- Build each onboarding video around one first win, and place it at the step where your analytics show users stalling, not on the welcome screen.
- Keep every onboarding video under about 90 seconds and open it with the finished result, so the viewer knows the payoff before they commit attention.
- Measure activation with a randomized holdout, not by comparing people who watched against people who didn't, because viewers are already more motivated.
The worst place for SaaS onboarding videos is the welcome screen, and it’s where most SaaS teams put them first. Someone signs up at four in the afternoon because they need to send one invoice, set up one dashboard, or import one spreadsheet. Then the product greets them with a four-minute overview of reporting, team roles and integrations, and they close the tab. The video was well made. It just answered a question nobody had yet.
Onboarding video works when it arrives at the moment a user is holding a specific question, and it fails when it asks them to learn the whole product before they have a reason to care. This is a playbook for the first kind. It covers how to find the step where users actually stall, how to build a short video around one task, where to place it, and how to measure whether it changed anything.
Why the welcome tour loses
A product tour is a map. New users need directions. A map is useful when you already know where you’re going, and at signup most people don’t. They have one goal and no yet-to-be-earned patience, so every minute spent on features they won’t use this week reads as friction, not help.
Three things make the welcome tour worse than it looks on paper. It’s generic, so the viewer can’t tell which parts apply to them, and generic content is easy to skip. It usually autoplays with sound, or starts with a slow logo sting, which signals “this will take a while.” And it sits between the user and the thing they came to do, so the cost of skipping feels low and the cost of watching feels high.
Consider what the user is holding at that moment. They have a task in mind and a browser tab open to finish it. A video that asks them to switch from the task to a general learning mode is asking for a decision they haven’t made yet. Most people will defer it, and deferred onboarding has a way of never coming back.
The fix isn’t a shorter tour. It’s a different unit of content altogether: one video per task, placed where the task stalls.
Start from the first win, not the feature list
The useful question is not “what should new users learn about the product?” It’s “what is the smallest outcome that proves the product is worth their time?” Call that the first win. For an invoicing tool it might be a sent invoice. For a dashboard it’s a connected data source with a chart on screen. For a support tool, it’s a reply sent from a shared inbox.
Finding the first win and the path to it is a short exercise that most teams skip:
- Write the first win as one sentence. “A user sends one invoice to a real client address” is a usable first win. “Users understand our billing system” is not, because you can’t observe it.
- Walk backwards from the win. List every action between signup and that outcome, in order. For an invoicing tool that might be: create account, connect a bank account, add a client, create an invoice, send it. Five steps, sometimes more.
- Mark where users actually stall. This is the step most teams guess at and should measure. Funnel analytics show where the drop-offs cluster. Support tickets show the same thing from another angle, because the people who stall are the people who write in.
- Make one video per stall point. Cover only the actions that point requires. If the stall is at “add a client,” the video doesn’t need to cover invoices.
Here’s a concrete version. Imagine an invoicing product where most users reach “add a client” and then about half of them abandon before creating an invoice. The analytics show the drop is concentrated in the line-item editor, which is hidden behind a button labeled with an icon rather than a word. That single stall deserves a 70-second video, and it’s worth more than a six-minute tour of the whole product, because it targets the exact confusion that’s costing you activations.
Placement decides more than production quality
A beautifully produced video in the wrong place will get fewer views than a rough one in the right place. Placement works because of the trigger: the viewer already has the question, so the video answers rather than sells. Here are the placements that tend to work, along with the length each one can tolerate.
| Placement | Trigger | Length target | Example |
|---|---|---|---|
| Empty state | User lands on a page with nothing in it yet | 45 to 75 seconds | “Create your first project” on an empty projects list |
| Checklist item | User opens the setup checklist and picks a step | 60 to 90 seconds | “Connect your data source” |
| Contextual help | User opens help while on a specific screen | Under 2 minutes | “Set up roles and permissions” on the roles page |
| Follow-up email | User hasn’t reached the first win 24 hours after signup | Under 90 seconds, one link | “Send your first invoice in 90 seconds” |
The empty state is the most underused. A blank list is where a user is most likely to feel lost, and a short video there replaces an empty screen with an obvious next move. Put the video directly in the empty space, not behind a help icon, because nobody goes looking for help on a page they think is working.
The follow-up email needs a discipline most teams don’t apply. Send one link, to one video, with the outcome in the subject line. Two links and a feature list turn a helpful message into a newsletter, and newsletters get archived.
Build each video around one task
Once you know the stall point, the video’s format is mostly mechanical, and the mechanics are what make a series look consistent. Every onboarding video I’d ship follows the same three-part shape.
First, show the end state in the first three seconds. The finished invoice, the connected dashboard, the sent reply. This answers the question “is this going to be worth it?” before the viewer has spent any attention. Second, walk through three to five steps, each one with the cursor and the zoom locked on the control being used. Third, close with one next action, not a menu of options. “Now add your first client” is an action. “Explore the other features” is a shrug.
