What problem does it solve?
Broadcasters run hundreds of live events across multiple leagues with a digital production team that cannot watch every feed at once, let alone clip, reformat and publish a moment before it stops being timely. Fox Sports says nearly 90% of its Digital content is consumed vertically, yet the traditional workflow, monitoring broadcasts across multiple screens, manually identifying key moments, clipping and reformatting them for vertical social platforms, does not scale to that volume without automation.
The moments still happen whether or not a broadcaster has the staff to capture them: a goal, a save, a controversial call. Without a scalable way to detect and clip those moments as they happen, the window to reach fans on social platforms in near real time closes, and the opportunity to grow audience and engagement around that moment goes unrealized.
How does it work?
- Analyse the live feed continuously. Computer vision and audio analysis watch the broadcast stream in real time and detect key moments, such as a goal or a celebration, within seconds of them happening on air.
- Harvest the clip automatically. A detection event triggers a pipeline that extracts the matching segment from a rolling buffer of the live stream and reformats it, for example cropped to 9:16 vertical video for social platforms.
- Surface it for review. The clip appears automatically in a web portal with AI generated tags and a short description, so an editor can find, search and filter incoming clips as the broadcast continues.
- Edit and assemble. Editors trim, merge or discard clips in a visual timeline, and can combine clips from across a broadcast into a highlight reel.
- Distribute. Finished clips and reels are published to the organization's fan facing channels; editor feedback on clip quality feeds back into tuning what the system detects.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Mainstream
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | at least 1000 | 1 | 1 organization |
Value drivers: Speed and cycle time, Employee productivity, Revenue growth.
Indicative value
A regional sports broadcaster covering 200 live events a year
USD 8333 to USD 106,667
Editor time cost avoided on live clipping per year
How this is calculated
Formula: liveEvents * clipsPerEvent * editorMinutesPerClipManual * automationShare / 60 * editorCostPerHour. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Live events covered per year liveEvents, events per year | 200 | 200 | The reference broadcaster. |
| Highlight clips produced per event clipsPerEvent, clips per event | 20 | 40 | Editorial assumption based on a typical number of clippable moments per match. Replace with your own. |
| Editor minutes to find, cut and format one clip by hand editorMinutesPerClipManual, minutes per clip | 10 | 20 | Editorial assumption for manual clipping and reformatting from a live feed. Replace with your own time study. |
| Share of that time an automated detection and clipping pipeline removes automationShare, fraction of editor time per clip | 0.5 | 0.8 | Editorial assumption, conservative against the evidence on this page (AWS describes a prototype it built for Fox Sports that surfaces clips in a review portal within 20 to 30 seconds of the moment occurring on air, and WSC Sports reports the NBA produces over 1,000 highlight packages in a few minutes), because neither source states a prior manual baseline in minutes per clip that this share can be checked against directly. |
| Fully loaded cost of a video editor editorCostPerHour, USD per hour | 25 | 50 | Editorial assumption for a broadcaster digital production role. Replace with your own. |
What it leaves out: Gross editor time cost avoided only. It leaves out the cost of the AI service and integration, the editor time still spent reviewing and publishing every clip, and any extra advertising or engagement revenue faster clips may generate.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Fox Sports
United States · Media and entertainment · 2026
Fox Sports worked with AWS, which built a solution for Fox Sports on its managed AWS Elemental Inference service (AWS calls it a prototype) that continuously analyzes a live sports broadcast, detects key moments such as goals and celebrations within single digit seconds, and automatically extracts, crops to 9:16 vertical video and surfaces the resulting clip in a review portal within 20 to 30 seconds of the moment happening on air. In a quote on the page, a Fox Sports executive describes it as having evolved from a hackathon concept into a production ready, machine learning driven solution integrated into the organization's live production and distribution workflows across multiple major sports leagues, and says the move responds to nearly 90% of Fox Sports Digital content being consumed vertically.
No outcome disclosed.
National Basketball Association (NBA)
United States · Media and entertainment · 2025
WSC Sports, whose platform automatically detects moments in games and cuts them into highlight clips for leagues and broadcasters, publishes a quote on its own site describing highlight package production for the NBA sped up to a few minutes for more than 1,000 packages, framed as letting the league create tailored content for every digital platform it operates on.
- Interactions handled: at least 1000, in a few minutes
"WSC Sports enables us to create unique content for every single digital platform that we touch globally. Now it takes a few minutes to create over 1,000 highlight packages."
Claimed by: organization
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A live video feed with enough delay or a rolling buffer to extract clips from
- A taxonomy of the moments worth clipping per sport, such as a goal, a penalty or a celebration
- Historical footage labelled with those moments, to train or tune detection
Systems to integrate
- Live video ingest and packaging (encoder, rolling buffer)
- Video transcoding for vertical reformatting and multiple output profiles
- A review and editing portal for the digital production team
- Distribution endpoints for social and app publishing
Complexity: Medium
Detecting a key moment in a live video feed and turning it into a broadcast quality vertical clip within seconds needs a low latency video pipeline, ingest, buffering, inference and transcoding, alongside the detection model itself. The harder work is usually the live production integration and the review portal, not the detection model.
- 1
Start with one sport and a short list of moments
Pick the highest volume sport or league and define which moment types matter, such as a goal or a red card, before expanding detection to other sports.
