Why this matters
You've built a product. You've written the scripts. You've shot the onboarding videos. Now you're about to ship them to 10,000 users who will decide in the first 90 seconds whether your product is worth their time.
Most teams guess. They watch the videos internally, ask a few colleagues, maybe run a survey. Then they push live and hope the activation metrics don't crater.
EmotionTrac gives you second-by-second facial expression data from real people watching your onboarding sequence. You see exactly where confusion spikes, where attention drops, where frustration builds. You fix those moments before your users ever see them.
The workflow
Step 1: Upload your onboarding sequence
Export your onboarding videos as MP4 files. Upload them to the EmotionTrac platform in the order users will see them. If you have multiple versions (A/B tests, different feature sets, localized content), upload each variant as a separate study.
The system accepts videos up to 10 minutes each. Most onboarding sequences run 2 to 5 minutes total across 3 to 6 clips.
Step 2: Define your panel and launch
Choose your audience criteria: geography, age, tech proficiency, industry background. EmotionTrac recruits opt-in panelists who match your target user profile.
Panel sizes typically range from 50 to 200 respondents per study. Larger panels cost more but give you tighter confidence intervals on the emotion data.
The platform schedules sessions, sends invitations, and collects responses. Panelists watch your videos on their own devices while their webcams capture facial expressions using the Facial Action Coding System (FACS). No one manually codes faces; the system processes everything automatically.
Step 3: Review the emotion timeline
Once responses come in, you get a timeline graph showing seven core emotions (joy, surprise, sadness, anger, disgust, fear, contempt) plotted second by second. Each emotion appears as a separate line. Spikes and valleys show you exactly when viewers felt what.
Look for patterns. Does confusion (a mix of surprise and fear) spike when you introduce a new interface element? Does joy drop off halfway through a feature demo? Does anger appear when you mention pricing or permissions?
Click any point on the timeline to jump to that moment in the video. Watch what was on screen, what the voiceover said, what UI element appeared. The cause is usually obvious once you see the data.
Try it: Schedule an EmotionTrac demo and see second-by-second emotion tracking in action. Or visit Innovation for more information.
Step 4: Identify drop-off risk zones
Pay attention to sustained negative emotion. A brief spike of confusion can be fine if it resolves quickly. A 20-second stretch of low engagement (flat lines across all positive emotions) means people are tuning out.
Compare emotion data to your internal goals for each segment. If you want users to feel confident about security at the 1:30 mark, check whether joy and surprise are rising or whether fear and disgust are climbing instead.
Cross-reference with any survey questions you included. If 40% of respondents say they don't understand a feature but the emotion data shows high joy during that segment, the disconnect tells you something about how people rationalize their feelings versus what they actually experience.
Step 5: Edit and retest
Make changes based on what you learned. Shorten the confusing section. Add a visual callout where attention dropped. Rewrite the script where anger spiked.
Upload the revised version and run a new study with a fresh panel. Compare the emotion timelines side by side. Did the edits move the needle? Did you accidentally create new problems?
Repeat until the data shows consistent positive engagement through the entire sequence. Most teams run 2 to 4 iterations before they're satisfied.
What you get
You get a heatmap of emotional response tied to specific timestamps. You know which 10 seconds are killing your activation rate before a single real user sees them.
You get demographic breakdowns. Maybe your onboarding works great for users under 35 but confuses everyone over 50. Maybe one gender responds well to your tone while the other finds it off-putting. You can't fix what you can't see.
You get comparative data if you test multiple versions. The platform shows you which variant generated more positive emotion, less confusion, better sustained attention. You're not guessing which version to ship.
You get a record you can share with stakeholders. Executives and investors understand emotion graphs. When you show them a spike in anger at the pricing reveal or a drop in joy when the UI gets cluttered, they grasp the problem immediately. No need to argue about subjective opinions.
Common pitfalls
Testing too late. If you wait until the week before launch, you won't have time to make meaningful changes. Start testing rough cuts as soon as you have a working draft.
Ignoring the panel composition. If your product targets enterprise IT buyers but you test with college students, the data won't predict real-world performance. Match your panel to your actual user base.
Overreacting to single data points. One person's spike in anger doesn't mean your video is broken. Look for patterns across the full panel. Outliers happen.
Testing only one version. You learn more from comparing two approaches than from testing a single cut in isolation. Even if you think you nailed it, test an alternative. You might be wrong.
Forgetting that emotion isn't the only metric. High joy doesn't guarantee comprehension. Low fear doesn't mean people will actually complete the onboarding steps. Combine facial coding with behavioral data (click-through rates, task completion, time to first value) to get the full picture.