Key takeaway: SR&ED for agtech startups runs through the same two-part CRA test every software or hardware claim faces, but agriculture creates its own version of a familiar trap: a season of field trials looks like R&D, feels like R&D, and often isn't. A crop-disease detection model that has to work on a camera bouncing down a combine at harvest speed, or a navigation system that has to hold a line in mud a wheeled robot has never seen — that's real technological uncertainty. Running this year's known agronomic protocol on a new field, or confirming a sensor reads the same as last year's, is routine testing, and the CRA excludes it by name. Precision ag and controlled-environment farming sit inside a software-and-hardware SR&ED category that already earns 42.6% of all credits allowed nationally — the money is there for the engineering half of the business, not the farming half.
Agtech founders tend to land on one of two wrong instincts. Some assume a program built for labs and semiconductor fabs has no room for a company that ends its year with dirt on its boots. Others assume that because farming is unpredictable and every field behaves differently, the whole operation must be "research" by definition. Both instincts miss the actual rule, and the rule doesn't care whether your product ships in a shrink-wrapped box or rides on the back of a tractor.
What follows is where that line actually falls: which parts of a precision-ag, ag-robotics, or controlled-environment stack clear the CRA's bar, which parts look technical but are agronomy as usual, how the contractor and capital rules apply to a farm-tech company specifically, and where teams lose the evidence for work that genuinely qualified.
Does SR&ED cover agriculture technology
Yes, for the engineering and computer science inside an agtech product — not for growing crops. The CRA applies one two-part eligibility test across every industry: was there a technological uncertainty that standard practice in the field couldn't resolve, and did the company try to resolve it through systematic investigation rather than by iterating until something worked.
Engineering, computer programming, mathematical analysis, and testing are all named among the eight recognized categories of support work, and none of them cares whether the output runs in a data center or on a soil probe. A computer-vision model for weed identification gets evaluated exactly like a computer-vision model for anything else. What's specific to agtech is the exclusion list sitting right next to that test — because farming trips several of those exclusions in ways a typical SaaS company never encounters.
The exclusions agtech runs into that software rarely does
Three of the CRA's named exclusions do most of the damage to an agtech claim, and all three show up under names that sound like normal parts of the business. Quality control or routine testing is the big one: running a known measurement protocol across a new field, a new season, or a new crop variety to confirm expected results is testing against a known standard, not resolving an unknown. Market research and sales promotion is the second. It catches more agtech work than founders expect: a pilot deployment run mainly to generate a case study or close a customer, rather than to test an unresolved technical hypothesis, falls on the wrong side.
Commercial production is the third, and the one that ends the most arguments. Once a system is deployed to grow a paying customer's actual crop for actual revenue, using an approach the company already knows works, that operation is commercial production, whatever uncertainty it carried the first time. A vertical farm that's raising and selling produce on a system it already validated isn't doing R&D by continuing to farm — it did the R&D earlier, when the environmental-control approach itself was still in question.
None of this is agriculture-specific guidance from the CRA; there isn't a farming-sector bulletin. It's the same exclusion list — which also rules out prospecting, social-sciences research, and style changes — applied honestly to a sector where "we ran a trial in the field" describes both genuine experiments and completely routine agronomy.
Where the real technical uncertainty shows up
Real uncertainty in agtech clusters where the physical world refuses to behave like a lab, and the honest test is whether a competent engineer could have predicted the outcome from documented methods. In practice, that means a short list of recurring problem shapes.
Computer vision for crop and pest detection is one. The hard part isn't a model trained once on a clean, curated dataset; it's a model that has to generalize across lighting, occlusion from leaves, and motion blur from a camera mounted on moving equipment. Autonomous navigation for ag robotics is another: holding a line through mud, uneven furrows, and variable crop density is a different control problem than warehouse or road navigation, and there's no off-the-shelf solution that transfers cleanly. Then there's sensor fusion for soil or plant health, combining spectral, moisture, and mechanical sensor data into a reliable signal under field noise nobody's characterized for your specific crop and geography.
