How to Link Video Viewing to Retention: Setting Up Video Analytics for SaaS

For three months, I told everyone our onboarding video was working.

I had the numbers to prove it. Time-to-value dropped from 14 days to 4. Activation rate jumped from 22% to 41%. Churn fell 34%. Those numbers were real. I wasn't lying.

But I couldn't prove that the video caused any of it. Correlation isn't causation. Maybe the video worked. Maybe the product got better. Maybe the market shifted. Maybe the cohort that received the video was different from the cohort that didn't. I had no way to isolate the variable. I was telling a story I believed, not a story I could defend.

Then our board asked a simple question: "How do you know the video is what moved these numbers?" I didn't have an answer. I had a narrative. Narratives don't survive due diligence.

That's when we decided to build a proper analytics stack—one that could link video viewing behavior to retention outcomes at the user level. Not aggregate. Not correlational. Individual. User by user. Play by play.

This is the story of how we did it. The tools we chose. The events we tracked. The reports we built. The dashboards we now show every quarter. And the mistakes that cost us six weeks of bad data. We did this work with LeoStudio, whose approach to clarity you can explore at leostudiohq.com. They were the ones who told us from the start that a video without measurement is a guess. We didn't listen. This is what we learned when we finally did.

The Problem: We Were Measuring the Wrong Things

Before the analytics stack, we measured video performance the way most teams do. We looked at views. We looked at average watch time. We looked at completion rate. Those metrics told us whether people were watching. They didn't tell us whether watching mattered.

The gap between "watched" and "mattered" is where most SaaS teams live. You know your video gets plays. You don't know if the plays are changing behavior. And if you can't connect the play to the behavior, you can't optimize the video. You're flying blind with a dashboard.

The dashboard we had was worse than useless. It gave us a false sense of confidence. We'd see a spike in views and assume the video was working. We'd see a dip and assume the video was failing. Neither assumption was grounded in anything real. We were reacting to noise.

We needed to answer questions like these:

Does a user who watches the video to completion activate faster than a user who watches 10 seconds?

Does a user who watches on mobile behave differently from a user who watches on desktop?

Does a user who watches the video twice retain better than a user who watches once?

Does the user who watches the video in the welcome email behave differently from the user who watches it in the app?

None of those questions could be answered with the aggregate metrics we had. They required user-level data. They required linking the viewing event to the behavioral outcome. They required a proper analytics stack.

This is a distinction that LeoStudio emphasizes constantly—video is not content, it's an intervention. And you can't measure an intervention with view counts. You measure it with outcomes. You can see that philosophy throughout their work at leostudiohq.com. It's the reason we stopped treating the video as a marketing asset and started treating it as a product feature with its own instrumentation.

The Stack: Tools We Chose and Why

We evaluated four tools. Each one served a different purpose. Here's what we landed on and why.

Product Analytics: Amplitude

We chose Amplitude as our primary product analytics platform. It's designed for user-level event tracking and cohort analysis, which is exactly what we needed. We could send custom events from our video player and then build cohorts based on whether users watched the video, how much they watched, and when they watched.

The key feature for us was cohort analysis. We needed to isolate users who watched the video from users who didn't, and then compare their retention curves. Amplitude's cohort tables are built for exactly that kind of comparison. You define an inclusion criterion—say, "watched onboarding video"—and a return metric—say, "logged in on day 7"—and the tool shows you the retention curve for that cohort versus the baseline. Research on cohort analysis confirms this is the standard approach for measuring how video viewing relates to retention.

Video Analytics: Wistia

We used Wistia for video hosting and player-level analytics. Wistia's JavaScript Player API let us capture granular events—play, pause, 25% watched, 50% watched, 75% watched, complete—and pass them to Amplitude as custom events.

The advantage of Wistia was the granularity. We could see exactly where viewers dropped off, not just that they dropped off. That data fed directly into our editing decisions. When we saw a 30% drop-off at the 45-second mark, we knew exactly which section needed work.

