A 2026 walk-through for Canadian founders who want to keep the full credit. T661, the new $6M expenditure limit, the technological-uncertainty narrative, and the math on when a consultant actually pays for itself.
Estimate your SR&ED credit →Direct answer: Yes, you can file Canada's Scientific Research and Experimental Development (SR&ED) tax credit yourself. The CRA does not require a consultant. For a Canadian-controlled private corporation (CCPC), the first $6 million of qualifying expenditures earns a 35% refundable federal credit — up to $2.1 million per year — and the underlying claim (Form T661 plus Schedule 31) is well within reach for a founder who can articulate the technological uncertainty in plain prose and reconstruct hours from existing time-records. DIY makes the most sense below roughly $300K in qualifying expenditures, where typical consultant contingency fees of 15-25% would consume a meaningful share of the credit. Above that threshold, a senior consultant frequently pays for themselves on technical-narrative quality alone.
Most founders meet SR&ED for the first time when an accountant offhandedly mentions it, or a consultant cold-calls offering to "find money" on contingency. Both framings obscure what the program actually does. SR&ED is not a grant. It is a refundable tax credit, defined in the Income Tax Act Section 248(1) and administered by the Canada Revenue Agency. You file it as part of your corporate tax return — specifically, on Form T661 (the SR&ED claim form) and Schedule T2 SCH 31 (the investment tax credit calculation). The CRA does not "award" SR&ED. You compute the credit, claim it on your return, and the CRA either accepts the claim or reviews it.
The reason it pays out cash even when your corporation is losing money is the refundable portion of the investment tax credit. For a CCPC earning the enhanced 35% rate, the credit is 100% refundable on qualifying current expenditures — meaning the CRA writes you a cheque if you have no tax owing. That's why it's the lifeblood of pre-revenue Canadian deep-tech and SaaS companies, where cash is scarcer than runway-extension stories.
Source: CRA — Scientific Research and Experimental Development (SR&ED) tax incentiveTo qualify, work must satisfy three criteria simultaneously, drawn from CRA Information Circular IC86-4R3 and the legislative definition. Pre-2021 audit guidance referred to "five questions"; the current framework is three:
The single biggest source of disallowed claims is work that is genuinely difficult but not technologically uncertain. CRA's Eligibility of Work Policy is explicit on this. The following are common DIY pitfalls:
The flip side: many founders systematically under-claim because they don't recognize qualifying work. If your team built a custom machine-learning approach because off-the-shelf models didn't converge on your data, that's almost certainly SR&ED. If you wrote a novel cryptographic protocol because none of the published ones met your performance constraints, that's SR&ED. If you built a bioreactor controller because the published control schemes didn't handle your culture's metabolic profile, that's SR&ED. The test is not difficulty — it's whether the publicly available knowledge of the field could give you a clear path forward.
Here's what you need to know about the SR&ED eligibility test in practical terms: it is fundamentally a writing test, not a technical-merit test. The CRA's Research and Technology Advisors (RTAs) are competent technical reviewers, but they don't audit your code or your lab results — they audit your description of them. A weak narrative around strong technical work loses claims; a strong narrative around modest qualifying work wins them. Most rejected first-time DIY claims fail not because the work didn't qualify, but because the founder described it in commercial terms ("we built X feature") instead of investigation terms ("we faced uncertainty Y, we hypothesised Z, we tested it via experiment Q, we learned R").
The 2025 federal budget — tabled by the previous government and largely confirmed in 2026 implementation — introduced four substantial SR&ED amendments. All apply to tax years beginning on or after December 16, 2024, which means most fiscal-2025 returns filing in 2026 are the first cohort to benefit.
This is the headline change. Before Budget 2025, a CCPC's enhanced-rate expenditure limit was $3 million per year, capped further by associated-group rules. Budget 2025 raised the limit directly from $3M to $6M — doubling the maximum federal refundable credit at the 35% rate from $1.05M to $2.1M per year.
Mechanically, this means a CCPC with $6M of qualifying current expenditures (effectively $3.87M of salaries plus the proxy 55% loading) earns the 35% refundable rate on the full amount. Before the change, expenditures from $3M-$6M earned only the 15% non-refundable base rate, creating a sharp cliff that distorted hiring decisions at scale-up companies. The new structure closes that cliff.
Source: Government of Canada — Budget 2025, Chapter 2: A Strong EconomyFrom 2014 to 2024, capital expenditures were excluded from SR&ED — you could deduct them from corporate tax in the normal way, but they earned no investment tax credit. Budget 2025 reversed this for tax years beginning on or after December 16, 2024. Capital used "primarily" (more than 50%) for SR&ED activities qualifies for the credit at the same rates as current expenditures.
For deep-tech companies running wet labs, fabrication shops, or specialised computing infrastructure, this is meaningful. A $400K bioreactor used 80% for qualifying work now generates a refundable credit of roughly $112K for a CCPC under the limit, where previously it generated zero. Document the SR&ED-use percentage with usage logs from day one — this will be a frequent audit target in the first few years.
