When an SR&ED claim is selected for review, a company will need documentation to support both the work performed and the expenditures claimed.
The problem is not always a lack of evidence. In many cases, the information exists, but it is spread across different systems, departments and file formats.
Technical records may be stored in Git repositories, Jira tickets, spreadsheets, test environments, meeting notes, emails or internal messaging platforms. The related financial information may be somewhere else entirely, such as accounting software, payroll systems, timesheets and project management tools.
Bringing all this together months after a project has ended can take a great deal of time. More importantly, it may be difficult to reconstruct a clear history of what happened.
This is one area where AI can be genuinely useful. It can help companies organize existing information throughout the year and make the supporting evidence easier to retrieve if the CRA decides to review a claim.
Organizing technical project records
AI-assisted tools can search and organize records such as:
- Project plans and technical requirements
- Design documents and architecture diagrams
- Git commits, pull requests and code review comments
- Jira, Azure DevOps or other project tickets
- Test plans, results and performance benchmarks
- Records of unsuccessful approaches
- Technical meeting notes and correspondence
- Version histories showing how the work developed
Once these records are grouped by project, activity and date, it becomes much easier to follow the progression of the work. The evidence can show what the team attempted, what results were observed and why a different approach was taken.
This is particularly important for projects involving several rounds of testing. The finished product does not always show the work that went into developing it. Earlier prototypes, failed tests and changes in direction may provide a much clearer picture of the experimental process.
Connecting time and labour records
Labour is often one of the largest expenditures in an SR&ED claim. It can also be difficult to support when employees record their time inconsistently or use descriptions that are too general.
Information may need to be gathered from:
- Timesheets and time-tracking systems
- Payroll records
- Project assignments
- Task management platforms
- Development and commit histories
- Engineering calendars
- Meeting records
- Employee work descriptions
AI can help compare these sources and bring potential issues to the company’s attention.
For example, it may identify time entries that fall outside the project period, employees assigned to a project without related activity, or technical work recorded in a project system but not reflected in the company’s time records.
This should not be used to recreate employee time after the fact. The real value is in identifying missing or inconsistent records while the information is still recent and can be verified.
Organizing materials, contracts and other costs
The same approach can be applied to financial evidence, including:
- Supplier invoices and purchase orders
- Records of materials used or consumed during testing
- Contractor agreements and statements of work
- Contractor invoices and payment records
- General ledger transactions
- Cost allocation schedules
An invoice may prove that a purchase was made, but it does not necessarily explain how that purchase relates to the experimental work.
Linking the invoice to a specific prototype, test, technical challenge or phase of the project creates a more complete record. It also makes the relationship between the technical work and the claimed expenditure easier to understand.
Finding documentation gaps early
Perhaps the most practical use of AI is finding gaps before they become a problem.
A company could use it to identify:
- Experiments without recorded results
- Test results that are not connected to a project activity
- Long periods with little supporting documentation
- Project expenses without related technical records
- Different project names being used across departments
- Missing invoices, approvals or contractor agreements
- Labour entries that require clarification
- Important technical decisions that were discussed but not recorded
Finding these issues early gives technical, financial and SR&ED teams time to resolve them through their normal documentation process.
It is much easier to ask an engineer about a test performed last week than one completed 18 months ago.
AI can organize evidence, but it cannot replace it
There is an important limit to what AI should do.
It can locate, classify and connect records that already exist. It should never be used to invent experiments, create unsupported activities or fill missing information with assumptions.
Any results produced with AI should be compared with the original records and reviewed by the employees who performed or supervised the work. Companies must also consider security, confidentiality, access controls and data residency before allowing an AI system to process technical, employee or financial information.
The CRA indicates that the strongest supporting evidence is dated and specific to the work performed. Good preparation therefore starts with reliable records created as the project progresses.
AI can make those records easier to manage. It can connect technical activities with labour and financial information, identify missing support and reduce the time required to respond to questions during a review.
It cannot compensate for poor documentation practices.
The best time to prepare for an SR&ED review is while the work is taking place, not after the review has started.
Is your SR&ED evidence organized, connected and ready to support your claim?