Work · 2026

One flow, three clients

A config-driven end-of-tuition wizard whose steps, copy and schema come from the backend, so web, iOS and Android render the same flow.

Work case study · 2 min read · event #000032

  • PHP 8.3
  • Laravel
  • Vue 3
  • TypeScript
  • Inertia.js
  • iOS
  • Android
  • MySQL

The problem

When a student wanted to stop tuition, the product mostly let them cancel. We learned little about why, and we gave no help to the ones who might have stayed with a different tutor or fewer lessons. Fixing that meant a multi-step flow on web, iOS and Android, and the risk with three clients is that they drift apart. Every copy tweak or new question would need three releases.

What I built

I made the flow server-driven. The backend returns the wizard configuration: the steps, their copy, the questions and a schema version. Each client is a renderer for that configuration. If the backend adds a step or changes some wording, all three clients show it without a release, and the schema version lets an older client recognise something it cannot render and fall back safely.

Behind the flow:

  • Outcome submissions are stored with a relationship status history, so we can see what happened to each tutoring relationship over time.
  • Each outcome (end tuition, switch tutor, fewer lessons, date-ranged cancellation of bookings) is processed by a queued job, so the side effects run reliably and can be retried.
  • On the web client, steps are composed in Vue 3 and TypeScript, with a session store holding the user’s answers as they move through.

The retention paths are where the product value is. A user who wants to leave can be shown an alternative-tutor carousel, with affordability logic so we only suggest options they could realistically book. Late-cancellation fee warnings are shown before they commit. Whatever they choose, the reason is stored as structured data rather than free text.

Along the way I refactored the booking repository into intent-named methods, which made the outcome jobs far easier to read and test.

What it gave us

Cancellation became a retention flow that records structured churn reasons, and the three clients stay consistent by construction rather than by discipline.

What I’d do next

Server-driven flows can become a hidden programming language if they are not constrained. I would add a published, validated schema for step types, with contract tests that each client must pass, and an experiment hook so alternative flows can be compared directly.

← Back to work