Key takeaway: SR&ED for adtech startups runs through the same two-part CRA test as any software claim, but the excluded category that trips up adtech isn't the one that trips up everyone else. Building the algorithm that decides which ad wins an auction in under 100 milliseconds can qualify. Running the A/B test that proves a new headline gets more clicks cannot — the CRA excludes "market research or sales promotion" by name, and adtech's whole business model sits right on top of that line. Software already takes 42.6% of all SR&ED credits allowed in a given year, and adtech engineering claims live inside that share — the opportunity is real, but so is the risk of claiming the half of the business that's actually marketing work wearing an engineering hat.
Ask an adtech founder what their engineers spend the year building, and you'll usually hear some version of "the whole stack is R&D — we're constantly testing, measuring, optimizing." That sentence is true and it's also exactly why adtech claims get into trouble. Testing and optimizing describe almost everything a marketing-technology company does, including the parts the CRA has named, specifically, as not SR&ED.
Here's the actual line: what kind of adtech engineering clears the two-part test, why "market research or sales promotion" catches more of this industry than most, where the real technological uncertainty tends to hide, and how the contractor and proxy rules apply once you've sorted the two apart.
Does SR&ED cover adtech and martech companies
Yes, but only for the work that resolves a genuine technological uncertainty — not for the work that measures a marketing outcome. The CRA applies the same two-part eligibility test to an ad-serving platform that it applies to a payments API or a game engine: was there uncertainty a competent professional in the field couldn't resolve using standard practice, and did the company try to resolve it through systematic investigation.
Computer programming and mathematical analysis both sit inside the eight recognized categories of support work, alongside engineering, design, operations research, data collection, testing, and psychological research. A bid-optimization engine or a fraud-detection model gets judged by the same standard as any other piece of software. What's different about adtech isn't the test. It's how much of the day-to-day work sits inside a category the CRA excludes outright, and how easy it is not to notice.
The exclusion that matters most for adtech
"Market research or sales promotion" is one of the CRA's named exclusions from eligible SR&ED work, and for an adtech or martech company it isn't a corner case — it's the center of the business. Measuring whether an ad performs, comparing one creative against another, surveying users about ad relevance, building a dashboard that shows a client their campaign metrics: all of that is market research or sales promotion, however much engineering effort went into the pipeline that produces the numbers.
The trap is specific to this industry. A fintech company building a fraud model doesn't usually confuse that work with sales promotion. An adtech company building an attribution model can, because the model's whole purpose is to answer a market-research question — did this ad work — even when building the model itself required real algorithmic uncertainty. The CRA's line isn't about how the output gets used downstream. It's about whether the activity you're claiming is the technical work of building a novel measurement system, or the act of using an existing system to measure a campaign. Those can happen in the same codebase, in the same sprint, done by the same engineer, and still land on opposite sides of the exclusion.
Two more named exclusions show up constantly in adtech claims and get missed just as often: "quality control or routine testing" catches A/B testing ad creative or landing pages once the testing method itself is standard (you're not developing a new statistical technique, you're running one), and "routine data collection" catches the ordinary work of pulling impression, click, and conversion data into a warehouse. Collecting the data is infrastructure work. Deciding what a novel model does with that data, under constraints nobody's published a solution for, is where SR&ED can start.
Where the real engineering uncertainty shows up
Real technological uncertainty in adtech clusters around problems that got harder, not easier, over the last few years — mostly because the data engineers used to lean on stopped being available. The test is the same one that applies everywhere: could a competent engineer elsewhere in the field have predicted the outcome before you ran the experiment, using published methods? In adtech, the honest "no" answers cluster in a few recurring places.
Take privacy-preserving measurement first, because it's the biggest one right now. An attribution or conversion-measurement system that has to hit a defined accuracy target without third-party cookies or persistent device identifiers — using differential privacy, on-device aggregation, or clean-room-style joins across parties that can't see each other's raw data — isn't wiring up a documented library. In most implementations, nobody's published a solution that actually hits the accuracy bar yet.
What about deciding which ad wins an auction? That's a fixed-latency problem, often under 100 milliseconds, with a constantly-updating model in the loop and real money lost on either side of a bad decision — under-bid and you lose the impression, over-bid and you erode margin. No off-the-shelf answer exists at scale for every bidding scenario a platform runs into.
Fraud and invalid-traffic detection is the grinding, unglamorous version of the same problem: a model that has to catch novel bot and click-fraud patterns specific to your traffic, without a false-positive rate that throttles real users. It's never solved once. The adversary adapts and the model has to adapt back.