The enhanced 35% rate phases out as a CCPC's prior-year gross revenue rises. Budget 2025 widened the phase-out window from the old $10M-$50M range to $15M-$75M. A CCPC with $30M prior-year revenue still qualifies for partial enhanced rate, where it would have been fully phased out under the old rules. This is a meaningful win for revenue-positive Canadian scale-ups that were previously punished for growing.
For the first time, certain Canadian public corporations can access the enhanced refundable rate — not the full 35%, but a partial benefit on a defined expenditure window. The detail here is technical and matters mostly for late-stage Canadian public companies; for the typical founder reading this guide, it is background.
The best path for a CCPC in 2026 is to file SR&ED yourself if your qualifying expenditures are under $300K, because the new $6M ceiling is irrelevant at that scale and consultant contingency fees of 15-25% would consume more than the marginal narrative quality is worth.
Because: at $200K of expenditures, a 35% refundable credit equals $70K. A 20% contingency takes $14K of that. The same hours spent on a careful narrative typically deliver the full credit, especially for first-year claims where the reviewer expects a less-polished file from a genuinely first-time claimant.
A handful of items widely reported as "Budget 2025 SR&ED changes" did not actually change. The base rate for non-CCPCs remains 15%, non-refundable. The proxy method still applies a flat 55% overhead loading on qualifying salaries. The 12-month filing deadline (claiming SR&ED late will not be considered) is unchanged. The technological-uncertainty test in IC86-4R3 is unchanged. Patent box and IP regime announcements from Budget 2025 are separate from SR&ED — they are a forthcoming proposal, not part of the current credit.
SR&ED consultants typically charge in one of three ways: pure contingency (15-25% of the credit, paid only if you receive the refund), hourly rates ($175-$350/hour), or hybrid models with a smaller contingency plus a fixed engagement fee. The headline question every founder asks — "Should I hire one?" — reduces, on the math, to whether the consultant's marginal value over your DIY effort exceeds their fee.
That marginal value lives in three places: (1) narrative quality, which raises the share of claimed expenditures that actually survive a review; (2) calculation accuracy, which prevents over-claims that get reduced and under-claims that leave money on the table; and (3) audit defence experience, which matters in roughly one in three first-year claims and one in five subsequent claims.
| Qualifying expenditures | 35% federal credit | Consultant cost (20% contingency) | Net to founder (consultant) vs. DIY |
|---|---|---|---|
| $100,000 | $35,000 | $7,000 | $28,000 vs. $35,000 |
| $300,000 | $105,000 | $21,000 | $84,000 vs. $105,000 |
| $1,000,000 | $350,000 | $70,000 | $280,000 vs. $350,000 (if DIY survives review) |
The naive read of that table says always DIY. The reality is more nuanced. The "DIY" column assumes your DIY claim survives review at the same dollar value the consultant would have produced. In practice, first-time DIY claims get partial reductions in 30-40% of reviews, often losing 20-50% of the claimed amount. A weak narrative, a missed expenditure category, or a poorly allocated salary can shift the comparison.
Below $200K of qualifying expenditures. DIY almost always wins. The contingency fee on a $70K credit is $14K — that's 80-100 hours of your time at the founder hourly equivalent. Spend the time. Read CRA's T4088 guide carefully. Get the narrative right. The marginal review survival rate from a consultant rarely justifies the fee at this size.
$200K-$500K. The toss-up zone. DIY makes sense if you have a technically literate co-founder, an organized fiscal year, and the bandwidth for 60-100 hours of focused work. A consultant makes sense if you're stretched thin, have multiple complex projects, or know a specific consultant with deep domain expertise in your area. Hybrid arrangements (consultant reviews your DIY draft for $3K-$8K) can capture most of the marginal benefit at a fraction of the contingency cost.
Above $500K. Consultant usually wins. At this scale, the cost of a partial reduction in review (20-50% of a $175K+ credit) outweighs the contingency fee. Senior SR&ED consultants who have survived dozens of reviews materially raise the share of expenditures that hold up. Negotiate hard on rate — 12-15% contingency is achievable at this size for clean claims, especially in subsequent years.
The best path for a sub-$200K claim is DIY, because the contingency fee is roughly equal to 100 hours of focused founder time and the marginal review-survival benefit doesn't repay it.
Because: at this claim size, the work involves one or two projects, a manageable number of employees to allocate, and a narrative that fits comfortably in the T661 box budget. CRA reviewers also bring more grace to first-year claims at this scale than to large ones.
We fundamentally believe that our economic resilience begins and ends with the Canadian entrepreneur. Their grit, their dedication, their innovative thinking and their willingness to take risks have built our country and continue to build it. Isabelle Hudon — President & CEO, Business Development Bank of Canada
Most founders underestimate how much a 20% contingency fee compounds over time. A company claiming $400K in qualifying expenditures every year for five years pays the consultant $140K in cumulative fees on $700K of refunds. That's $140K in dilution-free founder cash that could have funded a hire or extended runway. At year three, the templates are mature, the time-tracking is in place, and the consultant's marginal value has dropped sharply — but the contingency rate often does not. Renegotiate or switch to a fixed-fee arrangement once the work has stabilised.
