The anti-patterns that quietly kill role-based flows

Role-based first runs fail in predictable ways, and naming the failure modes up front is cheaper than discovering them in your activation numbers three months later. The first and most common is over-segmentation. Three to five roles you can route cleanly beats twelve micro-personas you cannot maintain, because every extra branch is a screen to build, instrument, and keep honest, and a branch nobody ships a real path for is worse than no branch at all. Start with the smallest set of roles whose first sessions genuinely diverge, and split a role only when the data shows two jobs hiding inside one option.

The second is the fake-route, where the survey branches the welcome screen but every role lands in the same product two clicks later. Users feel this instantly. If tapping "analyst" and tapping "admin" lead to the same empty canvas, you have built personalisation theatre, and theatre erodes trust faster than no personalisation at all because it makes a promise the product breaks. The route has to change the actual work, not just the wallpaper. The third failure is letting the role choice harden into a cage: a user taps the wrong option in the first ten seconds and the product then refuses to show them anything else. Always leave a visible "actually, I want to do something else" escape, and let AI override a stale survey answer the moment behaviour contradicts it.

The fourth, and the most expensive, is shipping the branches without the measurement. A role-based flow you cannot read per role is just a more complicated generic flow, because the entire payoff is the ability to find the one lagging role and fix its path alone. If you build only one thing from this playbook, build the per-role instrumentation, because every other improvement depends on being able to see which door is jammed. This is the same root-cause discipline that the broader growth-bottleneck work applies to the whole funnel: find the one constraint, fix it, re-measure, repeat.