Dynamic Onboarding

Definition

Dynamic onboarding is an onboarding flow where the LLM controls the order of questions and generates personalized follow-ups based on prior answers, instead of running a fixed wizard. The Wire RN pattern is two-phase: Phase 1 asks a small set of fixed base questions that every user sees, so the analytics baseline stays stable; Phase 2 hands control to the LLM, which produces AI-tailored follow-ups using a registered component vocabulary and the answers from Phase 1.

Example

A sleep-tracking app asks every user the same two base questions: primary goal and age. Phase 2 takes over. If the user picked sleep, the LLM picks a TextInputCard asking about typical bedtime. If the user picked focus, it picks a SelectionCard about distractions. Same registry, different paths, no hand-coded if/else.

// Two-phase dynamic onboarding with Wire RN
// Phase 1 = fixed base questions, analytics stay clean.
// Phase 2 = LLM picks the next component from the registry.
import { useWireAIThread } from "wireai-rn"; // wireai-rn@0.1.3

const BASE_QUESTIONS = [
  { id: "goal", component: "ChipSelectCard", props: { options: ["sleep", "focus", "stress"] } },
  { id: "age", component: "NumberStepperCard", props: { min: 18, max: 90 } },
];

function Onboarding() {
  const { messages, send } = useWireAIThread({
    systemPrompt:
      "You run Phase 2 of onboarding. Use prior answers to pick the next " +
      "registered component. One question at a time. Stop when you have enough.",
  });

  // After BASE_QUESTIONS finish, kick off Phase 2.
  return /* render BASE_QUESTIONS, then ComponentRenderer over messages */ null;
}

The split matters. Phase 1 gives the product team comparable funnel data. Phase 2 gives the user a flow that respects what they already said.

When to use it

• Apps where users come in with different goals — health, productivity, finance — and a single linear wizard wastes the majority of users.
• Products with a long onboarding (six or more screens) where personalization measurably lifts activation.
• Teams that already use generative UI for mobile and have a vetted component registry to draw from.
• B2C funnels where conversion to first action is the north-star metric and the cost of an extra LLM call is below the lifted activation revenue.

When NOT to use it

• Regulated onboarding — KYC, medical intake, insurance — where every user must answer the same questions in the same order for compliance.
• Funnels under three screens. The LLM round-trip latency and the additional failure surface buys nothing if the flow is already short.
• Apps that need offline onboarding. Without an on-device model and a local component registry, Phase 2 falls back to a static flow anyway.
• Teams without an analytics layer that can attribute activation lift to onboarding. Without that, you cannot tell if the personalization is helping or hurting.
• When the cost of an LLM call per onboarding session exceeds the lifetime value of an activated user. Cheap acquisition channels lose to fixed-flow onboarding.