A second hidden cost: many consultants have an institutional preference for over-claiming, because their fee scales with the claim and contingency contracts protect them from claw-back risk. If 15% of the claim is reduced in review, the founder bears 100% of the reduction; the consultant simply earns less on the lower amount. Read your engagement letter carefully on what happens to the consultant's fee in a partial reduction.
Here's what you need to know about consultant value: the most useful thing a senior consultant brings is not the narrative itself, but the calibration of what to claim and what to leave out. First-time DIY claimants tend either to under-claim (missing legitimate qualifying work in fear of audit) or over-claim (sweeping in non-qualifying integration work). A good consultant is a third filter that pulls those errors back to the centre. If you DIY, build that calibration deliberately — read CRA's published policy positions, study the SR&ED case law summaries, and ask your accountant to second-eyes the claimed-vs-non-claimed split.
The DIY-vs-consultant decision rarely turns on a single number. It depends on company stage, claim complexity, founder bandwidth, and the geography of your R&D team. The five profiles below cover the most common situations.
You're a technical founder, probably a sole developer or with one contractor, and you've spent 8-14 months building a product that involves real technical experimentation — novel ML, a custom protocol, a hardware-software integration where the off-the-shelf path didn't exist. Your fiscal-year salary equivalent (founder salary plus contractor invoices) is $80K-$180K. Your qualifying expenditures, after the proxy 55% loading, are $120K-$280K. Your refundable credit is $40K-$100K.
Path: File DIY. Read CRA's T4088 guide end-to-end. Spend a week reconstructing time records, a week drafting the narrative, and a few days finalising calculations and the form. Total: 60-90 hours, mostly evenings and weekends. The marginal value of a consultant at this scale is 15-25% of $40K-$100K, which doesn't pay for the loss of founder time spent briefing them. Use a tax preparer who has filed SR&ED before to review the final T661 — that's $1,500-$3,000 for a couple of hours of expert eyes, and it materially reduces the chance of a calculation error.
The company is at $2M-$10M ARR. Engineering team of 15-30. Multiple SR&ED-relevant projects per year — a novel multi-tenant architecture, a custom search ranking model, a new on-device sync protocol. Annual qualifying expenditures fall in the $700K-$2.5M range. Refundable credit, depending on CCPC status and gross-revenue phase-out, is $200K-$700K per year.
Path: Hybrid. Build the time-tracking and project documentation infrastructure in-house — this is a permanent operating function, not a once-a-year scramble. Engage a senior SR&ED consultant on a fixed-fee or capped-contingency arrangement to write the technical narratives and stress-test the calculations. Pay $15K-$40K for the engagement; never a 20% contingency on a $400K claim. The consultant's value is concentrated in narrative craft and review preparation, both of which fit comfortably in fixed-fee scoping.
You're building robotics, biotech, climate tech, novel semiconductors, or quantum hardware. Your team is mostly research-trained PhDs and senior engineers. Qualifying expenditures are typically $500K-$2M with significant capital expenditures (lab equipment, fabrication tooling, specialised compute) now eligible under Budget 2025. Refundable credit is $175K-$700K.
Path: Consultant, but choose carefully. A generalist SR&ED consultant will struggle with deep-tech narrative because the technological uncertainty involves field-specific experimental design that a non-domain reader can't articulate. Pay extra for a domain-aligned firm or an independent consultant who has previously filed in your specific area (synthetic biology, photonics, RF, etc.). Have your CTO write the first draft of each project narrative; the consultant edits, structures, and stress-tests it. Budget $30K-$80K for the engagement at this scale. The first-claim review survival rate gain easily repays the fee.
You're a Quebec CCPC. You're filing federal SR&ED and Quebec's R&D wage tax credit (CRDIS), which Quebec restructured in 2025 to broaden eligibility. The two claims share underlying documentation but file as separate returns — federal with the CRA on T661, provincial with Revenu Québec on the corresponding RD-1029.7 form. Combined effective rates often exceed 70 cents back on every dollar of qualifying R&D salary.
Path: Hybrid leans toward consultant. The dual-filing complexity, with two slightly different sets of eligibility rules and two separate audit pathways, is where DIY gets expensive in mistakes. A Quebec-based SR&ED firm that handles both filings as a package is usually worth the fee at any non-trivial claim size. Budget 15-20% combined fee on smaller claims, 10-12% on larger ones. The provincial credit's broadened 2025 eligibility means there is meaningful upside in getting a sophisticated reading of which projects qualify under the new rules.
You raised a pre-seed or seed round in the last 12 months. You're spending $400K-$1.2M annually, almost entirely on engineering and a small slice of equipment. You have no revenue yet. The refundable portion of SR&ED — the cash refund regardless of tax position — is structurally vital to runway. A delayed or denied claim is an existential issue, not a line item.
