Play-by-play is the sequential account of a sporting event, one action at a time, produced while the event is happening. The term began as a broadcast job description: the announcer who says what just occurred, as distinct from the analyst who explains why it mattered. It now carries a second meaning that matters more commercially. Play-by-play is also a data product, a machine-readable log in which every discrete occurrence in a game receives a timestamp, a type code, the identities of the participants and the resulting state of the contest.

Both senses describe the same thing at different resolutions, and advertising cares about the second. A structured play-by-play feed is the only source that knows, within roughly a second of the action, that a touchdown has been scored, a red card shown or a match sent to extra time. That knowledge is what allows an impression to be bought, decided and delivered against a specific moment in a game rather than against a scheduled break.

How a play becomes a record

Collection begins in the venue. An operator watching the field enters each event into a scoring application, choosing from a fixed vocabulary defined by the league or the data provider: shot, rebound, foul, substitution, pitch, tackle. Optical tracking supplies a second layer. Twelve synchronised high frame rate cameras in every Major League Baseball ballpark feed the Hawk-Eye system that has powered Statcast since 2020, tracking ball and player movement to a tolerance of a tenth of an inch, according to Hawk-Eye. Camera systems classify continuous motion; humans still adjudicate the discrete events that carry meaning, and the two streams are reconciled before publication.

The output is a record per event. A typical schema carries an event identifier, a sort order, the period and game clock, a wall-clock timestamp, a type code and description, the acting participants and the score after the play. Basketball adds coordinates for shot attempts. Ice hockey adds strength state. The naming differs by sport and vendor, which is why any system consuming several feeds has to normalise them into one internal shape before applying logic.

Delivery matters as much as content. In-play products require a push feed that streams events as they occur rather than polled requests that ask repeatedly whether anything has changed. Official feeds, licensed directly from a league, run faster and cleaner than scraped or observed alternatives, and they cost accordingly.

Where the signal enters the ad transaction

Until recently, live sport was bought at the level of the event. Buyers specified a match, a league or a window and accepted whatever happened inside it. Two developments in 2025 moved the unit of purchase downward.

The first was schedule-level infrastructure. Magnite launched Live Scheduler on November 18, 2025 inside SpringServe, sharing event name, timing, sport, league, broadcaster and concurrency estimates ahead of a fixture so buyers could plan against it. Twelve days earlier, Google Ad Manager introduced CTV live-biddable capabilities aimed at the inventory that appears when a game runs into overtime. IAB Tech Lab formalised the pattern through its Live Event Ad Playbook, whose Deals API and Concurrent Streams API sit as sidecars to OpenRTB, callable independently of the main bidding path.

The second was moment-level triggering, which requires play-by-play proper. PubMatic launched an AI-powered Live Sports Marketplace on July 17, 2025 built on analysis of live game data, arguing that conventional sports buying fails to distinguish high-engagement moments from lopsided fourth quarters. Genius Sports followed with Moment Engine, integrated into Magnite's ClearLine platform in an announcement dated March 2, 2026. The worked example the two companies published runs as follows: a decisive late scoring play during the NCAA basketball tournament is captured through official league data, converted into an advertising trigger inside ClearLine, and used to activate a pre-configured deal against relevant audiences across connected television inventory within seconds.

The chain runs: venue collection, provider normalisation, moment classification, injection into a sell-side platform, deal activation, bid, insertion. The last step is where older machinery takes over, since server-guided insertion resolves the creative just in time and concentrates ad calls into a narrow window at the break.

The timing problem

Play-by-play data travels faster than the picture. Betting operators target in-play latency below one second, because any wider gap invites a spectator in the stadium to bet on a result the market has not yet priced. Streaming runs far behind that. Standard segmented delivery typically produces 20 to 30 seconds of glass-to-glass latency, and measured lags on major streamed finals have run considerably longer.

The consequence for advertising is counterintuitive. A trigger fired the instant official data records a goal can reach a viewer before that viewer has seen the goal. Synchronising an advertisement with game action therefore means deliberately delaying the trigger to match a stream delay that varies by service, by device and by household. Systems that ignore this either spoil the moment or miss it.

Origin and evolution

The broadcast form arrived first. Harold Arlin, a 25-year-old Westinghouse engineer, described a Pittsburgh Pirates game from a box seat at Forbes Field on August 5, 1921 over KDKA, using a converted telephone as a microphone. Combat sports had gone out four months earlier as blow-by-blow, and cricket later settled on ball-by-ball. The variants share a structure: continuous narration keyed to discrete events.

The data form took six decades longer. Bill James and others started Project Scoresheet in 1984 to record every play of every game, a response to the Elias Sports Bureau declining to release play-by-play data publicly. David W. Smith founded Retrosheet in 1989 to computerise historic accounts, and its event files now cover most of major league history. Automated collection followed: pitch tracking debuted in the 2006 postseason and reached all 30 ballparks by 2008, Statcast arrived league-wide in 2015, and Hawk-Eye replaced the earlier radar and camera mix in 2020.

Rights consolidated in parallel. Leagues began licensing official data exclusively, and the resulting distributors, principally Sportradar, Genius Sports and Stats Perform, control the fast, authenticated feeds. Genius Sports has been the most explicit about moving that asset into advertising, unveiling Moment Engine at a NewFronts presentation on March 26, 2026 and extending it to Pac-12 football and basketball under a partnership announced nine days earlier.

