Designed for the agent-led era.
Onboarding lives in one place now.
Most SaaS products have six different help tools. A tour. A chat widget. A help center. A feedback button. A support form. And now, an AI chat. They don't talk to each other. Users get lost before they get started.
Agent-led onboarding puts all of it in one place: the same AI as the rest of your product. The 17 patterns below show what that looks like, grouped by where you're losing users.
Start where it hurts most.
Activation
1. Aha moment shortcuts
Most onboarding flows are designed around teaching the product. The ones that actually convert are designed around getting the user to a single specific moment of value, fast.
That moment has a name in growth circles: the aha moment. For Slack it's sending 2,000 messages. For Dropbox it's uploading the first file. For Facebook it was 7 friends in 10 days. Once a user hits that threshold, retention curves bend permanently upward. Before it, they're at risk of being a one-and-done signup.
The mechanism is the Peak-End Rule. People judge an experience by its most intense moment and how it ended. If your most intense moment is a 12-screen setup wizard, you have a problem. If it's a small win in the first 90 seconds, you have a product.
A number that stuck with me from a Rocketlane survey in 2025: 83% of B2B buyers say slow onboarding is a dealbreaker. Slow is relative; what they really mean is "I haven't felt anything yet."
What seems to work when shortening the path:
What I've seen go sideways:
- Designing the onboarding to introduce every feature. Most features can wait until session two.
- Requiring data import before first value. People drop off during the CSV upload.
- Hiding the aha moment behind a paywall or "talk to sales" gate.
The shift in posture I keep coming back to: onboarding isn't a tour of your product, it's the first session of using it. Treat session one as the demo, not as a precursor to the demo.
2. Contextual tooltips
Most onboarding tours fail because they teach things people don't need yet. The fix isn't a shorter tour. It's deleting the tour.
The pattern that works is contextual tooltips. Instead of an upfront slideshow, you show one short hint at the exact moment someone tries to use a feature. A Figma overlay that appears when you first land on the canvas. A Slack tip when you create your first channel. Same lesson, but it lands when it's relevant instead of two minutes before it's relevant.
There's a name for why this works: the Spacing Effect. People retain information better when they learn it at the moment of need, not all at once at the start.
Some numbers I came across in a Jimo case study. Genially switched from email-based feature announcements to in-app contextual tooltips and saw activation rise by 25%, with no product changes underneath.
What seems to work when building these:
What I've seen go sideways:
- More than two tooltips firing in the same session. Users tune out.
- Tooltips on features the user hasn't even looked at yet. Just-in-time means in-time, not just-in-case.
- Generic descriptions that read like a help article instead of a hint.
The shift in posture I keep coming back to: an upfront tour teaches people what your product does. A tooltip teaches people what they're already trying to do. Those are very different jobs, and only one of them actually sticks.
3. Personalized onboarding paths
Most onboarding flows treat every new user the same way. The data on this is brutal: average SaaS activation rates sit around 30%.
The fix is asking 2 to 3 questions at signup and routing each user through a different version of onboarding based on the answers. Canva asks "what will you be using this for" and reshapes the entire empty state. Mailchimp asks about email marketing goals and recommends specific first actions. Same product, different first 5 minutes.
The mechanism is the Cocktail Party Effect. People pay attention to things that feel personally relevant. A demo board that mentions their role lands harder than a demo board built for everyone.
Some data from Moxo's 2025 onboarding report: personalized onboarding flows lift retention by 40% versus generic flows. Clevry put time-to-productivity at 52% faster.
What seems to work:
What I've seen go sideways:
- Building twelve personas and one onboarding flow. The personas exist on paper and not in the product.
- Asking long questionnaires before the user has felt anything. Trust hasn't been earned yet.
- Storing the answers in a database and never actually changing the UI based on them.
The shift in posture I keep coming back to: your overall activation rate is a fiction. The real numbers are the per-segment ones, and most products are quietly leaving 20% of activation on the table because they're showing every user the same default flow.
4. Empty state design
The most dangerous screen in your product is the one users see immediately after signup. The one with no data, no content, no idea what to do next.
Wyzowl ran a study where 80% of users said they'd deleted an app because they didn't understand how to use it. In my experience, that confusion almost always traces back to one screen: the empty dashboard.
The mechanism is Blank Page Paralysis. The same reason writers struggle with a blank document but can edit a draft. An empty product looks broken even when it's working as designed. An empty product with a template feels like permission to start.