Data Pipeline: Segment

We used Segment to route events from Wistia to Amplitude. Segment acts as a single pipe for all our event data, which meant we didn't have to build custom integrations for each tool. When a user watched the video, Wistia fired an event, Segment caught it, and Amplitude received it with the user's identity attached.

The critical piece here was identity resolution. We needed to connect the anonymous video viewing event to the identified user in Amplitude. Segment handled that by passing the user ID through the event payload.

Dashboards: Mixpanel and Amplitude

We used Amplitude for retention cohorts and Mixpanel for engagement dashboards. Mixpanel's retention reports are designed for exactly the kind of analysis we needed—tracking how many users return after an initial event, and how that return rate changes over time.

We built a shared dashboard that showed four things: video watch depth, activation rate by watch depth, retention by watch depth, and churn by watch depth. That dashboard became the single source of truth for every conversation about the video.

The tool choice matters less than the event design. You can build this stack with Amplitude, Mixpanel, PostHog, or any combination. What matters is that you're tracking the right events and connecting them to the right outcomes. That's a principle LeoStudio drilled into us from the first call—the video is only as good as the data you collect around it. You can read more about their approach to measurement at leostudiohq.com.

The Events: What We Track and Why

The events are the foundation. If you track the wrong events, no tool will save you. Here's the event schema we built.

Event 1: Video Play

Event name: onboarding_video_play

Properties: video_id, video_title, trigger_source (welcome_email, empty_dashboard, inactivity_modal), device_type, user_id

Why we track it: The play event is the baseline. It tells us who started the video. But on its own, it's nearly meaningless. We use it as the denominator for every subsequent metric. What percentage of plays turn into completions? What percentage of plays lead to activation?

Event 2: Video Progress (25%, 50%, 75%)

Event names: onboarding_video_25, onboarding_video_50, onboarding_video_75

Properties: video_id, user_id, seconds_watched, percent_watched

Why we track it: Progress events tell us where viewers drop off. This is the single most useful metric for editing. When we saw that most viewers dropped at 45 seconds, we knew the section that started at 40 seconds was the problem. We cut it. Completion rate went up. These are the "heartbeat" events that many video platforms send at regular intervals to track watch depth.

Event 3: Video Complete

Event name: onboarding_video_complete

Properties: video_id, user_id, total_watch_time, completion_rate

Why we track it: Completion is the closest proxy we have for "the user got the message." A user who watches to the end has seen the one action and the one win. A user who drops at 30 seconds hasn't. We use completion as the primary segmentation variable in all downstream analysis.

Event 4: Activation

Event name: first_value_reached

Properties: user_id, time_to_value_days, activation_source

Why we track it: This is the bridge event. Activation is the moment the user connects a data source and sees their first insight. It's the outcome the video is designed to produce. Every analysis we run links video events to this event.

Event 5: Retention (Weekly Login)

Event name: weekly_active_user

Properties: user_id, cohort_week, activation_status

Why we track it: Retention is the ultimate outcome. We define retention as a weekly login. It's a crude proxy for "the user is still getting value," but it's the metric our board cares about. We link every video event to retention through the user ID.

The event design is the hard part. The tools are easy. If you get the events wrong, you'll spend months building dashboards that tell you nothing. If you get them right, you'll have answers in a week. LeoStudio helped us think through the event schema before we wrote a single line of tracking code. That conversation saved us weeks of rework. If you're planning a video analytics project, start there—at leostudiohq.com—before you touch a dashboard.

The Reports: What We Actually Show

We built four reports. Each one answers a different question. Each one is designed for a specific audience.

Report 1: Watch Depth vs. Activation Rate

Question: Does watching more of the video correlate with activating faster?

How we built it: We segmented users into four groups based on watch depth—0-25%, 25-50%, 50-75%, 75-100%—and calculated the activation rate for each group within seven days.

