All use cases

Retain / Save

Food-noise early-warning save

The constant chatter about food goes quiet on a GLP-1 medication — and when it starts creeping back, that is the first sign a patient is slipping, weeks before the scale shows it.

Get aheadBehavioral triggerRisk triggerChatPushWhatsAppGLP-1 & PeptidesHealth & Telehealth

What it does

The quieting of "food noise" — the intrusive, all-day thinking about food — is one of the first things patients notice on a GLP-1 medication, and its return is one of the first signs the effect is fading: during a taper, after a missed dose, or in the run-up to quietly stopping. By the time weight moves, the patient has often already drifted. The agent treats returning food noise as a leading churn signal in its own right: it tracks what the patient reports about appetite and cravings over time, catches the upward trend early, and opens a behavioral or clinical check-in before a stall ever shows on the scale.

How it works

  1. 1
    Trigger. the patient self-reports rising appetite, cravings, or "food noise is back" in chat, in a logged check-in, or in answer to a routine prompt; or a known risk context is live (an active taper, a recently missed dose).
  2. 2
    Decision. the agent compares the new report against the patient's own food-noise baseline and recent trend, weighs context from memory (titration stage, missed-dose history, whether a taper is underway), and decides whether this is a normal day or a genuine upward drift worth acting on. A judge reviews every unprompted message before it sends.
  3. 3
    Action. on a real upward trend the agent sends a short, supportive touch on the patient's channel — naming the signal, normalizing it, and offering a concrete behavioral step or a check-in — written to pull the patient back into the program's coaching conversation rather than replace it.
  4. 4
    Follow-through. if the pattern points to a clinical cause (a taper not holding, repeated missed doses) the agent routes it to the care team with the trend attached; it never adjusts or advises a dose. If the patient re-engages on their own, queued touches are cancelled, and lift is measured against a holdout.
Example
AgentYou mentioned food is back on your mind a lot this week. That quiet you had before is real, and it can get louder around a missed dose or a taper. It often settles again — want to talk through what is going on this week?
CustomerYeah, honestly I forgot last week's shot and now I'm snacking constantly. Starting to think it's stopped working.
AgentThat lines up with the missed dose, and it is a common pattern, not a dead end. I've shared this with your care team so they can look at the timing, and in the meantime let's set a protein-first plan for the next few days.

Configuration

How the agent is wired for this use case.

Triggera self-reported appetite / cravings / "food noise" entry in chat or a logged check-in, or an active risk context (taper in progress, a recently missed dose) — surfaced as a food_noise_reported signal or read from check-in logs.
Tools & actions
  • App backend · read self-reported food-noise and appetite entries and compare each against the patient's own baseline and recent trend.
  • Patient-record system · pull memory of titration stage, missed-dose history, and whether a taper is underway to judge whether the trend is real.
  • Messaging channel · send the early-warning coaching touch on the patient's channel, written to reopen the program's coaching conversation.
  • Clinical escalation path · route a trend pointing to a clinical cause (taper not holding, repeated misses) to the care team with the trend attached.
Autonomytrend detection and the supportive coaching touch run unattended, judge-gated on every unprompted message and limited to approved coaching content; under a detect-and-escalate clinical policy any clinical cause is surfaced to a clinician and the agent never adjusts or advises a dose.
Channelschat · push · whatsapp
Escalationa trend that points to a clinical cause, or any red-flag answer, routes to the care team; the agent surfaces the signal and never makes the dose decision.

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

food_noise_reported or appetite / cravings check-in entries, missed-dose events, taper-in-progress status, engagement-decay detection.

Data

the patient's food-noise and appetite history (for a personal baseline), titration stage, missed-dose and taper context, channel consent state, coaching memory.

Guardrails

judge review on every unprompted message; detect-and-escalate clinical policy — no dose advice, ever; clinical causes always route to a clinician; approved coaching content only; frequency caps per patient; stale-outreach cancellation; holdout measurement of the save lift.

Metrics it moves

  • churndown, by catching the earliest leading signal of fade and reopening the conversation before a stall or a silent stop.
  • save-rateup, because an appetite-return touch reaches the patient weeks ahead of the weight-based plateau play, when a save is still easy.
  • adherenceup, since a returning-appetite report often traces to a missed dose the check-in can help the patient and clinician address.

Related use cases

Retain / Save

Weight-stall plateau intervention coach

the later save once weight itself goes flat

Retain / Save

Off-ramp safe-transition agent

the structured path when a patient is actually stopping

Win back

Regain-triggered win-back

the same appetite signal used to reach a patient who already stopped

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