> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trackplay.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Analytics

> See the exact second they leave, which traffic pays, and whether the buyers watched differently.

Play counts and average watch time never told you the one thing that mattered: which
second moved the buyer. These tabs do.

Open any video and you get six tabs.

<CardGroup cols={2}>
  <Card title="General" icon="gauge">
    Your metric cards, each with a sparkline and a trend. Fully customisable. See
    [Custom metrics](/analytics/custom-metrics).
  </Card>

  <Card title="Retention" icon="chart-area">
    The drop-off curve, second by second.
  </Card>

  <Card title="Funnel" icon="filter">
    Page view → play → 25/50/75% → end → sale.
  </Card>

  <Card title="Sources" icon="coins">
    Which traffic actually pays.
  </Card>

  <Card title="Quality" icon="gauge-high">
    Whether the player is losing you sales.
  </Card>

  <Card title="Embed" icon="code">
    Your embed snippets.
  </Card>
</CardGroup>

## Retention

The curve of who is still watching, and when they left.

Two things make it more than a pretty chart:

**Buyers vs non-buyers.** Overlay the two curves. If your buyers watched past 4:10 and
your non-buyers did not, you have found the moment your video sells. That is the moment
to protect.

**Markers.** Your CTAs and conversion density are drawn onto the timeline, so you can see
whether the drop-off lands before or after the ask.

<Tip>
  The flat answer to "what should I cut?" is the cliff on this chart. Find the second
  where the line falls off, watch it, and cut what you find.
</Tip>

## Funnel

Where the viewer stopped, at every stage from page view to sale: a rate at each step,
and one overall.

A high play rate and a low 50% rate is a hook problem. A high 75% rate and a low
conversion rate is an offer problem. The funnel tells you which one you have.

## Sources

Which traffic pays.

| Metric   | What it is                 |
| -------- | -------------------------- |
| **EPC**  | Revenue per click          |
| **RPV**  | Revenue per play           |
| **AOV**  | Average order value        |
| **ROAS** | Revenue ÷ affiliate payout |
| **Net**  | Revenue − payout           |

Broken out by affiliate, by referring domain, and by traffic source.

<Note>
  ROAS is `null` (not zero) when no payout was captured for that source. A zero would
  read as "this source lost money", which is a different claim from "we do not know what it
  cost".
</Note>

You can also break any metric down by **campaign, ad set, ad, and creative**, so a
winning creative is visible as a creative and not only as a spike in the total.

## Quality

A slow player kills the sale before the hook lands. This tab tells you if that is
happening.

<AccordionGroup>
  <Accordion title="Speed">
    Time to first frame and startup time, at the median and the 95th percentile. The p95
    is the one that matters. It is the experience of your worst-served viewers, and they
    are the ones who leave.
  </Accordion>

  <Accordion title="Stability">
    Rebuffer rate, rebuffer ratio, dropped-frame rate.
  </Accordion>

  <Accordion title="Failure">
    Playback failure rate, exits before start, and a breakdown of the errors behind them.
  </Accordion>

  <Accordion title="Autoplay">
    How often autoplay actually started, and how often viewers unmuted.
  </Accordion>
</AccordionGroup>

A KPI with no samples reads **"No data"**, never `0%`. A zero would be a lie: it would
say "nothing went wrong" when the truth is "nothing was measured".

Errors are classified so the tab shows genuine failures, not noise. Auto-recovered
events (a buffer nudge, a stall the player fixed by itself) are collected under
**Diagnostics** rather than counted against you.

## Traffic quality

How much of your traffic is human.

On the video's General tab and on the workspace dashboard: the split between human, bot,
datacenter and VPN sessions, with a country breakdown.

<Warning>
  Bot and datacenter detection is reliable. **VPN detection usually is not**: it needs an
  IP-reputation add-on, and without it most VPN traffic shows up as datacenter or human.
  The card says so where you read it.
</Warning>

## Audience and hot leads

**Audience** in the sidebar lists identified viewers. Click one for a full profile:
lifetime watch time, every video they touched, their events, their conversions, their
revenue.

**Hot leads** scores viewers 0-100 on watch behaviour: did they reach the offer, did they
unmute, did they come back, did they click. Cold, warm, or hot.

<Note>
  The score is a transparent heuristic, not a black box. It is built from signals you can
  see on the viewer's own profile, so you can always check its work.
</Note>

## Export it

Your data is yours.

* **Webhooks** push events to you as they happen.
* The **[Analytics Export API](/api-reference/authentication)** pulls summaries,
  retention curves, raw events and conversions, with a token carrying `analytics:read`.
* Bulk exports run as a job and hand back a download URL.