What we found: The relationship was almost perfectly linear. Users who watched 0-25% activated at 18%. Users who watched 25-50% activated at 27%. Users who watched 50-75% activated at 34%. Users who watched 75-100% activated at 41%. The more they watched, the more likely they were to activate.

What it means: The video isn't just correlated with activation. It appears to be a driver. The dose-response relationship is strong evidence that the video is causing the change, not just accompanying it.

Who sees it: The product team. This report tells them whether the video is doing its job and where the video needs to improve.

Report 2: Completion Rate vs. Retention

Question: Do users who watch the video to completion retain better than users who don't?

How we built it: We created two cohorts—users who completed the video and users who didn't—and compared their 30-day and 60-day retention curves in Amplitude.

What we found: Users who completed the video retained at 2.3x the rate of users who didn't, at both 30 days and 60 days. The gap didn't narrow over time. It held steady.

What it means: Completion isn't just a proxy for engagement. It's a predictor of retention. A user who watches the whole video is a user who is going to stay.

Who sees it: The leadership team. This report is the one that justifies the video investment. It's the report that says, "This is not a marketing expense. It's a retention lever."

Report 3: Trigger Source vs. Activation

Question: Does where the user sees the video matter?

How we built it: We segmented users by trigger_source—welcome email, empty dashboard, inactivity modal—and compared activation rates for each group.

What we found: The empty dashboard trigger performed best, with a 41% activation rate. The inactivity modal was second at 36%. The welcome email was third at 31%. All three were above the 22% baseline, but the differences were significant.

What it means: Placement matters. The same video, in different places, produces different outcomes. The empty dashboard—where the user is staring at a blank screen—is the highest-impact placement.

Who sees it: The growth team. This report tells them where to invest in distribution.

Report 4: Video Events vs. Churn (Cohort Analysis)

Question: Do users who watch the video churn at a lower rate?

How we built it: We used Amplitude's cohort analysis to create a cohort of users who watched the video and a cohort of users who didn't. We then tracked their churn over 90 days. This is the standard approach for linking an event to a retention outcome—define the inclusion criterion, define the return metric, and compare the curves.

What we found: Users who watched the video churned at 34% less than users who didn't, over 90 days. The effect was strongest in the first 30 days and persisted through the full period.

What it means: The video isn't just a temporary boost. It's a structural change in the composition of the user base. Users who watch the video are the kind of users who stay.

Who sees it: The board and investors. This is the report that closes the loop. It links the video to the number that matters most—revenue retention.

The Dashboards: What We Watch Weekly

We built a single shared dashboard in Amplitude. It has four panels. We review it every Monday morning.

Panel 1: Watch Depth Distribution. A bar chart showing what percentage of viewers watch 0-25%, 25-50%, 50-75%, and 75-100%. This is the health check. If the distribution shifts left—more people watching less—something is wrong with the video or the trigger.

Panel 2: Activation Rate by Watch Depth. A line chart showing the activation rate for each watch depth bucket. This is the proof of impact. If the line is flat, the video isn't working. If the line slopes up, it is.

Panel 3: Retention Curve by Completion Status. Two lines—completers and non-completers—tracking retention over 60 days. This is the long-term proof. It's the chart that shows the video isn't just a short-term nudge.

Panel 4: Churn by Trigger Source. A bar chart showing 90-day churn for users who saw the video in the welcome email, the empty dashboard, and the inactivity modal, compared to a baseline of users who saw no video. This is the placement optimization chart. It tells us where to put the next video.

The dashboard is the single source of truth. Every conversation about the video—every decision to edit it, move it, or replace it—starts with this dashboard. We don't argue about opinions. We argue about the charts.

This is the discipline that LeoStudio brings to every project—the video is never finished until the data says it is. You can see that same rigor in how they approach production and measurement at leostudiohq.com. If you're building your first video analytics stack, read how they think about the full lifecycle before you design your events.