Airtable is the canonical example here. New users never see a blank spreadsheet. They see a row of templates (Marketing Calendar, Product Roadmap, Project Tracker) with sample data already populated. The "Start from scratch" option is there but it isn't the default. Most users pick a template, see immediate structure, and feel productive within thirty seconds.
What seems to work:
What I've seen go sideways:
- Empty states with motivational copy and no action. "Start your journey!" doesn't help anyone start anything.
- Sample data the user can't delete. Feels like the product is haunted.
- Empty states that look identical to error states. Users assume something broke.
The shift in posture I keep coming back to: an illustration plus a CTA feels intentional. A blank table feels like the product is broken. The difference is one design pass and a few hundred words of copy.
5. Onboarding checklists
Most onboarding checklists fail for the same reason that most tours fail: they're built for the team building them, not the user using them.
The version that actually works is short, visible, and pre-checked on the first item. Sked Social documented a 3x lift in conversion after switching to a 4-item checklist with the first task pre-completed (account creation). Pre-checking sounds like a parlor trick, but it taps a real psychological force called the Endowed Progress Effect. People are more likely to finish a task they've already started than one they haven't.
There are actually three mechanisms stacked here. The Zeigarnik Effect, where incomplete tasks nag at us. Goal clarity, where a small finite list outperforms a vague "explore the product" instruction. And the Endowed Progress thing on top.
What seems to work:
What I've seen go sideways:
- 12-item checklists. Users see the length and bail before starting.
- Tasks that don't actually link anywhere. "Configure your settings" with no link is just nagging.
- Checklists that stay visible forever, even after the user finishes them. Becomes wallpaper.
The shift in posture I keep coming back to: a checklist isn't a to-do list, it's a contract with the user about what activation looks like. Four specific completions and they get value. Twelve vague ones and they get fatigue.
6. Interactive product tours
Static product tours don't work very well. Interactive ones do, and the gap is bigger than I expected.
A static tour is the "click Next to learn about projects" thing. Tooltips, arrows, a guided slideshow. An interactive tour skips the narration and just has people do the action. Create the project. Name it. See it appear. Same lesson, but they're using the product instead of watching a demo of it.
There's a name for why this works that I actually like: the Generation Effect. People remember things better when they generate them than when they're shown them. It's why flashcards beat re-reading and why a sandbox beats a screenshot.
Some data I came across in a Jimo report recently. Across about 1,000 tours, median completion was 15%. The interactive ones hit 44%. Roughly 3x.
What seems to work when building one:
What I've seen go sideways:
- Tours longer than 7 steps. People bail around step 4.
- Generic placeholder data. Breaks the spell instantly.
- Forcing the tour on everyone. Power users opening a second account don't need the welcome wagon.
The shift in posture I keep coming back to: a tour isn't how you teach people your product. It's how you get them to their first real win. If they're not doing anything in there, they're not learning. They're just sitting through a demo with extra steps.
Retention
7. Behavior-based triggers
Time-based onboarding emails have a problem. They fire on a clock, not on a context. Day 3 reminder. Day 7 nudge. Day 14 check-in. They land regardless of whether the user has done anything since signing up.
Behavior-based triggers work the opposite way. They fire when a specific action happens (or doesn't happen) inside the product. Loom prompts you to share your first video the moment you finish recording one. Grammarly waits until you've written a sentence with a clarity issue, then surfaces the upgrade pitch. Same action, different timing, very different outcomes.
There's a Greek word for this: Kairos. The opportune moment. A behavior-triggered message lands much harder than the same message on a clock-based schedule. Jimo puts the lift at roughly tenfold.
What seems to work:
What I've seen go sideways:
- Triggering off vague events like "logged in 5 times." Not actually behavioral; just a usage threshold dressed up.
- Sending the same triggered message every time the event fires. Frequency caps matter.
- Triggers that nag users to do things the product has already done for them.
The shift in posture I keep coming back to: drip campaigns send messages on a schedule. Behavior-based triggers send messages on a signal. The signal is always more interesting than the schedule.
8. Time-delayed nudges
Behavior-based triggers fire when something happens. But sometimes the most important signal is when nothing happens.
If a user has been staring at the same screen for two minutes without clicking anything, they're not reading. They're stuck. The right move is a small, specific nudge. Not "need help?" but "Having trouble connecting your CRM? Here's the Salesforce integration guide."
