For $180 a year, Trace gives a family an automated highlight reel centred on one player. Before the match, the team supplies the roster and shirt numbers. After it, the system delivers each player’s clips.
In February 2026, the NBA demonstrated POV Mode: a live game reconstructed from a chosen player’s vantage point. The league described the on-stage version as a prototype whose animated figures would “ultimately be replaced” by visualisations of the actual players. It was a preview, not a product launch.
These are not equivalent products. Trace edits a recorded match hours later; the NBA is trying to synthesise a new view while the match is happening. That difference is the point.
The most useful way to understand sports video is not by asking which company has the best AI. Ask three narrower questions: How quickly must the answer arrive? Who supplies the player’s identity? What happens if the system is wrong? Those constraints explain the architecture—and much of the business model.
That frame is close to what the system actually sees. One shirt number is barely readable. No face is.
The hidden variable is time
Trace’s own support material says a 90-minute match can take almost four hours to upload on a 10 Mbps connection, followed by up to 12 hours of processing. The Bundesliga operates at the other extreme. Its current tracking system uses 16 to 20 stadium cameras and sends roughly 3.5 million data points to live applications within 150 milliseconds.
Give a system hours and it can batch work in the cloud, retry uncertain frames and put exceptions in a queue. Demand an answer in 150 milliseconds and the venue needs calibrated cameras, low-latency networking and compute already standing by. The deadline is not a feature added at the end. It determines the system.
The commercial models make that cost visible, although not perfectly. XbotGo Falcon advertises built-in AI and no subscription. Move ’N See similarly sells local recording and free streaming without a mandatory plan. At the cloud end, SportsVisio charges $149 or $189 a month for five uploaded games; its published plans imply per-game rates of $29.80 to $37.80. It promises the results within 24 hours.
That is not a universal law—subscriptions also pay for storage, support and product updates. It is a useful cost signal: recurring processing tends to produce recurring revenue.
| Job | Published deadline | Where identity starts | Commercial model |
|---|---|---|---|
| Personalised family clips | Up to 12 hours after upload | Roster and shirt number supplied by the team | $180 a year |
| Uploaded stats and highlights | Within 24 hours | Roster plus post-match video analysis | $149–$189 a month |
| Official live tracking | About 150 milliseconds | Known roster plus continuous, calibrated tracking | Enterprise and rights contracts |
The products sound adjacent. Operationally, they are different species.
Identity has to come from somewhere
At the family end, the customer does part of the labelling. Trace requires jersey numbers to be assigned before the match and warns that similar colours, poor light or low camera placement can break detection. Its troubleshooting guide is unusually candid: “if a human would have trouble identifying players, so will the computer”.
Other systems attach identity outside the image. Playback brings GPS data, video and player profiles into one workflow. The NFL’s live tracking system goes further: players carry two or three RFID tags in their shoulder pads, read by 20 to 30 receivers around the stadium ten times per second. The image does not have to solve the entire identity problem alone.
At the top end, identity can persist through continuous observation. Hawk-Eye’s SkeleTRACK uses a ring of roughly 10 to 14 cameras to track 29 points on each player’s body. With a known roster, multiple views and no gaps in coverage, the system can carry an identity from one frame to the next.
This is why “better face recognition” is the wrong shortcut. In wide-angle sport footage, faces are too small and blurred; research on soccer-player recognition treats shirt numbers, team appearance, position and tracking as complementary evidence. Rich and cheap systems face the same pixels. Expensive systems buy more ways around them.
More money can mean more humans
It is tempting to assume that higher budgets always mean more automation. The middle of the market shows why that is wrong.
Post-production analysis tools are often manual by design. A coach or analyst watches the match and presses buttons to tag events. Nacsport’s own beginner guide describes building buttons, then clicking them as actions occur.
The largest data suppliers also combine automation with people. In a 2021 filing, Genius Sports disclosed a network of more than 7,000 statisticians across more than 150 countries. Sportradar’s 2024 annual report said roughly half of its data was captured using advanced AI tools—but it listed proprietary systems, in-venue scouts and television coverage among its other collection methods. It did not say that the other half was all manual. Stats Perform is clearer about the operating principle: Opta Vision keeps a human in the loop at every stage.
These businesses automate less completely than a family highlight reel because the cost of error is higher. A wrongly tagged clip is an annoyance. A wrong event in an official feed can corrupt a broadcast graphic, a club decision or a betting market. The relevant metric is not the buyer’s budget; it is the cost of being confidently wrong.
At the top, payment reverses
Follow the money upward and the relationship changes.
A family pays Trace $180 a year. SportsVisio’s entry plan is $1,788 a year. Catapult said its average contract value per professional team was almost $27,000 in its 2025 financial year—an average across its platform, not a list price for one camera.
At the rights tier, the technology supplier may pay the sports property. Sportradar’s NBA agreement included annual licence fees and warrants equal to 3% of the company on a fully diluted basis, exercisable at one cent a share. The NBA became both supplier and shareholder. Genius Sports’ 2021 NFL deal likewise included 18.5 million penny warrants; later filings disclose additional warrants issued in 2025.
That reversal matters. At the bottom, technology is sold to the person holding the camera. At the top, exclusive access to the game is a scarce input that the technology company must buy.
A market with movable borders
This is also why neat “sports analytics market” numbers should be handled carefully.
| Research firm | Estimate | Reference year |
|---|---|---|
| IMARC | $1.73bn | 2025 |
| MarketsandMarkets | $2.29bn | 2025 |
| Mordor Intelligence | $5.28bn | 2026 |
These are not three measurements of an identical thing in an identical year. Mordor, for example, explicitly excludes standalone hardware. The spread is more useful as evidence of unstable category boundaries than as a precise demand forecast.
The business underneath
Sportradar’s 2025 results put the rights economics in plain view: €404.3 million of sports-rights expense against €1.290 billion of revenue. Roughly 31 cents of every euro of revenue went to rights holders.
This is a rights-led distribution business with computer vision inside it. Exclusive access is acquired, official data and products are sold to broadcasters, sportsbooks and other customers, and automation reduces the cost and increases the speed of collection. The model is a long way from selling a camera to a parent, even when both companies describe what they do as “sports AI”.
The causal chain is simpler than the market map makes it look. The deadline determines the architecture. The cost of error determines how much human and physical infrastructure stays in the loop. Exclusive rights determine who pays whom.
The cheapest camera shipped first because it solved a narrower job with more time, identity supplied by the customer and forgiving consequences. The NBA prototype is attempting a different job entirely. The surprising part is not that one arrived before the other. It is that the industry uses the same words for both.