All use cases

Prevent / Care

Off-label request escalation guard

A patient asks the chat to double their dose to catch up — the agent does not answer it clinically, it routes the request straight to the prescriber.

Close the loopNext best thingRisk triggerBehavioral triggerChatWhatsAppIn-appGLP-1 & PeptidesHealth & Telehealth

What it does

The most dangerous questions on a GLP-1 program are not symptoms, they are requests: "can I double up to catch up", "can I use my partner's pen", "can I take this for something it wasn't prescribed for". A helpful-by-default assistant treats those like any other question and answers them, which is exactly how a wellness product drifts into practising medicine. The agent classifies intent before it composes a reply: any off-label or out-of-plan request is recognized as a request to change the prescription, and instead of advising it captures the ask, hands the prescriber a clean summary, and tells the patient a clinician will decide. The hard boundary that keeps the product on the right side of the line is that the agent never takes an autonomous prescribing action, ever.

How it works

  1. 1
    Trigger. a patient message that reads as a dose, supply, or use change rather than a symptom report — double-dosing, sharing or borrowing a pen, using the medication outside the prescribed plan.
  2. 2
    Decision. an intent classifier separates an off-label or out-of-plan request from an ordinary support question; on a prescribing-change intent the agent does not reason toward a clinical answer, it routes, with conservative defaults that escalate when the ask is ambiguous.
  3. 3
    Action. the agent acknowledges without advising, assembles the request and the patient's context into a clean summary, and files it to the prescriber; the patient is told a clinician will review and respond.
  4. 4
    Follow-through. automated replies stand down on that thread once it is with the prescriber, the clinician's decision writes back to memory, and the patient is notified when the prescriber replies.
Example
PatientI forgot last week's shot — can I just take two this week to catch up?
AgentThat is a change to how your medication is prescribed, so it is not something I can advise on — but I have sent your question straight to your prescriber with your recent doses attached, and they will come back to you today. Please don't change your dose in the meantime.
PatientOkay, makes sense. I'll wait to hear from them.

Configuration

How the agent is wired for this use case.

Triggeran inbound message classified as an off-label or out-of-plan request — a dose change, pen sharing or borrowing, or non-program use — rather than a symptom report
Tools & actions
  • Messaging channel · acknowledge the request without giving clinical advice and notify the patient when the prescriber replies
  • Clinical/EHR system · assemble the request plus current prescription and dose history into a clean summary and file it to the prescriber
  • Memory store · log the request against the patient's history and write the clinician's decision back
  • Knowledge base · retrieve approved boundary-setting language for off-label and out-of-plan asks
Autonomyintent classification and routing run unattended under a clinical-safety policy class — detect and escalate only; the agent gives no dose or prescribing advice and takes no autonomous prescribing action, ever; every message is judge-gated
Channelschat · whatsapp · in-app
Escalationany off-label or out-of-plan request is a mandatory hand-off to the prescriber with full context; automated replies stand down once the clinician owns the thread

What you need

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

Signals

in-conversation requests to change dose, share or borrow a pen, or use the medication outside the plan; the patient's message history

Data

current prescription and dose history, program plan and stage, persistent patient memory, consent state

Guardrails

clinical-safety policy class — detect and escalate, never advise; no autonomous prescribing action ever; conservative defaults that escalate on an ambiguous request; judge gating on every message; automated replies stand down once a clinician owns the thread; full audit trail of every escalation decision

Metrics it moves

  • safe-escalation-rateup, as prescribing-change requests reliably reach a clinician instead of being answered by an assistant
  • contact-ratedown on the clinical team for the routine asks the agent acknowledges and routes cleanly, up only where a prescriber genuinely needs to decide
  • csatup, since the patient gets a fast, honest hand-off rather than an unsafe answer or a dead end

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