Let the next step be decided per-user, not by a fixed cadence

Create an AI-derived attribute 'next_best_action' on the User object, refreshed daily and on every significant behavioral event. Inputs: current lifecycle stage, engagement velocity (sessions trend), feature usage breadth, recent intent signals (pricing visit, feature first-click, support touch), and prior message engagement. Outputs: a structured object with (a) recommended_action one of [educate / nurture / offer / surface-feature / surface-recommendations / handoff-to-agent / wait]; (b) recommended_channel [email / sms / in_app / push / slack-internal]; (c) recommended_content_theme; (d) confidence (0-100); (e) expected_uplift_signal. The attribute is the decisioning brain — the journey routes from it at each gate.

Build a segment 'NBA-orchestrated users' capturing active users where next_best_action.confidence >= 50 AND user is not currently in another high-priority journey (no double-orchestration). Refreshed continuously — as users' NBA changes, they move between sub-cohorts of this parent segment.

Generate email content variants per recommended_action: (a) educate variant — short value tip matched to the user's stage; (b) nurture variant — case study/customer story relevant to user's segment; (c) offer variant — personalized incentive (discount / trial extension / credit) matched to plan and tenure; (d) surface-feature variant — feature spotlight with deep-link to the in-app destination. Each variant pulls dynamic content blocks based on the user's full profile, not just first name.

Generate in-app message variants for the surface-feature and surface-recommendations branches. Surface-feature: contextual tooltip pointing to the feature, with a 1-line value prop and 'try it' CTA, rendered in the relevant in-app location. Surface-recommendations: a personalized recommendation card pulling from the user's product affinity. In-app fires only during active sessions, never outside the app (avoids interruption fatigue).

Configure a recommendation surface 'NBA-driven product/content recs' that activates when next_best_action.recommended_action = surface-recommendations. Sources: the user's prior interaction history + similar-user purchase/usage patterns + currently-trending items in their segment. Render in: in-app sidebar + email content block + page personalization slot. Refresh weekly.

Build an adaptive journey wired to the NBA segment. At each gate (signup+7d, +14d, +30d, +60d, ongoing weekly), the journey reads next_best_action attribute and routes the user down the matching branch: educate → email variant a; nurture → email variant b; offer → email variant c with discount; surface-feature → in-app tooltip + email variant d; surface-recommendations → activate recommendation surface + email digest with recs; handoff-to-agent → trigger agent conversation; wait → skip touch, recompute next gate. Each branch's outcome (clicked / engaged / converted / ignored) feeds back into the next NBA computation so the decisioning learns. Exit on: conversion event (deal_created / subscription_created / activation_milestone — depending on lifecycle), unsubscribe, or sustained no-engagement (NBA returns wait 3 gates in a row → suppress).

Compose an NBA orchestration dashboard: distribution of recommended_action across users (which actions does the model favor — sanity check on model balance), per-branch engagement rates (which actions actually convert), confidence-vs-conversion correlation (does higher-confidence routing actually predict higher conversion — model-quality signal), uplift vs control (a 5-10% holdout that gets fixed cadence — the proof-of-value chart), and per-segment NBA quality (model may serve some segments better than others — informs retraining priorities).