Path: Either DIY with extreme care or a fixed-fee consultant. Avoid contingency arrangements at this stage; the fee structure incentivises aggressive claims that increase review risk, and a reduced first-claim sets a tone for subsequent years. Spend the up-front time to build a documentation system that survives audit. The single biggest mistake pre-revenue startups make is treating SR&ED as a year-end scramble; it should be a continuous operating discipline because the cash is too important to leave to memory and reconstruction.
| Claim size (qualifying expenditures) | DIY hours (year 1) | DIY hours (year 2+) |
|---|---|---|
| Under $150K | 40-60 hours | 20-30 hours |
| $150K - $500K | 60-100 hours | 30-50 hours |
| Over $500K | 100-180 hours | 50-90 hours |
T661 looks intimidating — it's a multi-page form with dozens of line items and a specialised vocabulary — but its underlying structure is straightforward. Six parts, in order: identification (Part 1), project descriptions (Part 2), calculation of SR&ED expenditures (Part 3), calculation of qualified SR&ED expenditures for the investment tax credit (Part 4), additional information (Part 5), and the claim preparer information (Part 6). The form is available from CRA in fillable PDF and has an accompanying T4088 guide that walks through each line. Read T4088 before you start filling anything; it answers 80% of the questions you'll have.
Source: CRA — Form T661, Scientific Research and Experimental Development (SR&ED) Expenditures ClaimLines 240/242/244 in Part 2 (the project narrative) and Lines 300-360 in Part 3 (salary expenditures, contractor expenditures, materials, and the proxy or traditional overhead calculation) account for roughly 90% of the total claim work and 100% of the technical-merit risk. The rest of the form is bookkeeping.
Part 2 is where you describe each SR&ED project. CRA limits each project description to 1,400 words. Line 240 asks "What scientific or technological uncertainties did you attempt to overcome?" Line 242 asks "What work did you perform in the tax year to overcome those uncertainties? Briefly state the hypotheses tested and the experiments performed." Line 244 asks "What scientific or technological advancements were you trying to achieve?" These three lines are the entire technical narrative. A reviewer reads them, assesses whether the work satisfies the three-part eligibility test, and either accepts or flags the project for technical review. If your narrative on these three lines is weak, no amount of polish elsewhere in the form will save the claim.
Part 3 calculates total SR&ED expenditures. The biggest line items are typically Line 300 (salaries and wages of employees directly engaged in SR&ED), Line 320 (contract expenditures), Line 340 (materials consumed or transformed), and Line 360 (the prescribed proxy amount, equal to 55% of qualifying salaries if you elect the proxy method). Line 360 is where founders most often leave money on the table by under-claiming; if you elect the proxy method correctly on Line 162, the 55% loading flows through automatically.
Standard corporate identification: name, business number, fiscal-year start and end, contact person. Line 162 is the critical election — it's where you choose proxy method or traditional method. The election is irrevocable for the year, so think about it before checking the box. For most founders the proxy method wins on simplicity alone (no overhead-allocation paperwork) and usually wins on dollar amount too because the 55% loading is generous relative to actual overhead in salary-heavy operations.
One project per page (digitally, one project section per submission). Lines 240, 242, 244 are the three narrative lines, each up to 350 words for a total project budget of roughly 1,050 words plus headers. CRA's published guidance is to write in plain language and avoid marketing or commercial framing. Use "we" not "the company"; describe technical actions, not business outcomes. The reviewer is a technical specialist in your field area and will catch puffery instantly.
Line 300 (salaries) requires a per-employee allocation of hours-on-SR&ED versus total hours. The CRA accepts time-sheets, calendar reconstructions, and project-based allocation methodologies. It does not accept "75% of engineering payroll" without supporting allocation. Line 320 (contractors) requires invoices that identify the SR&ED work explicitly — if a contractor's invoice says "consulting services" without project detail, get a revised invoice or supplementary letter. Line 340 (materials) covers components consumed or transformed in experimental work; office supplies and SaaS subscriptions are not material costs unless they are directly part of the experimental apparatus.
This part calculates the federal investment tax credit by applying the appropriate rate (35% refundable for CCPC enhanced, 15% for non-CCPC base) and reconciling against any government assistance received. If you got an IRAP grant, a federal R&D grant, or a Quebec provincial credit on the same expenditures, those amounts reduce the SR&ED-eligible expenditure base proportionally. Get this wrong and the CRA will catch it; the assistance-tracking is one of the easiest line items for a reviewer to audit.
Largely administrative. List of associated corporations, election questions, contact for the SR&ED file. Important: this is where you confirm the names and credentials of the technical staff. The CRA may use this list to schedule a technical review and will speak to the named technical contacts.
If a third party prepared the claim (consultant, accountant), they sign here. If you DIY, leave it blank or put your own info as preparer. CRA does not penalise self-prepared claims and the field is informational.