And then there's identity. Probabilistic identity resolution — inferring that two signals likely belong to the same person, at a confidence level you can defend, once the old approach of a shared cookie or login no longer exists — is a real statistical and engineering challenge, not a lookup. None of these four is eligible because it sounds technically impressive on a pitch deck. They're eligible candidates because nobody could have told you the outcome in advance from documented methods, and someone tested that systematically instead of guessing.
This mapping of the general CRA test onto adtech-specific problems is our read on how the rule applies to this industry, not a published adtech bulletin — there isn't one. The CRA sets a technology-neutral standard and expects it applied honestly to whatever a company's engineers actually attempted.
Two adtech companies, two different claims
Picture two Canadian ad-tech companies, both selling to the same kind of client and both convinced their engineering team does serious R&D.
Company A built a campaign-analytics dashboard on top of a standard cloud data warehouse. The team spends most of its year running structured A/B tests for clients — comparing ad creative, testing subject lines, measuring which landing-page layout converts better — and building increasingly polished visualizations of the results using well-documented BI tooling. The work is genuinely useful to clients, and the team iterates constantly. But nothing about the underlying technology was ever in question; the warehouse, the testing methodology, and the visualization stack all behave exactly as documented. This is market research and sales promotion delivered through software, and it's excluded regardless of how much engineering polish surrounds it.
Company B is losing the signal it used to measure attribution because the browsers and platforms it depends on have shut off the identifiers it used to rely on. The team has to build a new measurement approach that estimates campaign lift using aggregated, privacy-preserving signals instead — with no established method that hits the accuracy bar the business needs at the volume it operates at. They test several statistical approaches, most of which don't clear the accuracy target, before landing on one that does, and they document why the obvious first attempts failed. That's a genuine technical hypothesis, a series of experiments, and a result nobody could have predicted going in. Company B's claim has real substance behind it — and it would look the same whether the client vertical were retail, travel, or gaming, because the eligibility has nothing to do with which industry buys the measurement.
Most adtech companies are some mix of both. The claim only survives if the technical half gets separated from the marketing-services half honestly, project by project, rather than described as one undifferentiated "R&D team."
Contractors, agencies, and the 80% rule
Only 80% of what an adtech company pays an arm's-length contractor for genuine SR&ED work enters the qualified expenditure pool — the same rule that's applied to every SR&ED claim since 2012, whether the contractor is a specialist machine-learning consultancy or an offshore development shop. That 80% figure doesn't move because the invoice comes from a media agency instead of a dev shop.
Whether a payment counts as a contract payment at all turns on the substance of the engagement, not its label: did the contract require specific experimental work, did a fixed price shift technical risk onto the contractor, who owns what gets built, and is this a contract for services or effectively a purchase of a finished tool. An adtech company that pays an agency purely for campaign management and creative testing isn't buying SR&ED at any percentage — that spend was never eligible in the first place, contractor rule or not. The 80% haircut only applies to spend that would have qualified had the company done it in-house.
Where the proxy method and capital rules fit
Most adtech SR&ED claims run almost entirely on salaries — the engineers building the bidding, attribution, or fraud-detection systems — with the CRA's prescribed proxy amount covering overhead instead of requiring a company to itemize every cloud bill and SaaS subscription. That proxy has been 55% of the SR&ED salary base since 2014, and for a company running mostly on cloud infrastructure with a lean team, it typically outweighs the effort of tracking actual overhead expense by expense.
Capital equipment is a smaller factor for most adtech companies than for a hardware or biotech claim, since the infrastructure is usually rented cloud compute rather than owned hardware. It isn't zero, though: a company running its own GPU cluster for model training rather than a cloud provider's managed service, and buying that equipment after December 15, 2024, gets the benefit of capital expenditures being eligible again, at 40% refundable on the 35% rate. For most SaaS-model adtech businesses, though, the claim's size comes down almost entirely to how the salary pool gets split between eligible engineering and excluded measurement-and-optimization work.
What adtech companies get wrong about documentation
The most common failure isn't claiming the excluded work outright — most teams know a client-facing dashboard isn't a research project. It's failing to separate the technical uncertainty from the client-facing outcome inside the same project, because adtech teams document for the client, not for a tax claim.
A ticket that says "improve attribution accuracy for Client X" doesn't show whether the team was applying a known method to a specific client's data (routine, not eligible) or developing a new method because no known one hit the required accuracy under a new constraint (potentially eligible). A sprint retro captures what shipped, not which approaches were tried and discarded before the team found one that worked, or why the obvious first attempt failed. By the time someone sits down to build a claim, the distinction between "we optimized this campaign" and "we built a new way to measure campaigns without the data we used to have" has usually blurred back together in the write-up, even when it was clear at the time to the engineer who lived through it.
