User onboarding best practices are often reduced to a single instruction: remove steps until the flow is as short as possible. The instruction is easy to follow and easy to measure, which explains its popularity, but it leaves out a question that matters just as much, namely which of the steps are doing necessary work. A permission prompt, a security setting or an identity check adds friction in the first minute and also protects the user or the business later on. A welcome carousel with five screens adds friction as well, and nothing in the product depends on it.
This article treats onboarding as a sorting problem. It starts with the difference between friction that earns its place and friction that does not, then covers how to find the first value moment, how to classify each step as keep, defer or remove, which techniques reduce onboarding friction in practice, and where the evidence supports a longer flow. It follows the previous article in this series on how an app store product page influences downloads, which ended at the moment a person taps the download button. From that point the onboarding flow decides whether the person reaches the value the page promised.
User onboarding best practices begin by separating good friction from bad friction
Friction is any effort a person spends before the product delivers something useful. Typing an email address, granting camera access, choosing a plan and reading a tutorial all count. Product teams often treat the total number of steps as the figure to minimize, which works as a rough guide and fails in the cases where a step exists for a reason.
The Nielsen Norman Group draws that line in its guidance on mobile app onboarding. It recommends avoiding dedicated onboarding wherever possible and investing in a more usable interface instead. It also lists the conditions under which onboarding is warranted: when users need information to get started, with identity verification in a banking app as its example, when functionality depends on the user’s context, and when features are unusual enough that users would not recognize them. Those three conditions work as a practical test for any step in a flow, and a step that fits none of them is a candidate for removal.
Apple’s Human Interface Guidelines reach a similar position from the platform side. They state that people ideally understand an app by experiencing it, and that when onboarding is necessary it should be fast, fun and optional.
For the rest of this article, good friction means a step that the product cannot safely skip or that visibly improves the first result, such as an identity check in a regulated product or a goal question that changes what the app shows first. Bad friction means a step that asks for effort and returns nothing in the first session, such as a tour of features the person has no reason to use yet. The work then comes down to removing the second kind and keeping the first kind as light as the requirements allow.
Define the first value moment before designing the onboarding flow
Time to value is the time between a person’s first interaction with a product and the first moment the product delivers its core benefit. The moment itself differs by product. For a photo-cleaning app it is the first cleaned library, for a budgeting app it is the first categorized transaction, for a team tool it is the first task completed by another person, and for an insurance product it is a first quote the customer understands. Until a team names that moment for its own product, discussions about the onboarding flow tend to revolve around opinions about individual screens.
A workable definition has three properties. The moment can be observed as an event in analytics, which makes it possible to count how many new users reach it. It involves the product doing something useful for the person and not only the person completing setup. It also sits close enough to the start that a meaningful share of new users can reach it in the first session. Teams often find the right event by comparing users who stayed for several weeks with users who left in the first days, and then looking at which early action separates the two groups.
With the first value moment defined, every step between the download and that moment can be counted and assessed. The count matters because the product page has already made a promise, and the onboarding flow is where that promise is kept or broken. A page whose first screenshot shows a finished result sets an expectation that the first session will produce something similar, so the page and the flow are best planned together.
Sorting every step into keep, defer or remove
Once the first value moment is fixed, each step in the flow can be placed in one of three groups. A sorting pass like this is one of the more dependable user onboarding best practices because it replaces opinion with a repeatable test. The table shows the test question for each group and typical examples.
| Decision | Test question | Typical examples |
| Keep | Would the product be unsafe, unlawful or unable to work without this step? | Identity verification in regulated products, required consent, payment authorization, access the core feature depends on |
| Defer | Is the step useful later but unnecessary for reaching the first value moment? | Profile photo, notification preferences, teammate invitations, optional profile fields, plan upgrade offers |
| Remove | Does the step serve the team’s convenience more than the user’s goal? | Tours of rarely used features, duplicate confirmations, long marketing surveys, cosmetic customization at first launch |
The keep group tends to be smaller than teams expect, and the defer group tends to be larger. NN/g’s guidance on customization illustrates the difference. It places visual customization such as color schemes in settings and not in onboarding, while content customization can fit onboarding when the prompts are brief and explain why the data is needed. Apple’s guidelines say to postpone nonessential setup and customization steps and to provide reasonable default settings, so that most people can start interacting with the product immediately.
Permission requests belong in the keep group only some of the time. Apple’s guidelines suggest placing a request in onboarding when the app cannot function without the access, because that gives the product a chance to explain the benefit, and otherwise presenting it when the person first uses the feature that needs it. A camera permission for a scanning app fits the first scan, while a notification permission for a product whose value sits mostly inside the app can wait until the person has seen a reason to receive alerts.
Deferring a step has a cost that deserves a record. A step moved out of the first session still has to happen somewhere, and without a later prompt the product accumulates users who never completed it. Each deferred step therefore needs a named later moment, such as the third session or the first time the related feature is opened.
Techniques that reduce onboarding friction without removing what matters
Progressive onboarding spreads the learning across the life of the product and does not place all of it before first use. It rests on progressive disclosure, which NN/g defines as showing users a few of the most important options first and revealing a larger set of specialized options when they ask for them.
NN/g reports that the technique improves learnability, efficiency of use and error rate, and it recommends two levels, because designs with more levels tend to leave users lost between them. It names conditions for success as well: the split between initial and secondary features has to match actual usage, the path to the secondary level has to be obvious, and the selection should rest on evidence such as analytics and usability testing.
In onboarding terms, progressive onboarding means the first session covers what the first value moment requires and leaves the rest to contextual prompts. Apple’s guidelines support the same pattern by recommending a collection of context-specific tips in place of a single flow, with instructions placed near the part of the interface they describe. A tip that appears when a person first opens a feature reaches them when it is relevant, which a welcome carousel cannot do.