T661 attaches to the corporate T2 return for the same fiscal year. CCPCs have 12 months from the corporate filing deadline to file SR&ED — six months from year-end is the corporate deadline, plus 12 months equals 18 months from fiscal year-end, which is the absolute outside window for the SR&ED claim. Late SR&ED claims will not be considered. There is no extension. Plan to file alongside the corporate return, or within 6-9 months after, but do not let the 18-month clock expire.
Schedule T2 SCH 31 ("Investment Tax Credit") accompanies T661 and rolls the qualifying expenditure pool from T661 into the corporate-tax investment tax credit calculation. Most accountants generate Schedule 31 automatically once T661 numbers are entered.
The best path for a DIY filer is to spend roughly 60% of total claim time on Part 2 narratives, 30% on Part 3 expenditure allocation, and 10% on the rest of the form, because that's where claim risk and dollar value are concentrated.
Because: a clean Part 1 with a weak Part 2 will get reduced; a clean Part 2 with a sloppy Part 5 might get a follow-up email but will not lose dollars. Spend your hours where the credit lives.
The structural anatomy of a winning T661 Part 2 narrative is the same regardless of field: baseline → uncertainty → hypothesis → experiment → outcome. Five movements, totalling 700-1,400 words, written in plain technical prose. Most rejected first-time DIY claims fail because they collapse two of the five movements (typically baseline and uncertainty get merged into a vague "this was hard" statement, or hypothesis and experiment get merged into a generic "we built and tested" line).
Open with one paragraph naming the publicly available baseline. Open the second paragraph with the words "The technological uncertainty was..." and name a specific obstacle the public knowledge could not resolve. Then describe the hypothesis tested, the experiment run, and the technical result. Avoid commercial framing throughout. The CRA reviewer is a technical specialist in your field area — write to them, not to an investor.
Each movement of the narrative answers a different question for the reviewer. The baseline establishes that a competent professional in the field would not have a ready solution: it cites specific publications, vendor documentation, or open-source approaches and explains why each falls short. The uncertainty names the technical obstacle precisely, in the language the field uses. The hypothesis describes what you tested — what change to architecture, parameter, material, or method you predicted would resolve the obstacle. The experiment describes how you tested it — the experimental setup, the metric, the comparison. The outcome reports what you learned, including negative results.
Common writing mistakes: using brand names instead of technical concepts ("we built our X engine" rather than "we implemented a streaming differential synchronization protocol"); writing in past tense as a marketing recap ("we successfully launched") rather than as an investigation log ("we observed that..."); omitting failure modes (a narrative that reports only successes reads as commercial, not experimental); and using the word "novel" without specifying the comparator ("novel approach" is empty — novel compared to what published work?).
1. Baseline (one paragraph, 100-200 words). "At the project start, the publicly available approaches to streaming differential synchronization for multi-tenant collaborative editors were operational-transform (OT) variants documented in [Ellis & Gibbs, 1989] and conflict-free replicated data type (CRDT) families surveyed in [Shapiro et al., 2011]. Both approaches assume an underlying transport with bounded latency. In our deployment context — mobile-first clients on intermittent cellular networks — transport latency was unbounded and frequently exceeded application-layer session timeouts. Open-source implementations (Yjs, Automerge, ShareDB) all relied on bounded-latency assumptions for their conflict resolution semantics."
2. Uncertainty (one paragraph, 100-200 words). "The technological uncertainty was whether a hybrid CRDT-OT scheme could maintain causal consistency under unbounded transport latency without requiring full state reconciliation on each reconnection event. The publicly available literature surveyed reconnection strategies that scale linearly with document size; for our 50MB-100MB document working set, that approach was operationally infeasible. We did not know, prior to investigation, whether sub-linear reconnection was achievable while preserving the causal-ordering guarantees CRDTs provide."
3. Hypothesis. "We hypothesised that a vector-clock-indexed delta cache, persisted client-side, could reduce reconnection state transfer to O(log n) of unsynchronised operations rather than O(n) of total document size, while still enabling provably causal merge."
4. Experiment. "We implemented two prototypes — a baseline using Yjs's standard reconnection protocol and an experimental implementation using the proposed vector-clock delta cache. We ran controlled tests on a corpus of 1,200 simulated session reconnections drawn from production telemetry, measuring (a) bytes transferred on reconnect, (b) wall-clock reconnection time, and (c) merge correctness against a deterministic ground-truth document state."
5. Outcome. "Bytes transferred dropped by 87% on average (95% CI 82-91%); wall-clock reconnection improved by 71%. We identified a class of edit patterns — concurrent table-cell modifications in shared spreadsheet ranges — where the delta cache approach produced merge results that diverged from the deterministic ground truth in 0.4% of cases. We documented this divergence as a known limitation, scoped follow-on work to address it, and shipped the improved reconnection scheme for non-table workloads. The work resolved the original uncertainty for the targeted use case and identified a bounded successor uncertainty."