The mechanism is loss aversion, used positively. The user has already invested time on this page. A relevant nudge reduces the perceived cost of continuing past the friction. Walking away feels more wasteful than clicking the help link.
A small case study I came across from The Room, a hiring platform: they added a 90-second inactivity nudge to their CV upload flow and saw CV uploads climb 75% in ten days, from around 210 a week to 350. The nudge was the only change.
What seems to work:
What I've seen go sideways:
- Firing the same nudge on every page. Reads as a chatbot that has nothing to say.
- Nudges that link to a 40-article help center instead of the one specific article.
- Forcing nudges on users past their first 14 days. Veterans don't need them and find them patronizing.
The shift in posture I keep coming back to: silence is data. A user who hasn't clicked in two minutes is telling you something. The product gets to decide whether to listen.
9. Social proof in context
Most products treat social proof as a homepage thing. Customer logos, big numbers, testimonials. Once a user is inside the product, the proof disappears.
The version that actually moves usage is inline social proof, shown at decision points inside the product. Slack does this with "X people from your company are already here" right next to the channel join button. Frase shows "Join 30,000 content teams" inline with the feature it actually applies to. Duolingo turns the entire streak system into an ambient social proof engine.
The mechanism is informational social influence. When uncertain what to do, we look at what people similar to us are doing. The closer the comparison, the more useful the proof.
A finding from Forrester and Adobe: experience-driven businesses see twice the growth in retention and lifetime value than competitors. A lot of that experience advantage is just visible social proof at the moments when users are deciding whether to take an action or back out.
What seems to work:
What I've seen go sideways:
- Numbers that obviously include inactive accounts. Users can smell it.
- Irrelevant proof. A B2B PM doesn't care that consumer hobbyists love your tool.
- Proof as a modal interrupting the user. Proof should be ambient, not in the way.
The shift in posture I keep coming back to: a homepage logo wall is for convincing buyers. Inline social proof is for convincing users. They're different jobs and most products only do the first one.
10. Friction logging
Most product teams know roughly where users drop off. They almost never know why.
The reason is that almost no one asks while it's happening. Quarterly NPS surveys happen too late. Exit interviews happen too late. By the time anyone fills out a form, the user has already rationalized their way out of the product.
A 2024 Clevry study put it plainly: only 12% of users rate their onboarding as effective. Most companies don't know why, because they never ask during the experience.
The fix is contextual microsurveys. A single question, fired the moment friction shows up, dismissible. Grammarly does this on the Analytics page. Superhuman does it at key feature moments. The data they get back isn't just sentiment, it's specific narrative ("I couldn't figure out how to connect my calendar"). That kind of input goes straight into the backlog.
What seems to work:
What I've seen go sideways:
- Five-question forms dressed up as "quick surveys." Users notice and ignore.
- Surveys that only fire for power users. The people you most need to hear from are the ones already half-gone.
- Collecting answers and never closing the loop with the user. People stop responding the second they realize nothing changes.
The shift in posture I keep coming back to: usage data tells you where users dropped off. Microsurveys tell you why. Both are necessary; most teams only invest in the first one.
11. Comparative benchmarks
Vanity dashboards tell users they're using the product. Comparative benchmarks tell users where they actually stand. The first kind makes people feel busy. The second kind makes people change behavior.
HubSpot built one of the better examples of this. The Benchmark Dashboard shows your team's email open rates, conversion rates, and engagement next to "companies like yours" of similar size and industry. Suddenly the numbers have meaning. A 24% open rate looks fine in isolation. Sitting next to a 38% peer average, it looks fixable.
The mechanism is Social Comparison Theory. Festinger documented this in 1954. We evaluate ourselves by comparing to others, especially others similar to us. The product that shows that comparison is the product that drives behavior.
What seems to work:
What I've seen go sideways:
- Comparisons that shame. "You're in the bottom 10%" demotivates more than it motivates.
- Benchmarks based on irrelevant peer groups. A 5-person startup compared against enterprise teams isn't useful, it's discouraging.
- Showing the gap with no path to closing it. Comparative data needs an action attached.
The shift in posture I keep coming back to: a usage metric is a mirror. A comparative metric is a lever. Most products show the mirror and call it analytics; the products that actually shape behavior show the lever.