Production should be boring on purpose. Lock the browser window size and zoom, use the same demo account across the series, and name each file by task and version, such as invoice-line-items-v2. When every video uses the same window and zoom, the same button sits in the same place across the series, and a viewer who watched one video recognizes the next one immediately. That’s a real comprehension gain, and it costs nothing but a saved setting.
Timing matters as much as structure. For a 70-second video, a workable split is about five seconds for the outcome, about fifty seconds for the steps and about ten seconds for the next action. The steps are the variable part. If a single step needs more than fifteen seconds of on-screen work, it’s usually two steps that should be separate videos. Write the timings into the script, then record against them. A video that runs long is almost always one step too many, and the fix is to cut the step, not to speed up the narration until it sounds rushed.
Keep captions on by default. Onboarding videos are often watched at a desk, next to other work, with the sound off. A caption track is cheap insurance, and it helps people who are watching in a second language or a noisy room.
Put the outcome in the first three seconds: a viewer who can see the finished result knows what the next minute is for. A viewer who sees a logo and a title card is already deciding whether to leave.
Measure the watch, then the activation
Video analytics will tell you how many people pressed play, how far they got and where they left. Those numbers are useful for fixing a broken video, but they’re the wrong thing to optimize for. The metric that matters is whether the user reached the first win, within a window you define, such as three days.
Here’s the trap. If you compare people who watched the video with people who didn’t, the watchers will look better, and you’ll be tempted to credit the video. But watchers were already more motivated. They clicked the help link because they were engaged, and engaged users activate at higher rates anyway. The comparison is measuring who chose to watch, not what the video did.
The fix is a holdout, and it’s simple to run:
- Pick one trigger. Use one placement, such as the empty state on a single screen.
- Randomize at the user level. When a user hits the trigger, assign them to see the video or not, using a stable random split.
- Compare activation between the two groups. The difference, not the watch rate, tells you whether the video moved anything.
Consider a rough illustration with invented numbers. Suppose 400 users hit a stall each week, and 120 of them press play. If 54 of those watchers go on to create an invoice, that’s 45 percent. If 61 of the 280 non-watchers create one, that’s about 22 percent. It looks like the video doubled activation. A randomized split might show a real gain of a few points, or none at all, and that’s the number you should act on, because it’s the one that survives when you scale the placement to every user.
Run the holdout for at least a couple of weeks, long enough to cover your weekly rhythm, and keep the split fixed while you run it. Changing the video mid-test resets the comparison.
The mistakes that quietly kill good videos
Most failed onboarding videos fail for reasons that have nothing to do with the film itself. A few patterns recur.
Autoplay with sound. It feels attention-grabbing in a design review and feels hostile in a real browser. Users close the tab, and some learn to distrust the video element on your site. Let the viewer press play.
Gating. Forcing users to watch a video before they can continue creates resentment and gives you nothing you can measure cleanly. If the video is required to use the product, it’s a tutorial hiding inside a paywall, and users will find the workaround or leave.
The overview disguised as onboarding. A ten-minute video labeled “Getting started” is a product tour in a new coat. Split it into tasks, or cut it.
Videos that never get updated. This is the one that matters most over time, and it deserves its own warning.
Stale video is worse than none: a video that shows a button you renamed teaches users the wrong thing and sends them to support. Tie each video to a version of the interface and re-record it when that screen changes.
The maintenance habit is what separates a library that works after a year from one that’s a liability. Give every onboarding video an owner, list the screens it depends on, and check that list at each release. A re-record of a 70-second task video is a small job, and it’s far cheaper than the support tickets a stale one generates. Teams that already track repeat questions can use the same list to decide which videos to refresh first.
What good SaaS onboarding videos look like in practice
Put the pieces together and the final system is small. A handful of task videos, each under about 90 seconds. Each one opens on the finished result and closes with a single next step. Each sits at a specific stall point, shows up in the empty state or the checklist rather than on the welcome page, and gets captions by default. A randomized holdout tells you whether it moved activation, and a standing review keeps it current with the interface.
None of that requires a film crew. It requires knowing which step your users fail on and having a way to record that step cleanly, the same way, every time. Teams that already build tutorials people finish have most of the skill already. The onboarding version just needs to be shorter, tied to one moment, and measured against activation rather than views.
If you want to record a task video without spending an afternoon on zoom and cleanup, RecordlyPro records in your browser with nothing to install, and adds auto-zoom on every click, cursor effects, captions and chapters for you. Start with the one stall that costs you the most activations this month. Record that one task, put it in the empty state, and run the holdout before you decide whether to build the next one.
Keep reading
Async video: stop scheduling the meeting you could record
A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.
Sofia Meyer
October 6, 20262 min readAuto-zoom in screen recordings: what it is and why your demos need it
A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.
Sofia Meyer
October 13, 20262 min readHow to choose a screen recorder for product demos in 2026
A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.
Sofia Meyer
October 16, 20262 min readDofollow vs nofollow in 2026: what still passes value
The link attributes changed, Google's treatment changed, and a lot of advice is out of date. Here's what dofollow, nofollow, sponsored and ugc actually do now.
The Recordly Team
October 10, 20262 min read