ED claims. CRA has specific guidance for software-related claims and regularly reviews claims in this sector.
How to Calculate Your SR&ED Claim
Below is a worked example for an Ontario-based CCPC with three developers spending 100% of their time on qualifying SR&ED activities. In practice, most developers split their time between eligible R&D and routine work — you claim only the eligible portion.
| Line Item | Amount |
|---|---|
| Eligible salaries (3 developers × $100K) | $300,000 |
| Proxy overhead (55% of salary for T4 employees) | Included in calculation |
| Federal ITC (35% × $300K) | $105,000 |
| Ontario ORDTC (8% × $300K) | $24,000 |
| Total Refund | $129,000 |
Important: This is a simplified illustration. Actual calculations involve the proxy method or traditional method for overhead, adjustments for government assistance received, and potential clawbacks at higher income levels. Contractor costs are eligible at 80% of the invoiced amount. Consult with an SR&ED specialist or tax advisor for accurate calculations specific to your situation.
Many SR&ED consultants work on a contingency basis, charging 15–25% of the credit recovered. This means you pay nothing upfront and the consultant's fee comes from the refund. Even after consultant fees, the net recovery in our example would be $97K–$110K — essentially getting paid to do R&D you were already planning to do.
How to Document SR&ED Claims
Documentation is the single most critical factor in a successful SR&ED claim. CRA expects contemporaneous documentation — records created during the R&D process, not written retroactively at tax time. For AI and software projects, strong documentation practices are essential because the work is often intangible and difficult to demonstrate after the fact.
Project Descriptions
- Clear statement of the technological uncertainty at the project outset
- Description of what existing knowledge or approaches were considered and why they were insufficient
- Hypotheses formed about potential solutions
- Timeline of when R&D activities started and ended
Experiment Records
- Model training logs with hyperparameters, architectures, and configurations tested
- Performance metrics (accuracy, F1, latency, etc.) for each experiment
- Experiment tracking tool exports (MLflow, Weights & Biases, etc.)
- Analysis of results explaining why approaches succeeded or failed
Time Tracking
- Timesheets or time-tracking records showing hours spent on eligible activities
- Clear separation between R&D time and routine development time
- Per-project time allocation for employees working on multiple projects
- Records should be weekly or bi-weekly, not reconstructed annually
Technical Artifacts
- Git commit history showing the progression of R&D work
- Code review comments discussing technical approaches and uncertainties
- Architecture decision records (ADRs) or design documents
- Jupyter notebooks with experimental analysis and conclusions
A common best practice is to conduct brief “SR&ED check-ins” at the end of each sprint or development cycle. During these 15–30 minute sessions, the team documents what uncertainties they addressed, what experiments they ran, and what they learned. This creates a rich, contemporaneous record that dramatically strengthens your claim.
SR&ED + Other Programs: Stacking Strategy
One of the most powerful aspects of the SR&ED program is that it can be combined with other government incentives. However, stacking requires careful planning to ensure you maximize total recovery without running afoul of government assistance rules. When you receive funding from another program, you must reduce your SR&ED qualified expenditure pool by the assistance amount before calculating your credit.
SR&ED + NRC IRAP
IRAP funding for AI adoption can be combined with SR&ED, but IRAP contributions are considered government assistance and must be deducted from your SR&ED expenditure pool. For example, if IRAP funds $50K of a developer's $100K salary, you can only claim SR&ED on the remaining $50K for that developer. Strategy: allocate IRAP funding to non-SR&ED-eligible activities (market research, deployment, training) and reserve pure R&D salary costs for your SR&ED claim.
SR&ED + OIDMTC
Ontario's Interactive Digital Media Tax Credit can be claimed alongside SR&ED for Ontario companies building interactive digital products with AI components. The key is ensuring you do not claim the same expenditure under both programs. Typically, the interactive/UX development costs go to OIDMTC while the underlying AI research costs go to SR&ED. An experienced tax advisor can help structure the allocation to maximize total credits.
SR&ED + CDAP (closed) / BDC Data to AI
The Canada Digital Adoption Program (CDAP) closed to new applications in February 2024. Its successor, the BDC Data to AI advisory program, now provides digital adoption support. If a digital adoption strategy leads to AI R&D projects, those subsequent projects can generate SR&ED claims. The advisory or planning phase is not eligible for SR&ED, while the actual R&D implementation can be claimed separately. This remains a common pathway: digital adoption planning funds the discovery, SR&ED funds the research.