Expansion
12. Hotspots and pulsing dots
Static interfaces become invisible. Users learn the shape of your product in the first two weeks, then their eyes stop scanning for new things. Anything you ship after that is sitting in the UI without being seen.
The fix is small, intentional motion. A pulsing dot on a new or underused feature. A subtle "New" badge. Anything that breaks the visual habituation people have already built.
Typeform uses this on builder features that users haven't tried. Productboard pulses dots on contextually relevant features and pairs them with one-sentence tooltips. Miro tags new tools with "New" badges in the toolbar. In each case, the dot disappears the moment the user engages with the feature, so it doesn't become permanent noise.
The mechanism is attentional capture. Our eyes are drawn to movement and novelty. A static UI becomes invisible; a small pulse pulls the eye to one specific spot.
What seems to work:
What I've seen go sideways:
- Pulsing dots on every feature simultaneously. The signal collapses.
- Dots that fire for every user, including users who already use the feature.
- Animations loud enough to disrupt the rest of the UI. Subtle pulses outperform aggressive ones.
The shift in posture I keep coming back to: feature adoption isn't a marketing problem, it's an attention problem. Most products have plenty of features and zero ways to direct attention to them. A pulsing dot solves more of that than another announcement email.
13. Inline announcements and banners
Most product teams announce new features by email. Then they're surprised when adoption is low. The math is simple: average SaaS email open rates hover around 20 to 25%. That means roughly 75 to 80% of your users never see your feature announcement.
The version that works is announcing inside the product, in the surface where the feature actually lives. Notion drops "What's New" prompts in the relevant workspace area. Figma announces new toolbar features by tagging them in the toolbar itself. Slack uses persistent inline banners for changes important enough to warrant attention.
The mechanism here isn't psychological, it's structural. You can't open-rate your way out of an email that wasn't opened. In-product placement bypasses the entire delivery problem.
What seems to work:
What I've seen go sideways:
- Modal popups for everything. Modals are the highest-friction surface in the UI; saving them for actually critical news matters.
- Banners that link to a generic changelog post. The link should go to the feature, in the place where the user can try it.
- Sending the same announcement by email and in-product. Pick one channel based on the user's behavior in the last 30 days.
The shift in posture I keep coming back to: an email is a request to go somewhere. An in-product banner meets the user where they already are. The second one wins for adoption every time.
14. "What's New" widget
Banners interrupt. Emails get deleted. Changelogs at /changelog get ignored. The version of feature communication that actually works for engaged users is the in-product "What's New" widget. A bell icon with a notification badge, persistent, pull-based rather than push-based.
Linear set the bar for this. Their changelog is often cited as the gold standard. Entries are written user-first ("you can now do X"), each includes a screenshot or GIF, links go straight to docs, cadence is weekly to biweekly. The notification badge re-engages users every time they open the app.
Loom takes it further by attaching short Loom videos to each entry. Fifteen seconds of demo beats a paragraph of description, especially for visual features.
The mechanism is pull instead of push. Users who care will check the widget. Users who don't won't be interrupted. That distinction is the whole reason it outperforms banners and emails for non-critical updates.
What seems to work:
What I've seen go sideways:
- Posting once a quarter. The notification badge needs fresh fuel; long gaps train users to ignore it.
- Marketing-grade language. "Reimagining the future of work" reads as noise; "fixed the export crash" reads as a real product.
- Hiding the widget behind a settings menu. If users can't find it in 5 seconds, it's not pull-based, it's invisible.
The shift in posture I keep coming back to: emails go to users who might come back. A changelog widget meets users who already came back. Different jobs, different surfaces.
15. Feature suggestions based on usage
Most products surface advanced features the way Netflix surfaces movies in the 1990s. A homepage with one big banner that everyone sees regardless of what they actually watch.
Netflix solved this decades ago by recommending based on what each user actually watches. Almost no SaaS products do the same. Most show every user the same homepage tooltip, the same "premium features" tab, the same upsell modal.
The pattern that works is contextual feature suggestion. Surface the right next feature at the right moment, based on what the user just did. Grammarly does this brilliantly: "This sentence has advanced clarity issues. Upgrade to see suggestions." The pitch lands at the exact moment the user feels the limitation. Lemlist uses Jimo to fire similar prompts for users who haven't connected a custom domain yet.
