A Decision Framework for Keyword-Triggered Automation in Pharmacy WhatsApp

Why pharmacy keyword routing is harder than most verticals

Pharmacy WhatsApp messages arrive across a much wider range of urgency and specificity than most retail or service verticals. A restaurant message is usually a booking or a takeaway order — the range is narrow. A pharmacy message can be:

• A routine chronic-medication refill request.
• A question about medication interaction that requires pharmacist judgment.
• A report of a potentially serious side effect requiring immediate escalation.
• A parent asking whether their child's fever medication dosage is correct.
• An insurance-related question about NHIF/HMO/private insurance coverage of a specific medication.
• A stock availability question about a specific SKU.
• A prescription photograph submission for verification.

Each of these needs a different response. Automated response to the routine refill can be immediate and productive. Automated response to the side effect report is dangerous — it delays escalation to the pharmacist who needs to act. The routing decision has real consequences.

This is why generic keyword-triggered automation designed for other verticals fails in pharmacy contexts. The framework needs to be specifically designed for pharmacy message patterns. Here is the one I recommend.

First branch: check for emergency indicators

The first branch of any pharmacy keyword-triggered automation should check for emergency indicators before doing anything else. Not after processing the message; before.

Emergency indicator keywords in the pharmacy context include (in the primary languages of the pharmacy's clientele):

• Direct urgency: 'emergency', 'urgent', 'immediately', local equivalents in Swahili, Hindi, Arabic, Portuguese, Spanish, etc.

• Symptom severity: 'severe pain', 'bleeding', 'unconscious', 'difficulty breathing', 'chest pain', 'anaphylaxis', 'overdose'.

• Adverse reaction indicators: 'allergic reaction', 'side effect', 'reacting badly', 'swelling', 'rash spreading', 'vomiting continuously'.

• Family emergency indicators: 'my child', 'my baby', 'my mother collapsed'.

Detection of any of these triggers immediate escalation to the pharmacist on duty (or the on-call pharmacist if outside hours). The automation dispatches a specific escalation template: 'This message has been marked urgent and forwarded to our pharmacist. They will contact you within [X] minutes. If your situation is life-threatening, please call emergency services immediately at [number].'

This is not a nice-to-have. This is the difference between an automation system that supports pharmacy operations and one that damages them.

Every pharmacy WhatsApp deployment I've seen fail at some point has failed on this axis. The message that should have gone to a pharmacist immediately went to a routine queue. The client didn't get the response they needed. Bad outcomes followed.

Second branch: check for specific prescription-related workflow

Once emergency detection has passed (message doesn't contain emergency indicators), the second branch checks for specific prescription-related workflow indicators.

Prescription image detected: automation acknowledges receipt, files the image against the client's record for pharmacist review, and dispatches 'Your prescription has been received. Our pharmacist will review it and confirm within [X] hours.' Do NOT automate the pharmacist review — the pharmacist must actually review the prescription before dispensing.

Refill request for chronic medication (client has a prescription on file for the specific medication): automation coordinates the refill workflow — confirms current stock, dispatches payment coordination, arranges pickup or delivery. Human involvement minimal after initial setup.

Refill request for medication requiring new prescription: automation acknowledges the request but routes to human handling — 'To refill this medication, you'll need to see your doctor for a new prescription. Would you like us to help you schedule that?'

Question about interaction between medications: this requires pharmacist judgment. Automation acknowledges and routes to pharmacist. Do NOT auto-respond with generic interaction information; even if the specific question has an obvious answer, the pharmacist should verify the client's specific medication list before responding.

Question about a specific patient's dosage: pharmacist routing. Especially important for pediatric dosages and elderly patient dosages where individual factors matter.

Third branch: check for operational and administrative content

The third branch handles operational and administrative messages that automation can process without pharmacist involvement.

Stock availability inquiry for OTC product: automation can check the pharmacy's inventory system and respond directly. 'Yes, we have [product] in stock at [location].' Or 'That product is out of stock; expected back Tuesday. Would you like us to hold one for you?'

Delivery status inquiry: automation queries the delivery tracking system and dispatches specific status. 'Your delivery is with the courier, expected between 14:00 and 16:00 today.'

Hours and location inquiries: automation dispatches standard information with specific detail (holiday hours if applicable, specific branch closures, etc.).

Insurance coverage inquiry: this is a harder branch. If the specific medication is on the specific patient's covered list (data the pharmacy has), automation can respond. If the answer requires querying an insurer's system in real time, or requires human judgment about specific coverage details, route to human.

Payment confirmation and receipt requests: automation can dispatch payment coordination and post-payment receipts directly.

Appointment scheduling (for pharmacies offering vaccination or consultation services): automation handles routine booking; complex cases (specific patient histories, potential contraindications) route to human.

Fourth branch: general messages and fallback

The fourth branch handles everything that didn't match the previous three.

Greetings and small talk: automation can respond briefly and warmly, then ask how it can help. 'Hi, thanks for reaching out. What can we help you with today?'

Unclear or ambiguous messages: rather than guess, automation should ask a clarifying question. 'Can you tell me a bit more about what you need? We can help with refills, prescription submission, stock questions, or you can reach our pharmacist directly.'

Messages in languages the automation doesn't have templates for: route to human with a note about the language. Attempting automated translation for pharmacy content is risky; wait for human review.

Complaints or dissatisfaction: route to human. Automated responses to complaints damage the client relationship more than a slower human response.

Unrecognised keywords or nonsensical content: default to human review. The cost of routing a routine message to a human is small; the cost of the automation misfiring on a serious message is large.

The overall framework

The decision framework, in order:

1. Emergency indicator? → immediate pharmacist escalation with specific template.

2. Prescription-related workflow? → route by specific sub-branch (image/refill/interaction/dosage question), with pharmacist review always required for dispensing decisions.

3. Operational and administrative? → automate where automation has the data; route to human where it doesn't.

4. General or ambiguous? → clarifying question or fallback to human.

The order matters. Every message goes through branch 1 first. Emergency detection is never skipped, never deferred, never dependent on other processing succeeding.

Pharmacies that implement this framework report meaningful improvements in both operational efficiency (routine messages handled at scale) and safety (serious messages reach pharmacists reliably). Pharmacies that skip the framework — using generic keyword triggers designed for other verticals — encounter both operational chaos and safety incidents.

The framework is not automation-vendor-specific. It should be implementable in any competent BSP or CRM. If your automation platform can't support this branching structure, it's the wrong platform for pharmacy operations.