Signup and Onboarding Rebuilt Around Day-28 Activation

Signup rate is the wrong number to raise. Removing three fields will raise it and can lower the count of accounts still active on day twenty-eight, which is the number your board is actually watching. We agree an activation definition in the first week and freeze it for the engagement, first transaction signed for a wallet product, second session with data imported for a business tool, and report signup and activation side by side in every update. When the two move in opposite directions, that is the finding, not an inconvenience. Identity verification, where the loss usually is: For crypto and fintech onboarding the largest single drop is verification, and it is measurable in detail that almost nobody looks at. Sumsub, Onfido and Veriff all expose rejection reason, document type, country and retry rate. A 34 percent first-attempt rejection rate on one document type in one market is a specific, fixable thing, and until it is split out it hides inside a general complaint about onboarding drop-off. Retry behaviour matters as much as first-attempt pass rate: a flow that lets a rejected user try again with better guidance recovers a large share of them, and most flows do not. Wallet flows: The connect step across MetaMask, WalletConnect and Coinbase Wallet, network switching, the seed phrase screen where honest products lose people for good reasons, and the first signature where gas is the anxiety rather than the amount. We separate the user who abandoned from the user whose transaction failed, which requires console and RPC data alongside behaviour. Testnet rehearsals catch none of this. We work against mainnet sessions. Software onboarding: The email verification round trip and where it strands mobile users, SSO discovery for buyers whose IT will not allow anything else, teammate invitation placed before the product is worth showing rather than after, and the empty state that asks for data the new user does not have to hand in their first sitting. How experiments run here: Pre-registered. Hypothesis, primary metric, sample size, stop date, filed before the first user is bucketed, in GrowthBook, Optimizely, Statsig or your own assignment service. We do not change the activation definition mid-flight, and we do not stop a test early because it looks good. Losing tests are reported in the same format as winning ones, and the ratio between them is not a performance indicator we will accept being measured on. Not included: Lifecycle email and push messaging, which sit next to this work and belong to someone else; we will name a practice we trust. No referral loops or invitation incentives. No app store optimisation and no native app onboarding, which needs a different instrumentation stack than we run. Who should not buy this: Products before product-market fit, where onboarding is not the constraint and improving it hides the real problem for another quarter. Anything under roughly 300 signups a week, where a continuous test programme cannot honestly be run and the retainer would be theatre. Teams unwilling to fix the activation definition, because a metric that moves when the result is unwelcome is not a metric. How the retainer runs: One definition, one backlog, one test in flight at a time on the primary flow, a written readout every fortnight and a quarterly session where we argue about what to stop doing.