Amplitude for Streaming Services

Why Most Streaming Platforms Misread Their Churn Signal

Your data is there. The problem is you're looking at the wrong moments.

Most streaming teams track cancellations. Amplitude lets you work backward from cancellation to the exact behavioral pattern that predicted it — weeks earlier. That's the difference between reacting to churn and preventing it.

This guide covers how to configure Amplitude specifically for a subscription streaming service: which events matter, how to build segments that actually predict behavior, and which automations to prioritize first.

---

Setting Up Your Event Taxonomy

Get this wrong and every insight downstream is compromised. Streaming services have a specific set of events that, when tracked consistently, give you a complete picture of subscriber value.

Core Events to Instrument

Playback events are your foundation:

• `content_started` — include properties: `content_id`, `content_type` (series, film, documentary, live), `genre`, `language`, `season`, `episode_number`
• `content_completed` — add `watch_percentage` as a property; anything under 80% is a meaningful signal
• `content_abandoned` — timestamp + position where the user stopped; critical for identifying content-quality problems
• `autoplay_triggered` and `autoplay_cancelled` — autoplay cancellation spikes often precede churn by 14–21 days

Account and subscription events:

• `trial_started`, `trial_converted`, `trial_expired`
• `subscription_upgraded`, `subscription_downgraded`, `subscription_cancelled`
• `payment_failed`, `payment_retried`, `payment_recovered`
• `plan_page_viewed` — how often users visit pricing before cancelling

Discovery and search events:

• `search_performed` with a `results_returned` property (zero-result searches are a catalog gap signal)
• `browse_category_viewed`, `recommendation_clicked`, `recommendation_ignored`
• `watchlist_added`, `watchlist_removed`

Session-quality events:

• `playback_error` — include error code and device type
• `quality_changed` — user-initiated vs. adaptive; frequent adaptive drops signal infrastructure issues affecting retention
• `subtitle_enabled`, `audio_language_changed` — localization signals that help you understand engagement depth

User Identity and Properties

Set these as user properties in Amplitude, not event properties:

• `subscription_plan`, `billing_cycle`, `account_age_days`
• `primary_device` (mobile, TV, desktop, tablet)
• `content_preference_genre` (derived from watch history, updated weekly via a computed property)
• `household_size` (for family/shared plans)
• `days_since_last_stream` — your single most predictive churn variable

---

Segments That Actually Matter

Generic segments — "active" vs. "inactive" — don't give you enough to act on. Build these instead.

The At-Risk Subscriber Matrix

Combine two variables: recency (days since last stream) and engagement depth (average watch completion rate). This creates four segments:

1. Engaged and Recent — streamed in last 7 days, average completion above 70%. Your core. Protect them.
2. Engaged but Lapsed — high completion rate historically, haven't streamed in 15–30 days. Most recoverable with a targeted content hook.
3. Recent but Shallow — streamed recently but abandoning content under 30% completion. Catalog mismatch or content quality issue. Investigate before they leave.
4. Lapsed and Shallow — no streams in 30+ days, historically low completion. Highest churn risk. Intervention cost versus LTV calculation required here.

Build these as Amplitude Cohorts using behavioral conditions, then sync them directly to your CRM or messaging platform via the Amplitude Data Destinations integration.

Trial Conversion Segments

The 14-day trial window is where you win or lose a subscriber. Build cohorts based on:

• Days to first stream after signup — users who stream within 24 hours convert at 2–3x the rate of those who wait 72+ hours
• Content type at first stream — if a user's first stream is a live event or a flagship series, track whether that content type correlates with higher 90-day retention
• Number of distinct content titles in the first 7 days — users who sample 5+ titles during trial typically churn at half the rate of single-title viewers

---

Funnels and Retention Analysis

The Activation Funnel

Define activation for a streaming service precisely. A common mistake is treating "first stream" as activation. It isn't.

True activation for most streaming platforms looks like: streamed at least 3 separate sessions across 2 different content categories within the first 10 days.