The Mistakes That Cost Us Six Weeks

We didn't get this right the first time. Here are the mistakes that cost us the most.

We didn't set a user ID on the video player. The first version of our tracking sent anonymous events. We could see that 1,000 people watched the video. We couldn't see which users watched. We had to rebuild the entire tracking pipeline to pass the user ID through to every event. Six weeks of data were unusable.

We tracked plays, not progress. Our first event schema only tracked play and complete. We couldn't see where viewers dropped off. We couldn't see which sections were losing attention. We were flying blind on the editing side. Adding 25%, 50%, and 75% events was the single biggest improvement to our analytics.

We forgot to segment by trigger source. We tracked the video as a single entity. We couldn't tell whether the email video or the in-app video was working. Once we added trigger_source, the picture changed. The email was underperforming. We shifted investment to the in-app placement.

We measured views first. For the first month, we celebrated views. It was the easiest metric to look at. It was also the least useful. Views don't move retention. Watch depth moves retention. We wasted a month optimizing for the wrong number.

We didn't build the retention cohort until week three. We had the video events. We had the retention data. We hadn't connected them. Once we built the cohort analysis, everything clicked. The tool was capable of the analysis from day one. We just hadn't asked it the right question.

We forgot mobile. Again. Our first dashboard didn't segment by device. When we added the device_type property, we saw that mobile users watched less and activated faster when they watched at all. The mobile video was doing the heavy lifting. We just hadn't been measuring it.

The Framework: How to Build Your Own Video Analytics Stack

If you're starting from scratch, here's the framework we now use.

Step 1: Define the outcome first. What behavior is the video supposed to change? Activation? Retention? Feature adoption? Define the outcome before you define the event.

Step 2: Design the event schema. What events do you need to track? At minimum: play, 50% progress, complete, and the outcome event. Add properties for video ID, trigger source, device type, and user ID.

Step 3: Choose the tools. You need a video platform that fires events, a pipeline that routes them, and a product analytics tool that can build cohorts. Wistia, Segment, and Amplitude worked for us. Any equivalent combination will work.

Step 4: Build the cohort analysis first. Don't build dashboards. Build the cohort comparison. Watchers versus non-watchers. Completers versus non-completers. That's the analysis that proves the video works.

Step 5: Add the diagnostic reports. Watch depth distribution. Activation by watch depth. Churn by trigger source. These are the reports that help you optimize, not just prove.

Step 6: Review weekly. The dashboard should be a ritual, not a report. If you don't look at it every week, you'll miss the drift. The video decays as the product changes. The dashboard is how you notice.

Step 7: Refresh the video when the data says so. The analytics stack isn't just for proving the video worked. It's for knowing when the video has stopped working. When the watch depth distribution shifts left, it's time to edit. When the activation curve flattens, it's time to re-script. When the churn gap narrows, it's time to re-produce.

The Final Word

For three months, I told everyone our onboarding video was working. I believed it. I just couldn't prove it. Then we built the analytics stack, and the story changed from a narrative to a fact.

The video works. Users who watch it activate faster. Users who complete it retain better. Users who see it in the empty dashboard activate at the highest rate. Those aren't opinions. They're measured outcomes. They're the reason the video survived every budget review, every reorganization, and every skeptical stakeholder.

If you're trying to link video viewing to retention, don't start with the tools. Start with the outcome. Then design the events. Then build the cohort. Then review the dashboard. The tools are easy. The discipline is hard. But the discipline is what turns a video into a business case.

We built our stack with LeoStudio, a team that understood from the first call that the video was never the point—the outcome was. If you want to see how they think about measurement and clarity in onboarding video, spend time on leostudiohq.com. Read their case studies. Look at how they frame the problem. Then build your own stack. Then prove your own video works.

Tell them the SEO guy sent you. Tell them you already built the cohort. They'll know exactly what you mean.