Users who skipped an app's tutorial completed 94% of their tasks, users who read it completed 91%, and the readers rated the same tasks as harder, in Nielsen Norman Group's 70-person study published in 2020. That result is the right place to start a list of user onboarding best practices, because most onboarding advice still assumes the job is to explain the product. The data says the job is to get a new user to the one action that predicts they will stay, and B2B SaaS products manage that for 37.5% of sign-ups on average (Userpilot, 62 companies, 2025).
This guide turns the published evidence into 14 rules, grouped by where in the first week they apply, with product examples and the four events to measure. Vendor figures are labelled. Our SaaS onboarding statistics post holds the benchmark tables, and our SaaS onboarding flow design post, published the day before this one, walks through the screens; this post is about the rules behind both.
What user onboarding best practices are for
Every rule below serves one metric. Rachitsky and Timen define the activation milestone as the earliest point in onboarding that, by showing the product's value, predicts long-term retention; users who reach it should retain at least twice as well as users who do not. Their survey of more than 500 products puts the SaaS median at 30% and the average at 36%, and Userpilot's 2025 benchmark of 62 B2B companies averages 37.5%. ProductLed's survey of more than 600 SaaS businesses found only 34% track activation, so most teams cannot say where they stand.
| Rule | Practice | Evidence |
|---|---|---|
| 1 | Define one activation event before designing a screen | Rachitsky and Timen, 2x retention test |
| 2 | Instrument it, and the time to reach it, in the first release | ProductLed: 34% track activation |
| 3 | Design the flow backwards from that event | Userpilot: 37.5% average activation |
| 4 | Branch on intent, with two or three questions at most | Chameleon: 37.5% modal dismiss rate |
| 5 | Let people in before the account wall | NN/g login walls; Baymard 42% |
| 6 | Postpone setup; ship defaults and sample data | Apple HIG; NN/g empty states |
| 7 | One question per screen, with the reason it is asked | GOV.UK Design System |
| 8 | Decide the card wall on sales motion, not on taste | ChartMogul: 30% against 8% |
| 9 | Teach by doing, not by telling | NN/g 2020: 94% against 91% |
| 10 | Pull, not push: contextual help the user triggers | Chameleon: 2x engagement when self-triggered |
| 11 | Progressive disclosure, two levels deep at most | NN/g, 46 web applications |
| 12 | Text first, short, with a progress indicator | Chameleon: 44% text-only, 12% lift |
| 13 | A checklist of three to five items, one pre-ticked | Userpilot: 19.2% completion; Chameleon: 5 items |
| 14 | Get a second person in before day 14 | ChartMogul: trials decide by day 7 |
Sources: named in each section below. Chameleon, Userpilot and ChartMogul are vendor platform data.
Rules 1 to 4: decide what onboarding is for
Rule 1: define one activation event. The event should predict retention or revenue with at least a twofold lift, be something an individual user can do rather than a workspace outcome, and sit neither too early (finishing sign-up) nor too late (several purchases). Pick an early milestone that correlates with retention, then confirm it with experiments.
Rule 2: instrument it first. A team that cannot see its activation rate judges a redesign by opinion. Userpilot's 2025 report puts average time to value at 1 day, 12 hours and 23 minutes across its 62 companies, and that is the second number to track from day one: the clock between account creation and the activation event.
Rule 3: design backwards. Every earlier screen is then judged by whether it moves the user towards the event. The 30% to 37% band above is the honest default for a B2B product that has not measured its own.
Rule 4: branch on intent, briefly. Chameleon's 2025 benchmark, built on more than 550 million interactions on its platform, puts the average modal dismiss rate at 37.5%, with 38% of dismissals in under four seconds. A welcome survey is a modal, and it earns its place only when the answer changes the next screen. Two or three questions, each with a branch behind it, is the limit.
Rules 5 to 8: get people into the product before asking for anything
Rule 5: let people in before the account wall. Nielsen Norman Group has recommended against login walls since 1999: creating an account carries a high interaction cost, and users register far more willingly after they have received something. That is a 2014 article built on qualitative testing, a position rather than a measurement. Baymard's 2023 checkout research is the retail proxy: 42% of sites in its benchmark interrupt checkout to suggest an account, and in its testing the invitation performed best at the confirmation step, when it only needs a password. Ask once the user has something to protect.