Build this as a multi-step funnel in Amplitude:

1. `account_created`
2. `content_started` (session 1)
3. `content_started` (session 2, different `content_type` property)
4. `content_started` (session 3, at least 2 days after session 1)

Track your activation rate from this funnel weekly. A 5-point improvement in activation typically translates to 8–12% improvement in 90-day retention.

N-Day Retention Analysis

Use Amplitude's Retention Analysis chart set to N-Day Retention, not unbounded. Streaming services should watch:

• Day 7 retention — tells you if your onboarding is working
• Day 30 retention — tells you if your catalog depth is sufficient
• Day 90 retention — your real LTV predictor

Segment these retention curves by `subscription_plan` and `primary_device`. You will almost always find that TV-app users retain significantly better than mobile-only users — 15–25% better in most platforms we've seen. That has product implications.

---

Industry-Specific Challenges in Amplitude

Handling Anonymous and Logged-Out Sessions

Streaming services often have a browse-before-login pattern. Anonymous users browse, then sign up. Amplitude's Identify call lets you alias anonymous IDs to authenticated user IDs at the moment of account creation — but you must implement this correctly. If you don't, you'll split a single user journey across two profiles and break your activation funnel analysis.

Work with your engineering team to confirm the `amplitude.setUserId()` call fires immediately on successful authentication, not on page load.

High Event Volume and Sampling

A mid-size streaming platform generating 50M+ events per month can hit Amplitude's ingestion limits depending on your plan tier. Do not sample on the client side. Instead, work with Amplitude's data management settings to filter low-signal events server-side before ingestion — things like redundant `heartbeat` or `buffer_status` events that your infrastructure team may already be firing.

Content Catalog Freshness

Amplitude doesn't natively know when new content was added to your catalog. Pass `content_release_date` as an event property on every playback event. This lets you build analyses comparing engagement on new releases vs. catalog content — a critical metric for content acquisition decisions.

---

Automations to Build First

Prioritize these three Amplitude-triggered automations via your data destinations:

1. Lapsed trial alert — user reaches day 10 of trial without completing a second session → trigger an email or push featuring content from a genre they browsed
2. Payment failure recovery — `payment_failed` event fires → suppress cancellation survey trigger for 48 hours and route to payment update flow instead
3. Binge completion follow-up — user completes a full series season within 72 hours → trigger recommendation for next series within same genre before they enter idle state

---

Frequently Asked Questions

How do I measure the impact of a new content release on retention in Amplitude?

Create a user segment of subscribers who watched the new title within its first 7 days of release. Then compare their 30-day and 60-day retention curves against a matched cohort who didn't watch it. Use Amplitude's Personas or a simple cohort comparison in the Retention chart. This tells you whether tentpole content is actually retaining subscribers or just spiking viewership without changing behavior.

What is the right "active user" definition for a streaming service in Amplitude?

Avoid using session starts as your active metric. Define an active user as someone who completed at least 20 minutes of content in the measurement period. Set this as a custom active user definition in your Amplitude project settings. This eliminates accidental sessions, autoplay-without-watching behavior, and login-only visits from your active count.

Can Amplitude handle multiple profiles for shared streaming accounts?

Amplitude tracks by device ID and user ID, not by profile within an account. If your platform supports multiple profiles under one account (a common structure for family plans), you'll need to pass a `profile_id` as a user property on every event. This lets you segment by profile without conflating behavior across household members — essential for accurate genre preference and recommendation analysis.

How do we use Amplitude to reduce involuntary churn from failed payments?

Track `payment_failed` and `payment_recovered` as explicit events with a `days_to_recovery` computed property. Build a cohort of users who experienced payment failure and recovered within 7 days vs. those who didn't. Analyze what behavioral differences preceded the failure. In most platforms, involuntary churn accounts for 20–40% of total cancellations — isolating and addressing it separately from voluntary churn is one of the highest-ROI analyses you can run in Amplitude.