The Six Jobs of Mobile App Onboarding: Attribution, Personalization, Paywall, Permission, Social Proof, Commitment

The checklist

• Attribution: "how did you hear about us". The only place word of mouth becomes measurable
• Personalization: questions that shape the plan and make the paywall feel personal
• Paywall: one job of six. Its conversion is set by the other five
• Permission: the notification ask, embedded in a question the user answers for their own reasons
• Social proof: ratings, press, expert endorsement. Borrowed conviction at the moment of doubt
• Commitment: micro-yeses, holds, signatures. Make the user's intent physical before the ask

Attribution: the question ad dashboards can't answer

Paid attribution tooling (MMP, SKAN) can't see word of mouth, friend referrals, or social discovery. The user can. One low-stakes multiple-choice question produces the organic acquisition picture no ad platform reports. BitePal asks on screen 2 at 3.4M downloads. Yazio asks mid-flow across 700K monthly installs and feeds the answer into live ad-budget allocation.

Placement is a live tension, not a solved question. Bend asks at the end of its 22-screen personalization, betting that an invested user recalls their channel more accurately. The early ask instead captures everyone who abandons before the finish. Optimizing for coverage? Ask early. Optimizing for recall accuracy? Ask late.

Personalization: questions that shape the plan

YarnPal stacks 7 yes/no questions across its first 8 screens, each answerable in under a second, at 3.1M downloads and 360K installs a month. The questions do two jobs at once. They shape the plan, and they build the consistency bias that makes the paywall feel like the next yes (see commitment, below).

Catzy compresses the same investment into a single move: the user names their companion cat. Ownership faster than twenty screens of buildup.

Paywall: one of the six, not the finish line

The paywall is a job in the flow, not a gate after it. Its conversion is largely decided before it renders, by the five other jobs. Duolingo delays it past the first lesson at the end of a 38-screen flow and converts roughly 10% of ~60M monthly active users to paid.

The timing budget is real though. 90% of the subscription conversions an install will ever produce happen in the first 24 hours (the 24-hour window). The paywall belongs inside the first session.

Permission: embed the ask inside a question

A cold "Can we send you reminders?" asks the user to grant the app something. Bend embeds the ask in a habit question, "When do you usually exercise?", and the answer sets the reminder time. The native prompt then only confirms a choice the user already made. One answer implicitly answers two questions.

The same move recurs across the library at lower resolution. Airlearn reframes it as an identity declaration ("Are you committed to learning [language]?"). Flo asks "do you want a warning if something looks unusual?". YarnPal asks "how many reminders do you want?". Each wraps the permission in a question the user answers for their own reasons.

Social proof: borrowed conviction

Ratings walls, press mentions, expert endorsements, user counts: conviction borrowed from other people, spent at the user's moment of doubt. Bend carries New York Times and Wirecutter coverage and a 4.7-star rating into its flow.

The failure mode is harvesting proof before the product has earned it. Bend also asks for an App Store review before the first stretch, before any product experience. The teardown calls it the single most expensive misplacement in the flow. Display proof early. Request it only after value.

Commitment: make the yes physical

Oniri places a sustained press-and-hold at the exact moment personalization investment peaks, right after the third and final wave of questions. Prayer Lock uses a drawn finger signature, then stacks three permission requests immediately after it, betting on momentum from the yes state. Catzy substitutes a written agreement: "I commit to caring for myself every day."

It is the same consistency-bias mechanism as YarnPal's stacked micro-yeses, spent in one higher-cost, higher-conviction moment instead of several cheap ones.

How to run the checklist on your flow

Label every screen in your onboarding with the job it does. A screen doing none of the six is a cut candidate. A job missing entirely is a known mechanism left on the table.

Then check placement against the tensions that are genuinely unresolved: attribution early vs late, paywall timing. There the right answer depends on your flow's length and your product's value speed, not on a universal rule. That honesty is the point of the checklist. It tells you what must be covered, and it tells you where reasonable flows differ.