- 2
Build the live pipeline before the model
Get ingest, the rolling buffer and transcoding solid on real broadcast feeds first; the detection model only creates value if a clip can actually be produced within the broadcast window.
- 3
Put a review portal in front of every clip
No clip reaches a fan facing or social channel without an editor confirming it, tagging it correctly and checking it matches the moment.
- 4
Measure against the manual process it replaces
Track how many clips a human team produced per event before automation, and compare volume and editor time after.
- 5
Expand sport by sport
Each sport has different moments and pacing. Tune detection and add new moment types deliberately rather than assuming one model covers every sport equally well.
Guardrails
- Editor review before any AI detected clip reaches a fan facing or social channel
- A fixed, agreed list of moment types the system may detect, so it does not invent unofficial moments
- Rights and licensing checks before clips leave the organization's own channels, since a highlight is still copyrighted broadcast content
- Clear internal labelling of which clips were AI detected, so quality issues can be traced back to the pipeline
KPIs to instrument
- Clips produced per live event, before and after
- Time from the moment happening to the clip being ready for review
- Editor time spent per clip, before and after
- Share of AI detected clips an editor rejects or heavily edits
Human in the loop
Editors review every AI detected clip in the portal before it is published, decide which clips become part of a highlight reel, and can trim, merge or discard a clip that was wrongly tagged or badly framed.
Common failure modes
- Missed or duplicate moments
- A fast paced passage of play can be detected twice or missed entirely if the model's moment definitions are too broad or too narrow. Track detection precision and recall against a labelled sample of real broadcasts, not just the volume of clips produced.
- Clips leave the review portal unpublished
- A high clip count looks like an automation win in a dashboard, but only if the clips get published. Instrument publish rate per event, not only clips detected.
What are the risks and rules?
EU AI Act
Minimal risk
Detecting a sporting moment in video the broadcaster already owns the rights to, and cutting or reframing a clip from that real footage, is not listed in Annex III, so it stays minimal risk. The system AWS describes does generate synthetic text: AI generated tags and a short AI generated description of each moment for the internal review portal. That falls inside Article 50(2)'s scope for synthetic content, but the marking duty there sits with the AI system's provider, and the outputs stay internal for an editor to check before anything reaches a fan facing channel, so the practical exposure stays minimal. Cropping real footage to a vertical frame is standard editing, not the kind of content generation Article 50 targets.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Article 50 covers several separate duties: 50(1) disclosure when a person interacts with an AI system, 50(2) marking AI generated or manipulated synthetic audio, image, video or text so it is machine readable and detectable, 50(3) informing people exposed to emotion recognition or biometric categorisation, and 50(4) labelling image, audio or video content that is a deepfake, or AI generated text published to inform the public on matters of public interest, unless it goes through human editorial review and control. The AI generated tags and moment descriptions in this use case fall under 50(2), but that duty sits with the provider and the outputs stay internal to the review portal rather than reaching fans. Reformatting real footage into a vertical crop is standard editing, not synthetic generation. This would change if a deployment started publishing AI generated descriptions or commentary directly to fans, or generated synthetic video or voice rather than only extracting real broadcast footage; that would trigger 50(2) or 50(4) directly for the deployer.
Controls to put in place
- Editor sign off before publication of any AI detected clip
- A documented, agreed list of moment types the model may detect
- Rights and licensing review before any clip leaves the organization's own channels
- Regular sampling of published clips against the source broadcast for accuracy
- AI generated tags and moment descriptions stay internal to the review portal; only text an editor writes or approves reaches fans
Frequently asked questions
- How fast can AI turn a live sports moment into a clip?
- AWS describes a prototype it built for Fox Sports that detects a key moment within single digit seconds and surfaces the finished, reformatted clip in a review portal within 20 to 30 seconds of the moment happening on air. AWS says it plans to publish a reference implementation; how a production deployment performs depends on the live video pipeline's own latency, not only the detection model.
- Does automated clipping replace video editors?
- Not at Fox Sports. AWS describes a portal where Fox Sports editors review, search, tag, edit and distribute clips before anything reaches a fan facing channel. WSC Sports reports that automation lets the NBA produce over 1,000 highlight packages in a few minutes, but the NBA quote we found says nothing about editorial review, so we cannot say whether or how the NBA reviews clips before publishing them.
- Is automated sports highlight clipping high risk under the EU AI Act?
- Usually not. Extracting or reframing a clip from a broadcast the organization already owns the rights to is not an Annex III use, so this stays minimal risk. The AI generated tags and moment descriptions used inside the review portal fall under Article 50(2)'s synthetic content marking duty, but that duty sits with the AI system's provider, and the outputs never reach fans directly. It would raise the risk if a deployment started publishing AI generated text, video or audio to fans rather than only feeding it to editors reviewing real footage.
- Which sports and leagues use this today?
- AWS names Fox Sports for live vertical clipping across multiple leagues, and WSC Sports shows logos of the NBA, La Liga, MLS, DAZN, the NHL and several other leagues and broadcasters among its clients, though not every named client discloses a metric.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for automated sports highlights and clipping", last verified 30 September 2026, https://www.blits.ai/ai-use-cases/automated-sports-highlights-and-clipping. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: First published