for the founder who thinks “we’re just a SaaS” —
SR&ED for Canadian IT & AI companies
You call it Tuesday. CRA calls it technological uncertainty.
You’re solving problems that don’t have known answers. Most of your dev work probably qualifies — and the experiments that failed count too.
The biggest field for SR&ED
~40%
of all SR&ED credits go to software development (per CRA, 2024–25)
Pre-seed to scale-up. First-time claimants welcome — CRA’s new pre-claim approval gives an official determination in 8 weeks, before you spend the budget.
What qualifies · what doesn’t
✓ Qualifies
- Evaluating several model architectures to find one that holds up in production
- Fine-tuning on proprietary data where the outcome was genuinely uncertain
- Driving inference latency or cost below a threshold no one knew was reachable
- Integrating a system in a way nobody had tried before
- Failed attempts at horizontal scaling — failure counts
✕ Doesn’t qualify
Implementing a well-documented libraryBuilding a standard feature with no uncertaintyA UI redesignRoutine maintenance
not sure? that’s usually the uncertainty test passing. ask.
The common mistake
Your last claim probably came out of a Jira export. Fast and cheap — and CRA’s new AI triage now flags machine-scraped patterns. We interview your engineers and write the technical story the way a reviewer reads it. Across the program, the large majority of SR&ED claims are accepted — most as filed. A clear, hand-written narrative is how you give yours the best chance — and how it holds up if anyone asks. (Program-wide rates, not a guarantee for any one claim.)
Client pattern · anonymized
Hard problems with no off-the-shelf answer: customizing large language models for reliable, task-specific coaching; extracting structured data from arbitrary PDFs; re-architecting a monolith toward microservices for real-time scaling. We wrote the uncertainty and the dead ends into a narrative a reviewer could follow.
A few hours of their team’s time. We did the rest.
Real work we’ve claimed
Sound familiar?
Almost every one of these started the same way: the standard, off-the-shelf approach didn’t work, and nobody could tell them why. Figuring it out — including all the tries that failed — is exactly what SR&ED is for.
A team grabbed a popular pretrained vision model to flag when a worker was getting stressed on camera. It demoed beautifully — then fell apart on real footage because of glare and reflections. The months they spent cleaning up the image and retraining the model? That’s the claim, not the demo.
An AI team was using machine learning to run an industrial process nobody fully understood — messy data, gaps where readings went missing, conditions that drifted day to day. The textbook approach assumes clean, steady data. Theirs wasn’t. Building something that could cope with the mess was the uncertainty worth claiming.
A software shop tried to catch small-dollar fraud in a mountain of insurance claims. The catch: they only knew what some of the fraud looked like, and every false alarm meant a person had to go investigate. Getting a model accurate enough to trust took a string of approaches that didn’t work before one did.
A developer needed to push real-time alerts the moment data changed, without bolting on fragile config files that drift out of sync over time. The “normal” ways either slowed the database to a crawl or created a maintenance nightmare. The work to find a clean way around that is recoverable.
A team was told a popular low-code platform could host a system for thousands of anonymous users. It couldn’t — it assumes one login per person. Proving that, then building something that actually held up, is the kind of dead-end-to-solution path SR&ED rewards.
A security team wanted to spot fake look-alike web domains within hours, not days. The standard lookup tools were too slow and too stale. They had to invent their own real-time way to catch the subtle misspellings — and that engineering is the claimable part.
Most of your dev work probably qualifies. Let’s find out.
Book a 30-minute call with Amin →