Rule 6: postpone setup and ship defaults. Apple's Human Interface Guidelines, updated June 2024, say it plainly: postpone non-essential setup or customisation steps and provide reasonable defaults so most people can start immediately. For a web application the equivalent is sample data. Nielsen Norman Group's 2021 guidelines ask every empty state to communicate system status, explain what the area is for and offer a direct path to filling it; its Loggly example offers two routes, add your own log sources or explore with demo data.
Rule 7: one question per screen, with the reason. The GOV.UK Design System's question-page pattern starts from one question per page so users can focus on it, lets research decide when questions can be grouped, and asks designers to know why every question is asked and to make that reason clear on the page.
Rule 8: decide the card wall on your sales motion. ChartMogul's January 2026 survey of 200 software products found 20% of free trials require a card up front. Card-required trials convert at 25% to 35% free-to-paid for a median product against 4% to 6% without a card, but from fewer sign-ups: per 1,000 visitors the report models 35 sign-ups and 10.5 customers with a card against 45 and 3.6 without. These are self-reported estimates. If a salesperson follows up each trial, the card is a filter you want; if the product grows on usage, it removes the people who would have activated on day three.
Rules 9 to 12: user onboarding UI design that teaches by doing
Rule 9: teach by doing. Nielsen Norman Group's study gave 70 iOS users four apps with card-deck tutorials. The 35 who read the tutorial succeeded on 91% of tasks and the 35 who skipped it on 94%, not a significant difference; the readers rated the tasks harder, 4.92 against 5.49 on a seven-point scale, and that difference was significant. Apple's guidance agrees: people retain more when they perform the task than when they view instructions.
Rule 10: pull, not push. Nielsen Norman Group's 2023 comparison of tutorials and contextual help calls launch-time walkthroughs, feature card decks and "what's new" modals push revelations: they interrupt, get skipped, do not reliably improve performance and are quickly forgotten. Pull revelations, help triggered when a signal shows the user needs it, avoid all four problems if they are unobtrusive and easy to find again. Chameleon's data agrees, from a vendor that sells tours: launcher-driven tours complete at 67% against 31%, and a modal opened by a user action completes at 47% to 51% against 39% on page load.
Rule 11: progressive disclosure, two levels deep. Jakob Nielsen's 2006 article, based on 46 web applications, defines progressive disclosure as showing the few options most users need first and placing the rest behind a clearly labelled secondary screen. It improves learnability, efficiency and error rate as long as the depth stays at two levels: the first screen shows the activation action and little else, with "advanced" one click away and labelled.
Rule 12: text first, short, with progress shown. Chameleon's 2025 data gives the UI numbers: text-only modals complete at 44%, text with an image at 29%, text with video at 21%; a progress indicator lifts tour completion by 12%; tours beyond five steps lose more than half their users; embedded elements are up to 1.5 times more likely to be acted on than pop-ups, with 26 words as the recommended ceiling. Nielsen Norman Group's 2023 guidelines add: make help easy to dismiss and easy to find again, show it alongside each step rather than up front, and skip guidance for conventional patterns such as a gear icon.
| Pattern | Who triggers it | When it fits | Evidence |
|---|---|---|---|
| Launch-time tour | Product, on first login | Novel interaction models only | 94% against 91% task success without it (NN/g 2020) |
| Welcome modal | Product, on page load | One sentence and one action | 39% completion on load, 47% to 51% on user action (Chameleon 2025) |
| Hover tooltip | User, on hover or keyboard focus | Unfamiliar field or unlabelled icon | A pull revelation; keep it findable later (NN/g 2023) |
| Coach mark or inline tip | Product, at the moment of need | One hint, one action, shown once | Pull help avoids skip and forget (NN/g 2023) |
| Checklist launcher | User, when ready | Tasks that span sessions | 67% tour completion from a launcher against 31% (Chameleon 2025) |
| Empty state with sample data | Product, in the blank area | Any list or canvas that starts empty | Status, learning cue, direct path (NN/g 2021) |
Sources: Nielsen Norman Group 2020, 2021, 2023; Chameleon Benchmark Report 2025 (vendor platform data)
Rules 13 and 14: keep the second session alive
Rule 13: a short checklist with one item ticked. Userpilot's 2025 benchmark across 188 companies puts checklist completion at 19.2% on average and 10.1% at the median. Chameleon reports that engaged users complete about five items in a session. A checklist still earns its place because it survives the session boundary that kills a tour, but the data argues for three to five items led by the activation action, with the first already complete. Userpilot's examples post, a vendor article, shows Productboard doing exactly that: the workspace is created automatically and shown as the first task done.