Pro tip: Map out all your eligible programs at the start of the fiscal year. Create a master spreadsheet that allocates each developer's time and each project's costs to the most advantageous program. This prevents double-claiming and ensures you capture the maximum total benefit. Our free assessment identifies which programs your business may align with so you can plan your stacking strategy.
Ready to explore SR&ED for your business?
Take our free 10-minute AI readiness assessment to understand where you stand, then book a discovery call to discuss how we can help.
Related Articles
Frequently Asked Questions About SR&ED
What is the SR&ED tax credit?
SR&ED (Scientific Research and Experimental Development) is Canada's largest tax incentive program for R&D. It provides refundable and non-refundable investment tax credits to businesses that conduct qualifying research and development in Canada. The federal program is administered by the Canada Revenue Agency (CRA) and can be supplemented by provincial credits.
Can AI and machine learning projects qualify for SR&ED?
Yes. AI and machine learning projects can qualify for SR&ED if they involve genuine technological uncertainty, systematic investigation, and aim to achieve a technological advancement. Training custom models to solve novel problems, developing new architectures, or creating novel data pipelines with uncertain outcomes are common qualifying activities. Simply using pre-built AI APIs or applying known techniques with predictable results does not qualify.
What is the difference between refundable and non-refundable SR&ED credits?
A refundable credit is paid to you as cash even if you owe no taxes. CCPCs (Canadian-Controlled Private Corporations) with less than $500K taxable income earn 35% refundable credits on the first $4.5 million of qualifying expenditures. A non-refundable credit (15% for non-CCPCs) can only reduce taxes owed and any unused portion is carried forward up to 20 years.
How much can I claim through SR&ED?
There is no hard cap on total SR&ED expenditures. The enhanced 35% refundable rate applies to the first $4.5 million in qualifying expenditures for eligible CCPCs. Expenditures beyond $4.5 million earn the base 15% rate. When combined with provincial credits like Ontario's 8% ORDTC, a CCPC can recover 43% or more of eligible R&D spending. A three-developer team earning $100K each could generate approximately $129K in combined federal and Ontario refunds.
What documentation does CRA require for SR&ED claims?
CRA expects contemporaneous documentation that shows: the technological uncertainties you faced, the hypotheses you formed, the systematic experiments or analyses you performed, and the results (including failures). For AI projects, this includes model architecture decisions, training logs, performance metrics, experiment tracking records, and written descriptions of why known approaches were insufficient. Documentation should be created during the project, not retroactively at tax time.
Can I claim SR&ED and other grants at the same time?
Yes. SR&ED can be stacked with other programs, but you must reduce your SR&ED qualified expenditure pool by the amount of any government assistance received. For example, if IRAP funds $50K of your R&D salary costs, you subtract that $50K from your SR&ED eligible expenditures before calculating credits. SR&ED can be combined with IRAP, OIDMTC, and provincial programs when the eligible activities and costs do not overlap dollar-for-dollar.
How do I file an SR&ED claim?
SR&ED claims are filed with your corporate income tax return using Form T661 (Description of Scientific Research and Experimental Development) and Schedule T2SCH31 (Investment Tax Credit). The claim must be filed within 18 months of your fiscal year-end. Many companies engage SR&ED consultants who prepare the technical narratives and financial schedules, typically on a contingency fee basis of 15-25% of the credit recovered.
What happens if CRA audits my SR&ED claim?
CRA reviews a significant percentage of SR&ED claims. During a review, a Research and Technology Advisor (RTA) evaluates the technical merit of your claim while a Financial Reviewer examines the expenditures. Having strong contemporaneous documentation, clear project descriptions linking to technological uncertainty, and organized financial records is the best preparation. First-time claimants and claims above certain thresholds are more likely to be reviewed.
Disclaimer: This page is provided for informational purposes only and does not constitute tax, legal, or financial advice. SR&ED eligibility is determined solely by the Canada Revenue Agency (CRA) based on the specific facts of each claim. Program rules, rates, and eligibility criteria may change. The examples and calculations shown are simplified illustrations and may not reflect your specific situation. Always consult with a qualified SR&ED consultant or tax professional before filing a claim. Qyntral does not prepare or file SR&ED claims and does not guarantee eligibility or credit amounts. Information on this page was last reviewed in February 2026 and may not reflect subsequent changes to the program.