The mechanism is the Mere Exposure Effect. The more times a user encounters a relevant feature, the more likely they are to try it. Relevance is the unlock; repeated exposure without relevance just becomes noise.
What seems to work:
What I've seen go sideways:
- Generic upsell modals shown to everyone regardless of usage pattern. Reads as a banner ad inside your own product.
- Suggestions that fire too early. Telling a brand-new user about an advanced feature is the wrong moment.
- Stacking three suggestions at once. The brain picks zero.
The shift in posture I keep coming back to: feature discovery is a recommendation engine problem, not a marketing problem. Most teams build the feature, then write a help article, then wonder why no one finds it. The fix lives inside the product, fired by actual usage signals.
16. Win-back and re-engagement flows
Most products treat churn as an event. The day a user cancels. The day a credit card fails. The reality is that churn is a process that starts weeks earlier, when the user quietly stops logging in.
The good products detect it early and pull users back before they make a conscious decision to leave. Spotify uses "Made For You" playlists to surface based on listening habits. Dropbox sends emails framed as "Your files miss you" with specific quantification ("847 files and 23 shared folders waiting"). Duolingo built an entire brand around streak loss, including the now-famous menacing notifications.
The mechanism is loss aversion. The same psychological force that makes people irrationally hold onto stocks pulls users back toward products they've already invested in. The more value they've built up inside the product (data, settings, integrations, history), the stronger the pull.
According to Harvard Business Review, acquiring a new customer costs between five and twenty-five times more than retaining an existing one. Most teams understand this in principle and still spend almost nothing on disengagement flows.
What seems to work:
What I've seen go sideways:
- Generic "we miss you" emails with no specific value reminder. Reads as marketing automation, not a product.
- Re-engagement copy that quantifies vanity ("You logged in 30 times!"). Time spent isn't value.
- Win-back triggered at day 60 instead of day 7. By the time most products notice, the user has already moved on.
The shift in posture I keep coming back to: retention isn't a metric to measure, it's a flow to design. Most teams measure churn; the ones that actually move the number have built a specific sequence around the signals that precede it.
17. Usage milestones and progress celebrations
The longer a user stays, the easier they are to retain, but only if they can feel the value they've accumulated. Most products quietly stack up that value behind the scenes and never show it to the user. Hours saved. Reports generated. Deals tracked. The user keeps using the product but never gets a sense of what they've built.
The fix is explicit usage milestones, surfaced at intervals. Chameleon celebrates after a user closes 50 boards. Grammarly sends weekly Writing Insights with word counts and vocabulary scores. Some of those Grammarly emails go viral because users share them; the milestone became content.
The mechanism is sunk cost, used positively. The more value a user has visibly created inside the product, the harder it gets to walk away. They're not just leaving a tool, they're leaving 47 hours of work behind.
There's a useful side effect inside companies. When the user becomes the champion for the tool internally, milestone summaries become their evidence. "Our team has created 500 dashboards and saved 120 hours this quarter" is a defensible budget line.
What seems to work:
What I've seen go sideways:
- Vanity milestones. "You've used us 30 days in a row" is a streak, not value.
- Daily celebrations. Reward fatigue sets in fast; weekly or milestone-based is the right cadence.
- Milestones with no path to expansion attached. Either the user feels great and that's the end, or they feel great and you give them something to do with the feeling.
The shift in posture I keep coming back to: every product has value the user is creating; most products forget to tell the user about it. The ones that do, retain.
That's 17 patterns. Most products won't need all of them.
The ones that actually move activation pick two or three and ship them well.
What ties them together isn't the format. It's the surface. An onboarding tour, a help doc, an empty-state CTA, a behavior-triggered nudge. When those all live in different places, the user has to do the work of figuring out which one is for them. When they live in one agent, the product does.
If you'd rather not guess which two or three to start with, the Activation Audit is a 5-minute diagnostic that points you at the right ones.
Activation Audit
Audit your products onboarding in 5-minutes using Claude
This is the exact framework I use when I run activation audits for clients, packaged so you can run it yourself with Claude doing the heavy lifting.
If your users aren't activating the way you expected, let's find out why.
I offer a free 30-minute Discovery call where I look at your onboarding with you and tell you honestly what's broken and whether I'm the right fit to fix it.
Activation Guide
Fix your activation in 3 steps
A hands-on workbook for defining your Same-Day Win, clearing the path, and proving it worked.