Reading 100 successful T661 narratives reveals a small set of patterns. They cite specific public knowledge as the baseline (papers, vendor docs, open-source repositories), they name the uncertainty in field-specific language, they describe experiments in enough detail that another practitioner could replicate them, and they report outcomes including negative results. They do not read like product launches; they read like internal engineering retrospectives written for a peer audience.
Conversely, weak narratives share signals: heavy use of marketing terminology ("breakthrough", "best-in-class", "industry-leading"); generic descriptions of the work done ("we performed extensive testing"); absence of specific public-knowledge baselines; omission of negative or partial results. A reviewer reading a weak narrative typically requests a technical interview — which adds 30-90 days to claim resolution — or proposes a partial reduction.
Here's what you need to know about the narrative voice: write it as if you were sending it to a senior engineer in your field who is reading skeptically. They want to know what was hard, what you actually did, and what you learned. They do not want to know about market opportunity, customer impact, or the company's growth trajectory. The CRA reviewer reads tens of these per quarter and recognises commercial framing instantly — not as fraud, but as evidence that the writer doesn't understand what the program is for. The single most useful editing pass on a T661 narrative is to delete every sentence that would be true regardless of whether the work was technologically uncertain.
The best path for narrative writing is to draft a 1,500-word technical-retrospective version first, then ruthlessly cut to the 1,050-word T661 budget, because the cutting forces you to keep only the sentences that move the eligibility test forward.
Because: writing tightly to the form's word budget on the first pass produces narratives that omit critical baseline or uncertainty framing. Drafting long and cutting hard produces narratives that survive review.
The dollar calculation is the easy part of SR&ED. The hard part is correctly classifying expenditures and applying the right rate. Here we walk through the core arithmetic for a typical CCPC.
For each employee directly engaged in SR&ED work during the fiscal year, calculate (hours on qualifying activity / total hours worked) × (annual salary, including bonuses, vacation pay, and most benefits). The CRA accepts time-sheets, calendar reconstructions, and project-tag-based allocations from time-tracking software. It does not accept blanket percentages without underlying support. Specified employees (typically the controlling-group founders) face a salary cap of five times the year's maximum pensionable earnings under the Canada Pension Plan (CPP) — for 2026, that's roughly $361,500.
Subcontractors performing SR&ED on your behalf are eligible at 80% of the contracted amount (the remaining 20% is treated as the contractor's overhead, which is not refundable to you). The contract must be at arm's length and must clearly identify the SR&ED work performed. International contractors performing SR&ED outside Canada are not eligible for SR&ED expenditures — the work has to happen in Canada, with limited exceptions for Canadian contractors performing work abroad.
Components consumed or transformed in experimental work qualify. Capital expenditures are eligible again post-Budget 2025 if used primarily (more than 50%) for SR&ED. Office supplies, generic SaaS subscriptions, and rent are not materials.
The proxy method adds 55% of qualifying salaries (Step 1) to the expenditure pool as a deemed overhead loading. The traditional method requires you to track and allocate every overhead line item directly to SR&ED projects. Most early-stage CCPCs choose proxy — it's simpler and usually larger. Election is on T661 Line 162 and is irrevocable for the year.
Any non-SR&ED government assistance received on the same expenditures (an IRAP grant on the same project, a Quebec wage credit on the same hours) reduces the SR&ED-eligible expenditure pool. This is the single most common arithmetic error in DIY claims. Track every non-SR&ED government dollar that touched the same costs and net it out.
For a CCPC with prior-year gross revenue under $15M and no associated-group complications, the first $6M of qualifying expenditures earns the 35% refundable rate. Above $6M (or above the phase-out window) it's the 15% non-refundable base rate. The maximum federal refundable credit is $6M × 35% = $2.1M per year.
| Business profile | Federal rate | Refundable? |
|---|---|---|
| CCPC, prior-year revenue under $15M | 35% (first $6M) | 100% refundable |
| CCPC, in $15M-$75M phase-out | Sliding 35% → 15% | Sliding 100% → 0% |
| Public corp / non-CCPC | 15% | Non-refundable (carry-forward only) |
Assume a CCPC at $1.8M ARR, prior-year gross revenue $1.2M (well under the phase-out), with the following expenditure profile:
Calculation: $740,000 (salaries) + $740,000 × 55% = $407,000 (proxy) + $144,000 (contractors) + $22,000 (materials) = $1,313,000 qualifying expenditures. At the 35% refundable rate: $1,313,000 × 35% = $459,550 federal refundable credit. If the company is in Quebec, the provincial CRDIS adds roughly another 25-30% on the qualifying salary base, depending on the post-2025 Quebec rules and the project mix. Total combined credit: roughly $640K-$700K cash, on $1.3M of qualifying expenditures.
For a typical CCPC under the limit and not in the phase-out: total qualifying salaries plus contractors-at-80% plus materials, plus 55% of salaries (proxy method), minus any government assistance on the same costs. Multiply the result by 35%. That's your refundable federal credit. Cap is $2.1M per year.
