Video analytics
Video analytics, in streaming, is the measurement of how each video is delivered and watched. A small SDK in the player sends events such as play, playing, pause, seek, stall and error, and a backend groups them into views and turns them into metrics: startup time, rebuffering, bitrate, errors and watch time. Those metrics are broken down by device, country, network, CDN and title, so teams can see where playback fails and what people watch. The same phrase also names computer vision on camera footage, a separate field.
How does video analytics work?
It starts inside the player. An analytics SDK listens to the player's own events and sends them, with timestamps, to a collection service as small messages called beacons. Each beacon carries context: the video, the viewer ID your app passes in, the device, browser, player version and network.
The backend stitches those events into views. A view runs from the moment playback is requested until the viewer leaves, and every metric is calculated per view: how long the first frame took, how many times playback stalled and for how long, which renditions played, whether an error ended it. Views are then aggregated across dimensions such as country, ASN, CDN, device model or title, and exposed through dashboards, APIs and alerts.
The per-view design matters. An average across all traffic hides the one device model or mobile carrier that fails every time. Per-view data lets you find it.
What metrics does video analytics track?
Metrics fall into three groups. Quality of experience, or QoE, describes what the viewer felt: video startup time, exits before the video started, playback failures, rebuffer count and rebuffer ratio (time stalled divided by time watched), average bitrate and upscaling, how often a low rendition had to be stretched to fill a larger player.
Quality of service, or QoS, describes the network underneath: request latency and throughput for each segment, and which CDN served it. QoS explains why QoE moved. A rise in stalls on one ASN with falling throughput points at delivery, while stalls with healthy throughput point at the player or the device.
Engagement describes what people chose to watch: views, unique viewers, watch time, how far into the video they got and where they dropped off. Read engagement next to QoE. A drop-off at the same minute in every view is usually content. A drop-off that follows a stall is delivery.
Video analytics vs. web analytics
Web analytics tools track pages, clicks and sessions. They can tell you that someone opened a page with a video on it, and perhaps that they clicked play. They cannot see what happened inside the player afterwards: whether the first frame appeared, how long it took, how often it stalled, what quality it played at or which error ended it.
CDN logs miss the same things from the other side. They record bytes sent per request, but not whether those bytes turned into smooth playback. A segment can look fast in the logs and still reach a viewer after their buffer has run dry.
Video analytics sits between the two, collecting events from the player itself. That is why it can link a stall to a specific CDN, internet provider and device model, and why it is the data source for decisions about bitrate ladders, CDN routing and player changes.
How do you implement video analytics?
Start by covering every player surface you ship. A typical product has a web player, an iOS app, an Android app and perhaps a TV app, each on a different engine such as hls.js, Shaka Player, AVPlayer or ExoPlayer. If analytics only covers the web, the mobile and TV problems stay invisible.
Next, pass in metadata. Video title, series, content type, a pseudonymous viewer ID and a few custom fields specific to your product make the data usable. A course platform might tag course and cohort; a news site might tag desk and story.
Then decide how the data will be used. Set alerts on a few key metrics, such as playback failures and rebuffer ratio for your largest regions, so problems surface before support tickets do. If the business already runs a data warehouse, export event-level data so playback quality can be joined with sign-ups, retention or completion.
How does FastPix handle video analytics?
FastPix runs video analytics as Video Data. A player SDK collects each view and sends it to a dashboard and a Data API that returns views, metrics, breakdowns, errors, time series and comparisons. It is not tied to the FastPix Player. It covers 17 player surfaces: FastPix Player plus 11 third-party player SDKs, including Shaka, hls.js, Video.js and dash.js on web, ExoPlayer and AVPlayer on mobile, plus Samsung Tizen, LG webOS, Roku, React Native and Flutter. It works with or without FastPix delivery.
Each session carries 56 dimensions, 46 built in plus 10 custom, covering geography, ASN, CDN, device, browser, player, video details, DRM type and connection type. Per-view metrics include video startup time, player initialization time, page load time, buffer count, buffer ratio, dropped frames, request latency and throughput, average bitrate, upscaling, seek latency, exits before video start, errors with code and context, and live stream latency.
Every session gets an overall QoE score: 0.395 x stability + 0.330 x render quality + 0.275 x startup. You can drill from a low score to the event that caused it, break QoE down by country, region, CDN and ASN, send threshold alerts to PagerDuty, Slack or Opsgenie, and export event-level data to ClickHouse, Snowflake, BigQuery, Redshift or Databricks.
The first 100,000 plays a month are free, every month, with 30-day retention. Above that, the Growth plan is a fixed monthly price for a block of plays, then a rate per thousand plays. Start by adding the SDK to the player you already run.
