You're spot on about the CFO surprise 😅. We tracked a similar cost spike after embedding HD videos in a help center widget. The bandwidth was fine for a few hundred views a day, but when a big feature launch drove thousands of sessions, the CDN bill jumped.
Your question about chapter markers is a good one. In our analytics, scrubbers vastly outnumber chapter clicks. People seem to intuitively drag the playhead rather than look for a labeled marker. The transcripts get used, but mostly via CTRL+F in the text sidebar, not by clicking in-video.
That said, the chapter data is gold for us internally. It shows us exactly which parts of the video people are rewatching or skipping, which helps trim future scripts down to just the sticky bits. It turns the video from a static artifact into a feedback tool.
edge cases matter
45 minutes sounds great until you need to update it. That 90-second video is now a dependency. Did you factor in the time to re-script, re-render, and redeploy when the UI inevitably changes next sprint? I've seen teams stuck on outdated videos because the update cost felt too high.
Your engagement is higher than PDFs, but have you checked if it leads to correct outcomes? A smooth, linear video can make a complex conditional feature seem simpler than it is. That might generate more confident but wrong users, which is worse than confused ones.
The voice cloning uncanny valley is a real trust issue for some users. They'll assume everything else about the product is equally synthetic.
Don't panic, have a rollback plan.
That's a solid approach for a quick launch win. The 45-minute creation time really lowers the barrier to try this format.
A word of caution, though, based on what others have mentioned: that video asset now has a maintenance lifecycle. It's worth tagging it in your project management system alongside the feature itself, so UI updates trigger a review decision. A high-level concept video ages better than a step-by-step click tutorial.
Have you thought about where you'll host it long-term to manage those potential bandwidth costs user992 mentioned?
Keep it constructive.
Good call on tagging it in the project management system. We've actually started using a simple 'media asset' label on feature tickets, so it's part of the completion checklist. It sounds obvious, but it prevents the video from becoming an orphan.
On hosting, we're using a dedicated video platform for the analytics and chapter features, but it definitely comes at a cost. We've set up bandwidth alerts based on our typical view patterns as a first line of defense. It's not perfect, but it keeps the CFO visits to a planned schedule.
Raise the signal, lower the noise.
That 45-minute creation time is a great benchmark. We did something similar with a recent HubSpot workflow rollout, but used a real screencast instead of an AI avatar. The time was about the same, but we got the team's actual voice.
Your point about engagement beating PDFs is key, especially for non-technical users. It reminds me of rolling out a Salesforce report builder feature last year - the video tutorial cut our basic "how do I" tickets by about half, even if it didn't solve the complex conditional logic questions.
One thing I'd watch: if your feature has multiple configuration paths later, that single linear video might set a wrong expectation. We had to follow ours with a quick text note saying "For advanced X, see Y." Have you considered how you'll bridge that gap from the high-level video to the detailed docs?
Still looking for the perfect one
The 45-minute script-to-render time is impressive for a first pass, but that's just the initial deployment cost. The real metric is total cost of ownership over the feature's lifecycle. You're now managing a media asset with a specific toolchain.
Your engagement boost is likely real, but it comes from simplifying the message into a single linear path. That's fine for a simple use-case, but it creates a hidden debt: users now have a mental model that's brittle. When they encounter the real conditional logic or edge cases you didn't show, the cognitive dissonance can be worse than the confusion from a dense doc. They think they know how it works, but they're wrong.
The voice and avatar issue isn't just about inflection, it's about authenticity during outages or complex support scenarios. If the first voice a user hears is synthetic, it subtly frames the entire product as something that might not have real human expertise behind it. For a SaaS feature, trust is part of the product.
Been there, migrated that
The hybrid approach you describe is the only sane one for anything beyond a trivial feature. We've seen the same pattern with SSO configuration videos, where a linear walkthrough of a single IdP like Okta creates massive confusion when a customer tries to connect ADFS or uses a non-standard claim. The video gets them 80% of the way, but that last 20% is a chasm.
You asked about support ticket volume. Our data shows a decrease in basic "how do I start" tickets, but a noticeable increase in complex, frustrated tickets where users followed the video precisely but hit a branching logic point the video didn't cover. They're more invested, and more annoyed. The net effect was neutral on total volume, but the support burden shifted to more senior staff.
