All use cases

Retain / Save

Off-ramp safe-transition agent

The moment a patient decides to stop the medication is the moment most programs lose them — and the moment the weight quietly starts coming back.

Close the loopGet aheadBehavioral triggerRisk triggerTime-based triggerChatPushEmailWhatsAppGLP-1 & PeptidesHealth & TelehealthSubscription

What it does

Most patients regain a large share of the weight within a year of stopping a GLP-1 medication, yet the decision to stop usually happens off-platform and unmanaged: the patient cancels, the program sends a generic save offer, and the clinical risk of an abrupt unsupported stop goes unaddressed. Stopping is not one event with one reason — cost, side-effects, pregnancy, and "I've reached my goal" each need a different response, and treating them the same loses the patient and the safety moment at once. The agent turns the stop signal into a structured off-ramp: it classifies the reason, branches to the right path, intensifies behavioral coaching through the transition, watches for early regain, and keeps a door open for a clinician-led restart.

How it works

  1. 1
    Trigger. the patient signals stopping — a cancel-flow entry, a "thinking of coming off" message in chat, a refill that lapses, or an engagement-decay score crossing its threshold near a likely off-ramp moment.
  2. 2
    Decision. the agent reads the patient's memory (stage, current dose, side-effect history, stated reason) and classifies the stop-reason, then selects the branch — pregnancy routes to an immediate safe-stop and a clinician; cost surfaces brand, compounded, or maintenance options; side-effects open the symptom and titration path; goal-reached pivots to a maintenance conversation. A judge gates every unprompted message for tone, claims, and clinical safety.
  3. 3
    Action. on the patient's consented channel (chat, push, email, or WhatsApp) the agent runs the branch — never advising a dose change itself — provides safe-stop clinical context drawn from the knowledge base, intensifies behavioral coaching for the weeks around the stop, and starts tracking regain signals such as returning appetite or weight creep.
  4. 4
    Follow-through. outcomes and the chosen reason are written back to memory; clinical decisions hand off to a clinician with full context attached; if the patient stays, queued cancel-flow outreach is cancelled; if they leave, a nudge-back path stays armed for the regain-triggered win-back. Save and regain-rescue impact is measured against a holdout.
Example
AgentYou mentioned coming off the medication — that's your call to make, and I want to make sure you come off it safely rather than just stopping. Can I ask what's driving it: the cost, how you're feeling on it, or something else?
CustomerHonestly it's the price now that my dose went up. I think I'm just done.
AgentThat's the most common reason, and there are a few legitimate routes that cost less before you stop entirely — a lower-cost form, or a lighter maintenance plan that protects the progress you've made. I can lay those out, and if you still want to come off, I'll make sure your clinician sets up a safe wind-down. Which would you rather start with?

Configuration

How the agent is wired for this use case.

Triggera stop signal — a cancel-flow entry, a discontinuation-intent phrase in chat, a lapsed refill, or an engagement-decay score near an off-ramp moment — sourced from the program app, the subscription billing system, and the messaging channel.
Tools & actions
  • Customer record / CRM · read stage, current dose, side-effect history, and stated reason; classify the stop-reason; write the reason and outcome back
  • Subscription billing · read plan and lifecycle state; surface brand, lower-cost, or maintenance-tier options
  • Knowledge base · retrieve safe-stop clinical context and the reason-specific coaching material
  • Clinical escalation path · route pregnancy, abrupt-stop, and any dose decision to a clinician with full context
  • Messaging channel · run the branched off-ramp conversation and the behavioral-coaching touches
Autonomyreason-classification, coaching, and option-surfacing run unattended behind judge gating; this is a detect-and-route clinical posture — the agent never changes or recommends a dose, and pregnancy or abrupt-stop signals escalate to a clinician under the clinical-safety policy class.
Channelschat · push · email · whatsapp
Escalationpregnancy, a clinically unsafe abrupt stop, or any dose or wind-down decision hands off to a clinician.

What you need

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

Signals

cancel-flow entry, discontinuation-intent phrases, refill lapse, engagement-decay scores, early regain markers (returning appetite, weight creep)

Data

program stage and current dose, side-effect history, stated stop-reason, plan and pricing options, channel consent, off-ramp coaching memory

Guardrails

judge gating on every unprompted message; detect-and-escalate only, no dose advice; pregnancy and abrupt-stop hard-escalate to a clinician; frequency caps; stale-outreach cancellation on organic stay; holdout measurement of saves and regain rescues

Metrics it moves

  • save-rateup, by meeting the stop decision with a reason-matched path instead of one generic offer
  • churndown, as cost and side-effect stops convert into lower-cost or maintenance continuations
  • reactivation-rateup, because the armed nudge-back path catches regain early after a clean exit
  • ltvup, with patients carried into maintenance or restarted safely rather than lost on high-CAC economics

Related use cases

Upsell & Cross-sell

Maintenance-program upsell at goal weight

the positive-completion pivot the goal-reached branch routes into

Win back

Regain-triggered win-back

the nudge-back path armed when a patient does exit

Retain / Save

Food-noise early-warning save

the leading appetite signal that feeds regain tracking

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