Controlled-environment agriculture adds its own set of problems: an environmental-control algorithm that has to hold temperature, humidity, CO2, and light spectrum within tight bands for a new crop or a new facility geometry, where the interactions between those variables aren't documented for your setup, is a real control-systems problem. Variable-rate application systems that calculate application rates in real time from live sensor input, rather than applying a pre-set map, face a similar genuinely unsolved computation. And on the biology side, crop genomics and phenotyping work — building models or lab methods to predict a trait from genetic or imaging data where the underlying relationship isn't established — is legitimate scientific research, distinct from and running in parallel to the sensor and software work.
Two agtech companies, two different claims
Picture two companies that would both describe their year as "R&D-heavy."
Company A sells a soil-moisture sensor built on an off-the-shelf sensing module, paired with a dashboard that displays readings against published crop-water-need tables. The team spends the season running the sensor across a dozen partner farms, tuning the dashboard's UI, and adjusting alert thresholds based on customer feedback. It's useful, well-built software, and the field deployments genuinely tested whether farmers found the product valuable. But the sensing technology works as documented, the water-need tables are published agronomic science the company adopted rather than developed, and the uncertainty the team resolved was product-market fit, not a technological unknown. That's a strong go-to-market case study and a thin SR&ED claim.
Company B is building an autonomous weeding robot that has to distinguish crop seedlings from weeds of the same species at the two-leaf stage, in variable light, moving fast enough to cover a commercial field in a workable window. No published model reliably separates the two at that growth stage under field conditions, so the team designs and tests several vision architectures and sensor combinations, documents where each one fails, and only converges on an approach after multiple seasons of structured experiments against a technical hypothesis, not a sales goal. Company B's claim has real technological uncertainty behind it, and it would look the same whether the crop were corn, soybeans, or carrots — the eligibility has nothing to do with which field it was tested in.
Most agtech companies do a mix of both, in the same year, sometimes in the same field trial. The claim has to separate the two honestly rather than treat the whole season's activity as one undifferentiated R&D program.
Contractors, field trials, and the 80% rule
Only 80% of what an agtech company pays an arm's-length contractor for SR&ED work enters the qualified expenditure pool. That's the same rule that applies to any software or hardware claim, in effect since 2012, whether the contractor is a software shop or an agronomy consulting firm running trials on the company's behalf.
Whether a given engagement counts as a contract payment at all turns on the substance of the arrangement: did the contract require specific experimental work with an uncertain outcome, or a fixed-scope field trial run against an already-known protocol? Who bore the risk if the approach didn't work? An agtech company paying a university research farm or an independent agronomist to run trials should get that scoping into the contract itself, because a reviewer looks at what was actually commissioned, not the invoice description. A contract for "run our standard trial protocol on your acreage" reads very differently from a contract for "help us determine whether this sensing approach can detect the target condition at all."
Hardware, sensors, and capital equipment
Most agtech SR&ED claims — like most software claims — run primarily on salaries, with the prescribed proxy amount adding a 55%-of-salary overhead allowance rather than requiring a receipt for every sensor and cable. That matters more here than in a typical SaaS claim, because agtech teams often assume hardware costs dominate and salaries are an afterthought; usually it's the reverse.
Capital equipment deserves specific attention, though. Eligible capital property acquired after December 15, 2024 is claimable again, at 40% refundable on the 35% rate. For a company that bought a prototype robotics platform, a specialized sensor rig, or greenhouse environmental-control hardware specifically to run experimental work — not to operate an existing, validated system commercially — that restored eligibility is real money that didn't exist for equipment bought a few years earlier.
What agtech companies get wrong about documentation
The most common failure isn't claiming a routine field trial as SR&ED outright — most teams know last season's protocol run again isn't research. It's losing the evidence for the work that genuinely was uncertain, because field seasons don't generate a paper trail the way a sprint retro does.
A field log that says "sensor readings were off in the north plot, adjusted calibration" doesn't capture what the team hypothesized was causing the discrepancy, which calibration approaches they tried and rejected, or why the eventual fix wasn't obvious from the sensor's documentation. And agriculture has a documentation problem software doesn't: the evidence window is seasonal. If a hypothesis about a vision model's failure mode isn't written down during this year's growing season, the next chance to test it is a year away, and by then the technical lead who ran the trial may have moved on. The CRA's own position is that a project doesn't need to succeed to qualify — an approach the team tried in the field and abandoned is evidence, provided someone recorded why it was tried and why it failed, not just that the season happened.
