AI is a great tool in helping prepare stronger SR&ED claims in less time. However, the idea that AI can prepare the entire SR&ED claim without any SR&ED expert oversight is not the current reality for most SR&ED claimants.
A technical SR&ED write-up needs two things: the data (what was built, when, by whom) and the story (why it was uncertain, what was tried, what failed). AI is genuinely excellent at pulling the first one out of your GitHub, Jira or Project Management tools. It’s still weak at the second, which is why the approach and ways that you use AI matter.
At GrowWise Partners, a Canadian AI-enabled SR&ED consulting firm, we see this gap constantly. Founders come to us after trying an AI-only tool, or even just Claude or ChatGPT, expecting a finished claim. What they usually get is a document that’s technically accurate and strategically wrong. AI can be an incredible tool to assist in preparing a SR&ED claim, but in 99% of cases, you still need a SR&ED expert to guide the process. If you want a straight read on where your own claim stands before you commit to an approach, book a free consultation, and we’ll tell you plainly.
The two halves of a technical write-up: the data and the story
Every SR&ED technical narrative rests on two different kinds of information, and they don’t come from the same place.
The data is what happened: which projects existed, who worked on them, the quantitative improvement seen during different experiments, what the code or lab notes show. This is exactly what AI is built for. Point it at a code repository or a project management tool, and it will surface relevant activity faster than any human could.
The story is why it mattered: what the technological uncertainty actually was, why standard approaches didn’t resolve it, what was tried and failed before something worked. That almost never lives in the documentation itself. Software teams especially tend to keep very little detail in their commits and terrible notes on their own reasoning. This overarching information about the goals, struggles, challenges and iterations often lives almost exclusively in team members’ memories, which is the core issue of AI gathering insights only from documents.
In my opinion, the AI does a great job at gathering the data but not a great job at gathering the story. That is why our approach at GrowWise is to do technical interviews plus AI-enabled document ingestion.
Three kinds of “AI-powered” SR&ED help, and what each one actually gives you
Not every company calling itself AI-powered is doing the same thing with that AI. It’s worth knowing which one you’re actually evaluating before you compare price or hours.
Fully automated, document-only tools. There are several fully AI-only SR&ED consultants these days that connect to your GitHub or Jira, generate a draft from what they find, and stop there. No interview, and often no human review before you see the output. Fast, but it only ever gets you the data half of the write-up. These solutions typically promise the SR&ED claim will be complete in an hour and cost much less than traditional firms. That is because they are only looking at the documentation, not the story of the work.
Consultant-side “AI-enabled” firms. A growing number of traditional firms have added AI somewhere in their internal process, usually to help their own consultants classify data or draft narrative outlines faster. That’s a real efficiency gain for the firm. What it doesn’t automatically do is change what happens on your side of the table. These companies like to boast that they have a lot of AI, when in reality the process for you as a client is very similar to that when working with a traditional accounting firm: it’s manual, hours of meetings talking to the consultants explaining the work you do, preparing documentation for the consultants, and it takes a ton of your team’s time. Being AI-enabled internally isn’t the same as being AI-forward for the client, and it’s worth asking a firm directly how much of their advertised AI actually benefits you, versus how much only speeds up their own team.
Dual-AI platforms with human review. This is where GrowWise sits. AI handles document discovery across your tools, AI also runs the technical interview directly with your engineers, and a senior consultant reviews, provides expertise, and signs off before anything is filed. The AI reaches all the way to your team, not just the firm’s internal process. Our process focuses on using AI to expedite the information-gathering, but sticking to human expertise to do the parts of the process that require SR&ED knowledge, such as scoping projects, writing the technical narratives and reviewing everything.
Claude/ChatGPT AI Models: There is also the option of using tools like Claude or ChatGPT to assist in preparing your SR&ED claim. In some cases, if, for example, Claude is already deeply connected to the work your company does, you use Claude Code, and it has access to all of your information, it likely can support in preparing a decent first draft of your SR&ED claim. Again, where these tools typically fall short is in positioning projects in a way the CRA is looking for, and ensuring the narratives contain more than just the quantitative analysis side, but also the story and overarching narrative of the projects. The issue is that there are very few publicly posted examples of what a “good” SR&ED technical narrative looks like, so AI models do not have much data to use to understand how they should be positioned, structured or worded. Without much data, these AI models tend to not produce very high-quality outputs.
Can AI actually prepare an SR&ED claim on its own?
Short answer: no, not reliably, and definitely not in the way some new AI-only tools promise.
Some newer, AI-only SR&ED companies advertise claims prepared in as little as an hour. That’s not impossible, but it only happens under a very narrow set of conditions: one project, meticulous time tracking done by the company throughout the whole year, and simple financials with a single employee and no contractors, capital expenses, or materials to sort out. The moment a claim gets even slightly more complicated than that, it needs a person reviewing and shaping it, not just software extracting from it.
