Onboarding fails when a user returns after an interruption and the product cannot tell whether to resume, repeat, or start again. A progress bar is useful only if its steps correspond to durable, verifiable state.
Model onboarding as a workflow with recoverable transitions. Distinguish information entered, external work requested, work confirmed, and setup complete. That makes both the interface and support diagnosis clearer.
Store outcomes rather than screen visits
Define the minimum durable facts for each step: account created, workspace established, required settings saved, integration authorised, or first useful action completed. Visiting a settings page does not prove the settings are valid.
Give repeated actions an explicit policy. Reopening onboarding should not create a second workspace or duplicate an external subscription merely because the prior response was interrupted.
Recover a fictional study-product setup
Imagine a founder signs in, names a workspace, creates a first study, and connects a paid-access flow. They close the browser after the study is saved but before the interface displays confirmation.
On return, the application should inspect durable state and offer the existing study rather than creating another. If the payment or integration step remains pending, show that state with a concrete next action. Do not mark the whole account complete because the user once reached the final screen.
Test interruption before and after each write, repeated clicks, a second browser, expired session, and a changed workspace role. Verify the final records and any external references. These are proposed checks, not a claim that a specific product has passed them.
Separate recoverable gaps from manual-review cases
Some failures can retry automatically, while others need user input or staff review. An invalid business setting should not loop forever in a background retry. A temporary provider outage should not force the founder to re-enter information that was already saved.
Provide support with a workflow identifier, current state, last attempted transition, and failure category. Keep private form contents out of general logs. If support changes state, record the real actor and reason.
The tradeoff is convenience versus state-model complexity. A very small pilot can have fewer onboarding steps, which reduces failure combinations. It still benefits from a clear definition of complete and safe repeated actions. Avoid building an elaborate workflow engine when a small explicit state model is sufficient.
The StackFlight case study provides context for connected study creation and paid access. The interrupted setup here is fictional and does not describe a reported incident.
Questions about onboarding recovery
Should users always restart after a failure?
No. Reuse verified saved state where appropriate and identify only the incomplete step. Restarting can create duplicates and unnecessary frustration.
Can the browser own onboarding progress?
It can display progress, but durable server and external-system outcomes should determine completion for consequential steps.
What is the first recovery test?
Interrupt confirmation after a successful write, then return from another session. Verify that the existing result is recovered once. Bring failures to an onboarding review.