Who should think twice before claiming
Skip this, or claim narrowly, if most of the engineering effort goes into running structured tests for clients using established statistical methods and well-documented tooling, however sophisticated the reporting layer looks. Measuring a campaign is not researching a technology, no matter how much infrastructure sits underneath the measurement.
The mirror mistake shows up just as often: adtech companies that assume a marketing-adjacent business can't possibly qualify for an R&D tax program, and never look closely at what their platform engineering team actually built. A company solving a real measurement, matching, or bid-optimization problem under constraints nobody's published a solution for is doing exactly the kind of work this program exists to fund — independent of whether the client relationship on top of it is a marketing service.
What SR&ED for adtech startups is actually worth
The rates don't change because the engineering happens to sit inside a marketing-adjacent business. A Canadian-controlled private corporation earns the enhanced 35% refundable rate on up to $6 million in qualifying expenditures a year, for tax years beginning after December 15, 2024 — up to $2.1 million at the full limit. Current expenditures like salaries at that rate are 100% refundable up to the limit for most CCPCs, cash back even against zero tax owing, though excluded corporations receive 40% instead.
Adtech companies with a permanent establishment outside Ontario, BC, Quebec, or Alberta should check their own province's rate before assuming none applies — but wherever a provincial credit does apply, it isn't additive with the federal rate. The s.127(18) grind rule reduces the federal expenditure base by the provincial credit received, so a claimed combined rate needs the real math behind it, not a straight sum.
The clock runs the same as any other software claim too: a corporation's SR&ED reporting deadline is 18 months after its fiscal year end, with no extension available under any circumstance. An adtech company mid-fundraise or mid-integration with a new ad exchange is exactly the kind of team that lets that date slip while everyone's focused elsewhere.
That's the part of the job that has nothing to do with which industry the codebase serves: continuous documentation, captured from the tools engineers already use, reviewed by a qualified independent SR&ED expert before anything is filed. Automation on our side handles the collecting; a named human still stands behind what goes in front of the CRA. It won't decide whether a given attribution model clears the bar — that's still the two-part test, applied project by project — but it means the distinction between the measurement engineering and the marketing outcome gets captured while the engineer who built it still remembers which parts were genuinely uncertain.
Frequently asked questions
Does building ad-tech software automatically qualify for SR&ED? No. The test is whether the work resolves a genuine technological uncertainty through systematic investigation, per the CRA's eligibility guidance — not whether the software happens to serve advertisers. A company can run a technically ordinary ad platform with little eligible work, or build one genuinely novel measurement system inside an otherwise routine product.
Is A/B testing ad creative or landing pages eligible for SR&ED? Generally no. The CRA names "market research or sales promotion" and "quality control or routine testing" as excluded activities, and running a standard statistical test to see which creative performs better falls squarely into both, however much infrastructure supports it.
Can building a campaign-reporting dashboard ever count as SR&ED? Rarely, on its own. Pulling and visualizing performance data with documented tools and methods is "routine data collection," a named exclusion. It can be different if the underlying measurement problem itself — for instance, estimating attribution without the identifiers you used to rely on — required genuinely new methods, but the dashboard is the output, not the eligible work.
Does losing third-party cookies make measurement work SR&ED-eligible? Not automatically, but it's exactly the kind of constraint that can create real eligibility. Building a privacy-preserving measurement system that has to hit a defined accuracy target without the signals engineers previously relied on is a genuine technical problem if no established method gets there yet — the loss of the old approach is what creates the uncertainty.
Does outsourcing engineering or data science work change what an adtech company can claim? Yes, if the contractor is arm's length and the work itself would have been eligible in-house. Only 80% of that contract payment enters the qualified expenditure pool. Paying an agency purely for campaign execution or creative testing isn't SR&ED spend at any percentage, since that work was never eligible to begin with.
Do capital costs like GPU hardware matter for an adtech claim? Only for companies running their own hardware rather than cloud infrastructure. Eligible capital property acquired after December 15, 2024 is claimable again, at 40% refundable on the 35% rate — relevant mainly to a company that bought its own training cluster rather than renting compute from a cloud provider.
Adtech doesn't get a harder or easier SR&ED test than any other software vertical. It gets an industry where more of the day-to-day work happens to sit inside a named exclusion, and where separating the engineering from the marketing outcome takes more discipline than it does almost anywhere else.
See what your platform's engineering work could be worth — estimate your refund or check your eligibility.