The bigger issue isn’t speed. It’s positioning. AI tools tend to frame work around business challenges or product features rather than the underlying technical uncertainty the CRA is actually looking for. We’ve had founders come to us asking for a review of a claim an AI tool produced, and the honest answer was that we needed to start over, because the tool never scoped the projects correctly in the first place. It didn’t know when a large project should be split into two, or when two smaller ones actually belonged together under one core technical challenge.
This isn’t limited to fully automated tools, either. Some AI-enabled consulting firms use AI mainly to speed up their own consultants’ internal prep work, rather than as something a client interacts with directly. That’s a legitimate use of AI, but it doesn’t solve the core challenge of SR&ED: companies want a solution that is trustworthy, reliable and doesn’t take much time from their team.
Does the right approach change by industry?
Short answer: yes, and it comes down to how naturally your industry documents its own reasoning.
Well-documented, research-heavy industries like biotech and medtech tend to have an easier time with AI-forward tools. Lab notes typically lay out the experiment, the hypothesis, and the result in a way that already resembles a technical narrative, so there’s more of the “story” sitting in the documentation to begin with.
Software companies are the opposite case. A code repository shows what was built, not why it was hard. This is usually where human-led scoping matters most from the very start, identifying which projects are even worth writing up before anyone touches the technical narrative.
How much time should SR&ED prep actually take?
Short answer: Some AI-only tools out there claim to prepare full claims in 1-3 hours, but when it comes to SR&ED, you need to consider quality and quantity (hours required)
Model | Typical time from your team |
AI-only tools | Claiming 1 to 3 hours |
GrowWise | 3 to 6 hours |
AI-enabled full-service firms | ~10 hours |
Traditional consultants | 20 to 40 hours |
Almost every AI-only SR&ED consulting platform today advertises a time commitment in the same few-hour range. On its own, that number doesn’t tell you much, because a firm can land at five hours by asking fewer questions and preparing a lower-quality SR&ED claim.
A shorter process does not necessarily mean less of a headache; if the shorter process results in a low-quality claim that ultimately gets audited, you’ll end up spending significantly more time on SR&ED. At GrowWise, our goal is to make the process as simple and quick for our clients while still balancing the need for detailed information and details to create a strong, high-quality claim that will withstand CRA scrutiny.
Is a document-only AI tool ever the right call?
Short answer: sometimes, but only for small, well-documented claims, and you should still get an expert to check the work.
You should always want someone who actually knows SR&ED involved in your process. If you’re using a self-serve AI tool, pay a reviewer to look at what it produces before you file. That said, a document-only tool can be a reasonable fit if:
- Your expected claim is around $20,000 or less
- Your documentation is genuinely detailed, not just present
- You have one project, not several that might need to be split or combined
- Your financials are simple: one or two technical team members, no contractors, no capital purchases, no materials to allocate
Outside of that, the risk shifts to you. A thin claim that gets challenged on review costs a lot more time than the hours an AI-only tool saved you up front.
The GrowWise take
AI is genuinely useful in preparing SR&ED claims, there is no doubt about it. Some of our clients send us their entire code repository, and our tools dig through all of it to extract exactly what’s relevant to a specific project. A few years ago, that either wouldn’t have happened at all, or it would have taken a consultant days. That part of the process has gotten dramatically better.
Where the market oversells it is the claim that AI can do the whole job. If a company is telling you your claim will be ready in an hour, ask what your documentation would need to look like for that to be true. For almost everyone, the honest answer is that it wouldn’t be, and a claim built that fast is usually a claim built shallow, which is exactly the kind that draws CRA attention.
One day we see AI being useful as a complete AI agent, living inside all of your data, extracting SR&ED-relevant details as they happen and preparing extremely well-documented, audit-proof SR&ED claims. We are working on building that future state, and see huge applications for AI to continue benefiting the simplification of preparing SR&ED claims, but in the meantime we aren’t quite there yet. SR&ED still requires human oversight, but AI is incredible at facilitating information digestion to expedite the aspects that humans still do best.
Quick-reference FAQ
Does using AI in my SR&ED prep increase my audit risk? Not on its own. Risk comes from whether the claim actually reflects a genuine technological uncertainty, not from whether AI touched the documentation along the way. A shallow claim, AI-assisted or not, is the real risk factor.
Can I switch from an AI-only tool to a hybrid platform partway through a claim year? Yes. Most of what an AI-only tool has already gathered, your commits, tickets, and time data, carries over. What usually needs rebuilding is the project scoping and the narrative itself.
What should I ask an “AI-powered” SR&ED company before signing up? Ask directly whether their AI is something you interact with (client-facing) or something that only speeds up their internal team, and ask who reviews the output before it’s filed.
Where this leaves you
This post covers general patterns. Your own claim depends on how your team actually documents its work, and that’s worth a real look rather than a guess. Every post here covers general rules, and your specific situation may differ, which is exactly what a free call is for.
We don’t charge anything upfront, and you pay only when CRA pays you. Want to learn more about our AI-enabled approach at GrowWise? Book a call with me.