Each component has subtleties. Salaries need per-employee hour allocation. Specified employees (founders with significant ownership) face a $361,500 cap in 2026. Contractors are at 80% to capture only the labour component, not the overhead. Materials must be consumed or transformed in experimentation, not used in production. Capital expenditures are eligible again post-Budget 2025 if SR&ED-use is primary (more than 50%). The proxy 55% loading replaces all overhead tracking; it does not stack with traditional-method overhead. Government assistance subtraction is dollar-for-dollar against the affected expenditures, not the credit.
Above $15M prior-year gross revenue, the enhanced 35% rate begins to phase out linearly until full phase-out at $75M. The expenditure limit phases out on the same curve. At $30M revenue, you retain roughly 75% of the enhanced rate window; at $50M, roughly 42%; at $75M+, you're at the base 15% rate. The phase-out is mechanical — CRA's calculator on T661 Schedule 31 walks the math.
If your corporation is part of an associated group (controlled by the same person, related persons, or a chain of corporations), the $6M expenditure limit and the gross-revenue phase-out are computed at the group level, not per corporation. Founders running parallel ventures that share investors or directors should map associations carefully — missing this analysis can produce a claim that the CRA reduces because the limit was already consumed by an associated entity. The CRA's IT-64R4 (Associated Corporations) is the canonical reference.
Provincial R&D credits stack on top of the federal credit but are typically net of the federal credit on the same expenditure base. Each province publishes its own rate and rules. Quebec's CRDIS, restructured in 2025, is the most generous; Ontario's ORDTC and OBRITC are smaller; BC and Alberta have meaningful provincial credits. The total combined effective rate for a Quebec CCPC can exceed 70% on qualifying salaries; for an Ontario CCPC, roughly 45-50%; for a BC CCPC, roughly 40-45%.
The best path for the calculation is to elect the proxy method on Line 162 unless your salary intensity is unusually low, because the 55% loading produces a larger and simpler claim than traditional overhead tracking for most early-stage CCPCs.
Because: traditional method is paperwork-heavy and rarely produces a meaningfully larger claim for salary-heavy operations. The 55% proxy is generous on a salary-loaded cost structure and saves dozens of hours of allocation work.
The CRA does not require you to submit documentation with the T661. The form goes in clean. Documentation comes into play if and when a review opens — financial review of the expenditure pool, technical review of the project narrative, or both. At that point, you have 30-60 days to assemble the supporting record. Founders who reconstruct documentation in that window typically lose 20-50% of the claimed amount. Founders who hand the reviewer a tidy folder of contemporaneous records typically come out at 90-100% of the claim.
Maintain three living artifacts during the year: (1) per-employee time records, ideally in a tool like Harvest or Clockify with project tags that map to T661 projects; (2) per-project running notes capturing baseline, uncertainty, hypothesis, experiment, outcome — updated monthly, not at year-end; (3) per-expenditure receipts and contractor invoices that name the SR&ED project explicitly. If a reviewer can trace any claimed dollar to a primary record, you survive.
The CRA's evidentiary preferences are well-documented in IC86-4R3 and the SR&ED Eligibility of Work Policy. Reviewers favour contemporaneous records over reconstructed ones because the former cannot be retrospectively shaped to fit the claim. A GitHub repository with commit messages dated throughout the year is gold; a markdown file written the week before the audit is mud. The ideal documentation system runs continuously in the background of normal engineering operations: time-tracking by project tag, monthly retrospective notes by project, archived design docs and code review threads, and a per-project folder where receipts and invoices are filed as they're paid.
Engineers resist time-tracking, often vocally, and founders frequently capitulate. This is a costly mistake. Even rough hour-level granularity (8 hours on Project A on Tuesday, 4 hours on Project B on Wednesday) is sufficient if it's contemporaneous. Annual reconstructions are not. The cost of asking engineers to spend 5 minutes per day tagging hours is roughly 1% of their salary; the cost of losing a $100K claim reduction in review can run to many percent.
Per-employee, per-day, per-project granularity. Tooling: Harvest, Clockify, Toggl, Tempo for Jira, or even a structured weekly survey form. Tags should map directly to the T661 project IDs. Specified employees (founder-owners) should track too — their time is often the largest single salary line in the claim. Retain time records for at least six years post-filing (CRA's standard records-retention period).
One document per project, updated monthly. A simple template works: header lines for baseline, current uncertainty, current hypothesis, experiments run this month, outcomes. At year-end, this document compresses into the T661 Part 2 narrative with minimal effort. Critically, the running log captures the experimental flow as it actually happened — including the dead ends, which a reviewer will want to see as evidence of systematic investigation.
For software work: GitHub commits, pull requests, design docs in Notion or Confluence, code review threads, architectural decision records (ADRs). For lab work: bench notebooks (paper or electronic), instrument logs, sample tracking sheets. For hardware: schematic version history, board revisions, test plans, measurement archives. The reviewer is not going to read every commit, but a representative sample plus the running log establishes the systematic nature of the work.
Payroll registers showing per-employee compensation; contractor invoices that name SR&ED projects (request revised invoices if the original lumps work as "consulting"); material receipts; capital-expenditure records with usage logs proving SR&ED-primary use. File these into a per-project folder as they happen, not at year-end.