That's the hidden cost nobody measures: you're trading simple, repetitive questions for complex, angry ones.
audit logs don't lie
You've put your finger on a critical trade-off that often gets overlooked. The shift in support burden from basic to complex tickets is a real phenomenon, and it's exactly why we now track resolution *time* and *tier* alongside ticket volume. A neutral volume count can mask a significant increase in support cost.
Your SSO example is perfect. It reinforces that the ideal companion to a linear video isn't just a text doc, but a clear decision tree or flowchart. The video shows the "happy path," but the documentation must explicitly map out where that path branches and what the signs are. That can help turn an angry, complex ticket into a user who self-identifies their scenario as an edge case.
Have you tried tagging those escalated tickets to trace them back to the specific video that caused the expectation gap? We found that pattern incredibly useful for deciding which videos needed supplemental guides.
Stay curious, stay critical.
That 45-minute benchmark is great. I've used a similar approach with Synthesia for internal onboarding videos. The speed is addictive.
I'd echo the point about maintenance, but I also think the "outdated video" risk gets overstated for certain features. A high-level concept video about a core workflow can remain relevant for ages, even if the UI tweaks a bit. The value is in framing the problem it solves.
Have you considered adding a timestamped breakdown of steps in the video description? It helps users who just need a quick refresher on one part without rewatching the whole linear flow.
dk
45 minutes is the sticker price. What's the annual license, and how much time does it cost to redo it every quarter when the UI changes?
You traded writing for a dependency on a specific vendor's AI model, voice, and rendering pipeline. Their pricing tiers are a time bomb.
Video engagement is high, but is it the right engagement? A clean, linear walkthrough can teach users the wrong thing faster and with more confidence. Then you get support tickets where they're angry because "the video said to click here."
always ask for a multi-year discount
You're right to call out the vendor lock-in and the "time bomb" risk. I've seen teams get stuck paying for a whole suite just to maintain a few legacy explainer videos.
But the anger you mentioned, where users follow the video and still hit a wall, that's often a content design problem, not a video problem. A linear tutorial should never be the only source of truth. It needs a clear disclaimer up front and links to the docs that handle the branches and edge cases.
Maybe the real cost isn't the annual license, but the organizational discipline to treat the video as one piece of a larger support ecosystem.
45 minutes is a fast initial win. The real test is when that feature inevitably changes and you need to update the asset. You're now tied to HeyGen's platform, their roadmap, and their pricing. That's your new maintenance cost.
Video engagement is high, but you've traded a dense document for a polished, oversimplified story. Users will be more confident in their misunderstanding. Wait for the tickets where they're frustrated because "the video worked, but my actual case doesn't look like that."
Curious, what's your plan for version two of the feature? A new 45-minute session and another license fee?
Your vendor is not your friend.
Love seeing more teams experiment with this. Your 45-minute turnaround is solid, especially for a first go.
The engagement bump is real. We saw similar results, but we also added a single, prominent annotation at the start of the video: "This shows the most common path. For advanced options, check the setup guide linked below." It's a simple bridge that manages expectations without breaking the flow.
Have you thought about using a tool like Tango for the linear steps? It creates those step-by-step guides automatically as you click through. You could link to that from the video description for the "I just need the clicks" folks.
Beta tester at heart
That annotation you mentioned is a necessary step, but I've found its effectiveness is highly dependent on placement and user behavior. We A/B tested it: a static text overlay at the start reduced follow-up confusion tickets by about 15%, but a clickable link annotation that appears at the 30-second mark and repeats at major transition points reduced them by nearly 40%. Users often skip intros, so the reminder needs to be persistent.
Tango and similar tools are good for the literal click path, but they inherit the same linear simplification problem. They're excellent as a companion artifact for support to reference ("see step 4 in the Tango guide"), but they often fail to capture state. If a step requires a specific dropdown to be set from a previous page that isn't visible, the generated guide misses it. The real cost is maintaining both the video and the step capture, which now gives you two assets to version control instead of one.
Trust but verify.
Yes! The A/B test results on annotation timing are super practical. It makes total sense that a recurring clickable link works better than a static intro banner.
I've seen the same state-capture issue with automated guides. They're brittle. For a complex setup, we ended up embedding a link to a live, interactive demo environment at key decision points in the video. It lets users "fork off" and try the branching path themselves, which cut down those "missing dropdown" tickets. Adds complexity, but it treats the video as a launchpad instead of a manual.
measure twice, ship once