Engage / Lifecycle
Proactive bubbles and personalized push
A small bubble or push that knows what the user just did, says something worth tapping, and opens straight into a conversation.
What it does
Most app notifications are templates fired on a schedule, and users learn to swipe them away. This replaces them at both layers. Inside the app, contextual bubbles appear at the right product moment — a meal just logged, a streak about to break — and open an entry point into the AI chat instead of a dead-end banner. Outside the app, template push gives way to messages the model writes per user, and the same model judges whether the message is worth sending at all, so silence is a valid output. Small, efficient models keep the per-message cost low enough to run this across the full audience, which is what normally blocks personalized push at scale.
How it works
- 1Trigger. a behavioral or time signal from the product —
meal_logged, a streak or inactivity timer, any moment the client defines as worth a word. - 2Decision. the agent drafts a short, personal message from the user's recent activity and remembered context; a quality judge scores relevance, timing, and tone, and drops anything that doesn't clear the bar.
- 3Action. an in-app bubble at the moment of the action, or a personalized push through the client's own provider — tapping either one lands in a live conversation, not a static screen.
- 4Follow-through. if the user opens the app first or the moment passes, the queued send is cancelled before it fires; opens and replies write back to analytics, and lift is measured against the templated baseline.
Configuration
How the agent is wired for this use case.
meal_logged, a streak or inactivity timer, any moment the client defines as worth a word.- Product analytics · ingests the triggering events and receives opens and replies back for measurement.
- Memory layer · reads recent in-app activity and remembered preferences to ground the draft.
- In-app messaging / bubble surface · renders a contextual bubble at the product moment that opens into live chat.
- Messaging channel · sends the per-user push through the client's own provider.
What you need
The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.
Signals
product events worth a moment (meal_logged, workout_completed, streak timers, inactivity thresholds).
Data
recent in-app activity, the user's remembered preferences and goals, notification consent and device tokens.
Guardrails
judge review of every unprompted message with "don't send" as a first-class outcome, frequency caps and quiet hours, instant opt-out, cancellation of stale sends.
Metrics it moves
- click-through-rateup, a message written for one user about what they just did beats any template.
- dau-mauup, each tapped bubble or push is a session that starts in conversation.
- contact-ratedown, the bubble answers the question at the moment it forms, before it becomes a ticket.
- opt-out-ratedown, relevance and the judge's silence option replace notification fatigue.
Related use cases
See it on your own customer journey
Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.
Book a demo