Patent applications, conference paper drafts, technical blog posts, customer-facing technical documentation, third-party benchmarking reports. Not required, but powerful when present — they corroborate that the work was advancing the field, not just integrating known parts.
| Claim profile | Approx. review rate | Common review type |
|---|---|---|
| First-time CCPC, any size | ~30-40% | Combined financial + technical |
| Repeat CCPC, under $250K | ~10-15% | Financial only (spot check) |
| Repeat CCPC, over $1M | ~25-35% | Technical, with financial follow-up |
Note: Review rates are CRA-published guidance ranges from public SR&ED program statistics; actual rates vary by industry, region, and prior compliance history.
Here's what you need to know about documentation discipline: the work you do during the fiscal year to maintain primary records compounds in value. Year one feels like overhead. Year three, when your claim is well above $500K and the review is technical, the running notes and commit history you've been keeping make the difference between a 100% claim and a 50% reduction. Founders who treat SR&ED documentation as a continuous discipline routinely outperform those who treat it as a year-end project, even when the latter hire consultants to compensate.
The best path for documentation is to build the time-tracking and running-narrative system in the first month of the fiscal year and run it continuously, because reconstruction at year-end loses 20-50% of dollars in any meaningful review.
Because: the cost of continuous tracking is small (5 minutes per engineer per day) and the cost of a reduced claim is large. Documentation discipline is the highest-ROI behaviour change a founder can adopt for SR&ED.
If your claim is selected for review, the CRA assigns a Research and Technology Advisor (RTA) for the technical side and a Financial Reviewer for the cost side. They contact you by letter or call, schedule an interview, and request supporting documentation. The review can resolve in any of three outcomes: full acceptance (claim paid as filed), partial acceptance (claim reduced to a lower amount), or denial (claim disallowed, with appeal rights). The dominant outcome for well-prepared first-time claims is full or near-full acceptance. The dominant outcome for poorly prepared first-time claims is partial reduction in the 20-50% range.
The RTA interview typically runs 60-120 minutes. The RTA will have read your T661 Part 2 narratives in advance and prepared questions tied to the eligibility criteria. Common question patterns:
The Financial Reviewer focuses on expenditure allocation, government-assistance netting, contractor eligibility, and specified-employee caps. They will request the underlying time records, payroll registers, and contractor invoices. Hand them a single PDF or zip with everything organised by T661 line item; do not make them dig. Common findings:
Small businesses are the lifeblood of our economy and integral to the success of communities across the country. Small Business Week is an opportunity to highlight their tireless work and recognize their tremendous contributions, especially in challenging times. The Honourable Rechie Valdez — Minister of Small Business, Government of Canada
| Profile | Federal credit (35%) | + Quebec CRDIS | Combined effective |
|---|---|---|---|
| $300K qualifying expenditures, Ontario CCPC | $105K | n/a (Ontario ORDTC: ~$10K) | ~$115K (38%) |
| $300K qualifying expenditures, Quebec CCPC | $105K | ~$75K-$90K | ~$180K-$195K (60-65%) |
| $1M qualifying, Quebec CCPC, $200K salaries | $350K | ~$50K-$60K (CRDIS on salaries) | ~$400K-$410K (40-41%) |
Here's what you need to know about review tone: the RTA and Financial Reviewer are public servants doing technical compliance work, not adversaries. Most reviewers spent years in industry before joining the CRA — in software, biotech, manufacturing, materials — and they read narratives in good faith. Adversarial founders consistently produce worse outcomes than cooperative ones. Treating the review as a peer technical conversation, with the RTA as a senior reviewer at a different lab, produces dramatically better outcomes than treating it as a tax dispute.
The best path through a first-claim review is to over-prepare the documentation packet, let the technical team speak technically, and concede small reductions cleanly while defending the substantive claim, because the review is a technical-credibility test more than a numerical dispute.
Because: reviewers calibrate the entire file based on the first 20 minutes of interview. A CTO who walks fluently through one experiment establishes credibility that carries the rest of the review. A founder who hedges or improvises invites deeper scrutiny.
Here's what you need to know about timing the DIY decision: the cost-benefit math flips around the $200K-$300K refund threshold. Below that, consultant fees of 18-25% wipe out the leverage advantage of expert representation, and a competent founder can produce an audit-defensible file in 30-50 hours of focused work. Above $400K, consultant value re-emerges — not because you can't file solo, but because the marginal hour spent on cleaner narratives, sharper expense allocation, and pre-emptive review preparation produces returns the consultant fee absorbs cleanly. Most CCPCs filing under $300K should file themselves; most filing over $500K should at least get a strategic review from a fee-capped consultant before submission.
Plug your fiscal-year salary, contractor, and material spend into our free SR&ED calculator. Get an instant estimate of your refundable federal credit and provincial layering, plus a 1-page summary you can share with your accountant.
Open the SR&ED Calculator →