How Much Would the NHL's Puck-Tracking Engineers Claim in SR&ED?
The NHL's puck-tracking system captures thousands of data points per game. Most fans think it's magic. It's actually a decade of painstaking R&D by Canadian companies like Sportlogiq and Stathletes — and it probably qualifies for millions in tax credits.
Thousands of data points per game. None of it existed ten years ago.
The NHL's puck-tracking system — officially called NHL EDGE, powered by Amazon Web Services — captures thousands of unique data points per game. Shot velocity. Passing networks. Defensive zone exits. Time-on-attack metrics. Every one of those numbers flows from a camera system that didn't exist in any commercially viable form a decade ago.
Behind the broadcast overlay is a mountain of genuinely hard R&D. Montreal-based Sportlogiq — one of the NHL's primary analytics providers — builds computer vision models that identify players, track pucks, and derive contextual events from broadcast footage. Waterloo-based Stathletes, founded by McMaster grads Sara and Meghan Normandeau, turns that same footage into structured datasets that power betting markets, team strategy sessions, and broadcast commentary.
Here's the twist almost nobody notices: the R&D that makes sports analytics work isn't the fancy model. It's the grunt work of making that model function on real, noisy, unpredictable footage — where arena lighting changes mid-period, camera angles switch without warning, and players constantly block each other. That's where SR&ED lives.
Why tracking a hockey puck is harder than tracking a car
Computer vision is pretty good these days. Your iPhone can identify your face in a dark room. Autonomous cars can track pedestrians, cyclists, and other vehicles in real time. So how hard can it be to track a hockey puck?
Extremely hard, as it turns out.
Players wear identical uniforms — same colours, same numbers, same helmets. The puck moves at 100+ mph and disappears behind skates, boards, and goalies for significant stretches of every game. Glass reflections create phantom objects. Arena lighting varies by venue. And every time the broadcast cuts to a different camera angle, the system has to re-identify every player from scratch.
In a 2023 interview with Sports Business Journal, Sportlogiq CEO Craig Buntin revealed that his R&D team spent 18 months — eighteen months — iterating on player re-identification algorithms alone. Published computer vision methods existed. They just didn't work on hockey footage. The noise profile was wrong. The occlusions were too frequent. The uniform similarity broke every standard approach.
A computer vision team tries a published deep-learning approach for multi-object tracking on NHL broadcast footage. The model crushes it on clean test datasets. On real footage, it fails catastrophically: glass reflections create fake players, inconsistent arena lighting breaks the colour model, and camera cuts during line changes make the system think half the team just disappeared. The team tests four modified architectures over eight months, documents why each one fails on specific noise types, and eventually builds a hybrid approach combining optical flow with appearance-based re-identification. The failed attempts, the noise characterization, and the hybrid solution are all logged in real time. This isn't 'using AI.' This is systematic investigation. And it's textbook SR&ED.
The most common mistake sports analytics companies make: claiming the entire platform as R&D. The dashboard that displays tracking results? Product engineering. The API wrapper that serves data to broadcast partners? Standard integration. The billing system that charges teams by seat? Definitely not R&D. SR&ED applies only to the specific technical obstacles where existing knowledge couldn't provide a solution. The rest is just software — valuable software, but not SR&ED.
What a real sports analytics claim looks like
A typical sports analytics company with 12 engineers probably has 3–4 people doing core R&D: model architecture, training pipeline improvements, and edge-case handling. The other 8–9 are building infrastructure, adding product features, and managing client integrations. That's normal. The key is knowing which work qualifies and documenting it properly.
- Qualifies: Developing a tracking approach for occluded players where published methods fail, then systematically benchmarking against real footage and documenting why each attempted method failed. That's R&D.
- Doesn't qualify: Building the dashboard that displays tracking results. That's product engineering.
- Qualifies: Training a predictive model for shot quality where no existing benchmark exists for your specific camera angle and rink configuration, then experimenting with feature engineering and documenting which approaches generalize.
- Doesn't qualify: Applying an existing shot-quality model to a new season's data. That's routine data processing.
CRA cares about systematic investigation, not commercial value. A model that generates $10 million in licensing revenue is not more eligible than one that generates $100,000. What matters is whether the development required genuine experimentation because existing approaches couldn't resolve the specific technical constraints you faced.
The money question: how much would they actually get back?
Let's run the numbers on a realistic sports analytics company.
12-person company. 4 engineers on qualifying R&D. Average salary $140,000. Each spends roughly 60% of their time on core research. That's $336,000 in qualifying wages per year.
At the enhanced 35% federal rate plus Ontario's 8% OITC, that's approximately $144,000 in annual refundable credits. Over three years of development — which is a realistic timeline for building, iterating, and commercializing a sports analytics platform — that's $432,000 in non-dilutive capital.
To put that in context: Sportlogiq raised $5 million in Series A funding in 2016. A company of that scale doing core computer vision R&D would typically be eligible for significant SR&ED credits. Those credits, if fully claimed, would have represented meaningful non-dilutive capital — potentially 8–10% of their total capital stack at that stage. That's the kind of capital structure that lets a founder sleep better at night.
This guide analyzes hypothetical SR&ED scenarios based on publicly available information about Sportlogiq, Stathletes, and NHL analytics partnerships. Actual claim values depend on specific company circumstances, documentation quality, and CRA review. Consult a qualified Canadian CPA for advice tailored to your situation. Learn more at sredy.io.
Turn this guide into a claim package
SREDY.IO walks you through eligibility, project narratives, supporting costs, and evidence so your CPA has a cleaner file to review.
More guides
What Actually Makes Software Work SR&ED-Eligible?
Most software companies doing real R&D leave SR&ED credits on the table because they don't recognize their own qualifying work. Here's the eligibility distinction that matters.
EligibilityAI Projects and SR&ED: What Changed, What Didn't
AI and ML work can qualify for SR&ED — but not all of it. The CRA's eligibility test hasn't changed. Here's what that means for model training, fine-tuning, and architecture decisions.
DocumentationThe Evidence Habit: What to Save Before Year-End
SR&ED claims are significantly easier to defend when you've been saving the right artefacts throughout the year. A practical pre-year-end checklist for Canadian tech companies.
Ready to check if your work qualifies? Take the free eligibility assessment.
Check eligibility — free →