Several other techniques shorten the path to first value. Sample data lets a person see a populated version of the product before entering anything of their own: a project tool that opens with a clearly labeled sample project allows someone to explore the main views within the first minute, and the sample can be deleted afterwards. Smart defaults serve the same purpose for settings, since a reasonable preset for currency, language or notification frequency removes several decisions from the first session. Interactive walkthroughs suit complex workflows, since Apple’s guidelines advise teaching through interactivity. Deck-of-cards tutorials are the format NN/g does not recommend.
Deferred registration lets a person use the product before creating an account, and one widely cited experiment shows what that can change.
What Duolingo’s delayed sign-up test shows about onboarding flow design
Duolingo offers one of the better documented examples of how the order of steps in an onboarding flow changes results. Gina Gotthilf, then the company’s VP of Growth, described the test in an interview with First Round Review.
Moving the sign-up screen back a few steps, so that people experienced part of the product before being asked for an account, produced about a 20 percent increase in daily active users. She noted that this first test predated her time at the company. About three years later, when the user base was much larger, further changes to the sign-up wall produced an 8.2 percent increase in daily active users. Those changes included soft walls that users can skip and a hard wall that requires sign-up to continue.
Two cautions apply when reading these figures. They describe one company’s language-learning product, where the first lesson delivers value within minutes, and the interview does not state the measurement windows or sample sizes. The second round also produced a smaller gain than the first on a much larger base, which points to diminishing returns once the main change has been made. A 2026 review of onboarding flows by Appcues describes the current Duolingo flow as running the first lesson before asking for an account, although it does not state when or how the earlier figures were measured.
The reasonable reading is that registration is a step many products can defer without losing the data they need, provided the product can deliver something before it knows who the user is. Products whose first value depends on the user’s identity or data, such as a bank account, cannot always use the same sequence.
When user onboarding best practices say to keep the flow long
Some flows should stay long, and the clearest cases involve regulated products. NN/g lists identity verification in a banking app among the situations that justify onboarding. In financial services and insurance, identity checks and consent collection often come from regulation, so the question becomes how to complete them with the least abandonment and not whether to include them.
The scale of that abandonment has been measured. Signicat’s 2022 report The Battle to Onboard surveyed 7,600 consumers in 14 European countries and found that 68 percent had abandoned a financial services application in the past year, up from 63 percent in its 2020 survey. Of the respondents, 21 percent said they abandoned because they had to provide too much personal information, and 92 percent said they were concerned about how much data they share with their bank. The report is a survey of stated behavior across European markets and covers financial applications in general, so it indicates the scale of the problem and does not isolate a single step such as identity verification.
The findings point to where effort can go. Because concern about data sharing is widespread, explaining why each item is requested addresses a stated reason for abandonment, and NN/g makes the same recommendation for customization prompts. Asking only for what regulation requires at account creation, and collecting optional details after approval, reduces the personal data requested before the person has seen any result.
Letting an applicant save progress and return later suits flows that depend on documents the person may not have at hand. In some products, regulation permits showing a quote, a demo account or a read-only view before verification, which moves the first value moment ahead of the heaviest step. What regulation permits differs by market and by product, so any change of sequence needs review with compliance specialists first.
Measuring onboarding success with activation rate, Time to value and drop-off
An onboarding change counts as an improvement when it moves a metric tied to the first value moment. The table lists four measures that together describe a flow.
| Metric | What it measures | How to read it |
| Activation rate | Share of new users who reach the first value moment within a set window, such as seven days | The main outcome metric. A rise after a change indicates that more people reach value |
| Time to value | Elapsed time from first open to the first value moment, usually reported as a median | A falling median indicates a faster path. Read it alongside activation rate, since a shorter time among fewer people can mislead |
| Step drop-off | Share of users who leave at each step of the flow | Shows where the flow loses people and which steps are candidates for deferral |
| Early retention | Share of activated users who return on day 1 and day 7 | Indicates whether the value delivered was enough to bring people back |
Interpretation needs some care. Drop-off at a step is a problem only when the step is not required, and heavy drop-off at identity verification may reflect the regulation more than the design. Changes are easier to attribute when they are tested one at a time against a control group. Numbers and observation also answer different questions, since analytics show where users leave and a moderated session with a real user shows the reason. A routine that pairs a weekly review of the funnel with a small number of moderated sessions on the steps losing the most people covers both questions at modest cost.
Putting user onboarding best practices into practice with Varmeta
Onboarding decisions depend on the product, its users and its regulatory context, which is why the same flow rarely transfers cleanly from one product to another. Varmeta has built many kinds of digital products for clients across industries, from entertainment and consumer apps to business tools, and its portfolio shows the range. That experience covers how first-session flows, core features and release planning fit together, and Varmeta also works with AI in the products it develops. Teams planning a new product, or reworking a flow that loses too many people in the first session, can discuss scope and priorities with Varmeta through its contact page.
Conclusion
User onboarding best practices start from a question about each step and not from a target length. A step belongs in the first session when the product cannot work safely without it or when it visibly improves the first result. Other steps can move to a later moment through progressive onboarding, be replaced by sample data or smart defaults, or be dropped. The first value moment gives the sorting its reference point, and activation rate, time to value and step drop-off show whether the changes worked.
The evidence also shows limits. Duolingo’s delayed sign-up produced measurable gains in a product that delivers value quickly, while the Signicat survey shows that abandonment in financial services is high even where some steps are unavoidable. In both situations the practical work consists of reducing onboarding friction where the product allows it and making the remaining steps clear, short and explained.