Why the term matters for marketers

The commercial argument rests on where attention sits. Sports streaming has become the growth surface: 76% of sports streamers had watched NFL content live in the twelve months before a March 2026 study, with NBA streaming viewership rising from 33% to 41% over two years. Money has followed. Magnite reported 5,800 new live sports advertisers alongside a 56% rise in live sports spend, and Amazon's advertising leadership has described live sport and Prime Video as a single business rather than adjacent line items.

Play-by-play is what converts that audience into addressable units smaller than a match. It also feeds the viewer-facing products that now surround the stream: Prime Video's Key Moments and Rapid Recap features, announced on September 30, 2025, curate highlights from the same event stream that betting overlays consume.

Limitations and disputes

Accuracy is the first problem. Play-by-play events involving judgement are logged by people, and scorekeeper bias is documented. Research by Matthew von Bommel and Luke Bornn, published in Data Mining and Knowledge Discovery in 2017, found inconsistencies in how NBA scorekeepers record assists and blocks, with consequences for any product priced off box score statistics. Substitution logs in less-resourced leagues carry error rates high enough to require algorithmic repair before lineup analysis becomes possible.

The evidence for moment-based advertising is largely vendor-supplied. Genius Sports reported research conducted with MediaScience finding unaided brand recall roughly twice as high when an advertisement followed an emotionally intense play, and identified the 60 seconds after a high-impact moment as the peak effectiveness window. The study was commissioned by the company selling the trigger, and no independent replication has been published.

Cost and access form a third dispute. Official feeds are exclusive, expensive and concentrated among three suppliers, which puts moment targeting out of reach for buyers without a route through a curating platform. Brand safety is a fourth: a signal that fires on peak emotion fires equally on a serious injury or a crowd disturbance, and taxonomies distinguishing celebratory moments from distressing ones remain immature.

Disambiguation

Colour commentary is the analytical voice alongside the play-by-play announcer, supplying interpretation rather than description. The distinction is a division of labour inside one booth.

Tracking data records continuous spatial position for every player and the ball, sampled many times per second. Play-by-play records discrete events with semantic meaning. Tracking answers where; play-by-play answers what.

SCTE-35 cue messages mark splice points for advertisement insertion inside the transport stream. They signal that a break is available, not that anything happened in the game.

Concurrency data, standardised through the IAB Tech Lab work that IAB Europe's CTV guide describes, counts how many devices are watching. It measures audience, not action.

Recent developments

Verification has become the selling point. DeepIntent built a pharmaceutical live sports package around verified live inventory in June 2026, arguing that live signals mixed into general bidstream inventory offer reach without confidence. The 2026 football World Cup supplied the largest test to date, with Magnite publishing programmatic guidance on March 9, 2026 covering deal structures and failure handling, and FIFA approving advertising during three-minute water breaks across all 104 matches, an inventory category defined by an in-game event rather than a clock. Whether buyers will pay a premium for a signal they cannot independently audit remains the open question.

Timeline

  • April 11, 1921: KDKA broadcasts a boxing match described blow by blow, the first live sports event on commercial radio
  • August 5, 1921: Harold Arlin calls a Pittsburgh Pirates game on KDKA, the first play-by-play radio broadcast of a major league baseball game
  • 1984: Project Scoresheet begins collecting play-level records after Elias Sports Bureau declines to release its data publicly
  • 1989: David W. Smith founds Retrosheet to computerise historic play-by-play accounts
  • 2006: Automated pitch classification debuts in the Major League Baseball postseason
  • 2008: Pitch tracking hardware is installed in all 30 major league ballparks
  • 2015: Statcast is deployed league-wide after a partial trial the previous year
  • 2017: Research documents scorekeeper bias in NBA assist and block recording
  • 2020: Hawk-Eye optical tracking replaces the earlier radar and camera combination
  • July 17, 2025: PubMatic launches its Live Sports Marketplace for game-moment targeting
  • November 6, 2025: Google Ad Manager introduces CTV live-biddable capabilities
  • November 18, 2025: Magnite launches Live Scheduler within SpringServe
  • February 19, 2026: IAB Tech Lab opens public comment on the LEAP Forecasting API
  • March 2, 2026: Genius Sports integrates Moment Engine with Magnite ClearLine
  • March 17, 2026: Genius Sports and the Pac-12 announce a moment-based advertising partnership
  • March 26, 2026: Genius Sports presents Moment Engine at its NewFronts session
  • May 2026: Genius Sports publishes MediaScience research on recall around live moments

Summary

Who. Leagues own the underlying events and license them. Sportradar, Genius Sports and Stats Perform collect and distribute the feeds. Sell-side platforms including Magnite and PubMatic convert them into advertising signals. Demand-side platforms and agencies buy against those signals; broadcasters and streamers deliver the impressions.

What. A sequential account of every action in a sporting event, existing both as spoken broadcast narration and as a structured data record carrying timestamps, event type codes, participant identities and resulting game state.

When. The broadcast form dates from August 1921. The structured data form was assembled from 1984 onward through Project Scoresheet and Retrosheet, automated from 2006, and turned into an advertising trigger between July 2025 and March 2026.

Where. In the venue where events are logged, in the provider infrastructure that normalises and distributes them, inside sell-side curation platforms, and finally in the live stream where the resulting advertisement is inserted.

Why. Live sport concentrates simultaneous attention in a way little else does, and play-by-play is the only signal precise enough to locate the seconds within a match where that attention peaks. It also sets the limits of the practice, since the data outruns the picture and its commercial value rests largely on evidence supplied by the companies selling it.