Rule 14: a second person before day 14. ChartMogul's analysis of about 2,500 SaaS companies, covered in our statistics post, found trial conversions peak around day 7, and its 2026 survey found 62% of products run 14-day trials. The invite belongs inside that window, after the activation event and not before it. Our SaaS onboarding email sequence post, published the day before this one, carries this rule outside the app.
User onboarding examples that follow the rules
The examples come from pages we opened: Growth.Design's walkthrough of Trello, Nielsen Norman Group's empty-state review, and Userpilot's 2026 examples article, a vendor post. None publishes conversion data, so read the table as illustration, not proof.
| Product | What it does in onboarding | Rule | Observed in |
|---|---|---|---|
| Trello | Lets users explore before confirming email; a use-case question personalises the first board | 5, 4 | Growth.Design walkthrough |
| Loggly | Empty state offers two paths: add log sources or explore with demo data | 6 | Nielsen Norman Group, 2021 |
| Datadog | Empty favourites area reads "Star your favorites to list them here" | 6 | Nielsen Norman Group, 2021 |
| Calendly | Role-based setup questions create a usable meeting link early; optional learning is separated from required setup | 3, 7 | Userpilot examples (vendor) |
| Productboard | First checklist task pre-completed because the workspace is created automatically | 13 | Userpilot examples (vendor) |
| Linear | Onboarding tasks arrive as native issues in the user's own work queue | 9, 10 | Userpilot examples (vendor) |
| Figma | Welcome survey branches to tailor the workspace and templates to the stated goal | 4 | Userpilot examples (vendor) |
| Miro | One work-email field with SSO options beside it | 5, 7 | Userpilot examples (vendor) |
Sources: Growth.Design, Trello user onboarding case study; Nielsen Norman Group, Designing Empty States in Complex Applications, 2021; Userpilot, Best User Onboarding Experiences in 2026
One pattern recurs. The products praised for onboarding put the user's first real object (a board, a meeting link, an issue) on screen within the first minute, and ask questions only where an answer visibly changes that object. Trello's 15-option persona question is the counter-example: it asks more than the branching behind it can use.
How to measure whether the rules are working
Four events cover the 14 rules. Wire them before the first screen is designed.
| Event | What it tells you | Benchmark to read against |
|---|---|---|
| Account created, with entry model | The top of the activation funnel | 4.5% website-to-sign-up for a no-card trial (ChartMogul 2026, survey estimate) |
| Activation event completed, with time since creation | Activation rate and time to value | 30% to 37.5% activation; about a day and a half to value (surveys and Userpilot) |
| Checklist item completed, with item name | Which tasks users actually do | 19.2% average completion, 5 items per engaged session (Userpilot, Chameleon) |
| Tour, modal or tip dismissed, with step number | Where the UI is pushing too hard | 37.5% modal dismiss; completion falls past 5 steps (Chameleon 2025) |
Sources: ChartMogul SaaS Conversion Report 2026; Userpilot Benchmark Report 2025; Chameleon Benchmark Report 2025; Rachitsky and Timen 2022
Onboarding lives inside the application, and we build applications in Next.js; the auth and billing plumbing behind rules 5 and 8 is covered in our Next.js and Supabase SaaS post. On an onboarding project we would write these four events into the specification before any screen is designed, as our MVP spec template asks for the success metric before the feature list.
What this means for a founder scoping an onboarding flow
Treat the 14 rules as a sequence, not a menu. Choose the activation event and prove it clears the twofold retention test. Instrument it. Then cut everything in front of it that neither branches the flow nor is needed for the product to run: the email wall moves to the first invite, setup becomes sample data, the tour becomes two or three tips the user triggers, and the welcome survey shrinks to the questions with a branch behind them. Keep the checklist to five items with one ticked, and get a colleague in before the trial's second week. Read the result against 30% to 37.5% activation and a day and a half to value; every benchmark here is vendor data or a survey, not your users.
We would scope onboarding as part of the application build, with the activation event and the four measurement events in the first release. Our web app development service page explains how we scope it.