Who should think twice before claiming
Skip this, or claim narrowly, if the company's field activity is mostly running established agronomic protocols, applying published crop science, or validating a product with farmers rather than resolving an open technical question. A season of field trials that confirms what the team already expected is market validation, not R&D, no matter how much mud was involved.
The mirror mistake shows up just as often: agtech founders who assume a farming business can't possibly qualify for a program built for labs, and never look closely at what their engineering or data science team actually built. A company with a real computer-vision, robotics, sensor-fusion, or environmental-control problem at its core — one that took structured experimentation to solve, not just iteration — is exactly the kind of work this program funds, whether it ships on a tractor, in a greenhouse, or on a drone.
What SR&ED for agtech startups is actually worth
The rates don't shift because the R&D happens to end up in a field. A Canadian-controlled private corporation earns the enhanced 35% refundable rate on up to $6 million of qualifying expenditures a year, for tax years beginning after December 15, 2024, with the basic 15% rate applying above that — as much as $2.1 million a year at the enhanced rate, and current expenditures like salaries at the 35% rate are 100% refundable for most CCPCs, though excluded corporations receive 40% instead.
Provincial credits stack on top, but not additively: a company also claiming a provincial R&D credit sees its federal base reduced under the s.127(18) grind rule, so an Ontario agtech company's real combined rate lands closer to 40.2%, not a naive 43%, and the same non-additive math applies in every other province. The filing clock is identical to any other sector too — a corporation's SR&ED reporting deadline is 18 months after its fiscal year end, with no extension process, and a company mid-harvest is exactly the kind of team that lets a filing date slip while everyone's in the field.
That last part is where continuous documentation earns its keep, and it's the one piece of this that doesn't care whether the codebase controls a greenhouse or a payment system: evidence captured from the tools a team already uses — commits, tickets, field logs, lab notes — reviewed by a qualified independent SR&ED expert before anything is filed. Automation on our side handles the collecting, season after season; a named human still stands behind what gets submitted. It doesn't decide whether a navigation algorithm or a control system qualifies — that's still the two-part test — but it means the evidence from this year's field season is still there next spring, instead of living only in the memory of whoever ran the trial.
Frequently asked questions
Does running field trials automatically make agtech work eligible for SR&ED? No. A field trial that applies a known agronomic protocol or confirms an expected result is quality control or routine testing, which the CRA specifically excludes. A field trial only supports a claim when it's testing a genuine technical hypothesis whose outcome wasn't known in advance.
Is a computer-vision model for crop or pest detection SR&ED eligible? It can be, if the model has to solve a problem no published approach reliably handles under your field conditions — variable lighting, occlusion, motion, or a specific crop and pest combination. A model trained on a clean dataset using a standard, documented technique is applying existing technology, not advancing it.
Does using off-the-shelf sensors or an existing farm-robotics platform disqualify a company? Not automatically, but it raises the bar. Using a sensor or platform exactly as documented is adopting existing technology. Work that pushes past documented limits — a novel sensor-fusion approach, a control algorithm the platform's own guidance doesn't cover — can still qualify even on top of off-the-shelf hardware.
Does outsourcing field trials or research to an agronomist or research farm change what a company can claim? Yes, if the contractor is arm's length. Only 80% of that contract payment enters the qualified expenditure pool, and whether it counts as a contract payment at all depends on whether the work involved genuine experimental uncertainty, not the invoice description.
Does capital equipment like sensors or greenhouse control systems qualify? It can, if acquired after December 15, 2024 for experimental work rather than to run an already-validated commercial operation, per the CRA's restored capital-expenditure eligibility. Capital expenditures at the 35% rate are 40% refundable.
Is a vertical farm or greenhouse operation itself SR&ED eligible? The operation of a validated growing system to produce and sell a crop is commercial production, which the CRA excludes. The R&D happened earlier, when the environmental-control approach, growing methodology, or sensing system was still an open technical question — that work can qualify even if the resulting farm now runs profitably.
Agriculture doesn't get a softer version of the SR&ED test or a harder one. It gets the same two-part rule, applied honestly to the half of an agtech company's year that's genuinely engineering research, and not to the half that's running a farm.
See what your agtech engineering work could be worth — estimate your refund or check your eligibility.