I Built an AI-Powered WhatsApp Refill System for a Supplement Store — Here’s How It Works
By Rohit KumarAI Automation & Digital Growth ConsultantBuild Log I didn’t build this because AI automation is fashionable. I built it because a simple business process was becoming repetitive: customers buy supplements, eventually need a refill, and somebody has to remember when to follow up, find the previous order, answer the customer and move the order to the next stage. That is exactly the kind of process where automation makes sense. The objective was not to replace the business owner or turn every customer message into an AI conversation. The objective was much simpler: connect the repetitive parts of the refill journey so the business can spend more time dealing with customers and less time moving information between systems. My rule while building this: Don’t use AI for a problem that a simple automation can solve. The business problem I was trying to solve A supplement store has a natural repeat-purchase cycle. Someone buys whey protein, creatine, a mass gainer or another product. At some point, that customer may need the same product again. The difficult part isn’t sending one WhatsApp message. The difficult part is remembering the right customer, finding the previous order, understanding the reply, checking the order state, handling payment and keeping the business record updated. When those steps are handled manually, the process becomes dependent on memory. As order volume grows, that creates unnecessary administrative work and increases the chance of missed follow-ups. The question I asked was simple: How much of this journey can happen automatically while keeping the business owner in control when something requires a human decision? What I actually built The system combines a customer/order dashboard with an automation layer. The workflow shown in the build uses webhooks, order data, WhatsApp, conditional logic, payment-related actions, AI-assisted message processing and status updates. The dashboard gives the business a single operational view instead of forcing someone to inspect individual conversations and spreadsheets every time an order changes stage. Dashboard from the demonstrated refill automation build. Customer information should be anonymized before public publication. What the dashboard is actually tracking The dashboard screenshot gives a useful view of the operational state of the system. In the demonstrated snapshot, it showed 14 total orders, 14 reminders sent, no pending approvals, one active delivery and no completed deliveries. Thirteen orders were shown in the reminder/chatting stage, while one was shown as booked or out for delivery. The purpose of these numbers is to demonstrate the workflow state at the time of the snapshot — they are not being presented as business performance or revenue results. 14Orders in dashboard snapshot 14Reminders shown as sent 1Active delivery 0Pending payment approvals The complete customer journey Instead of treating WhatsApp, orders, payments and delivery as separate activities, the system connects them into a single workflow. Previous Order → Refill Reminder → WhatsApp Reply → Order Lookup → Payment → Delivery Where the automation starts The workflow contains separate webhook-driven paths for different business events. A refill reminder can enter the system, be normalized into the expected data format, trigger a WhatsApp message and then update the corresponding order. This is important because the workflow doesn’t need an AI model to decide everything. A scheduled reminder is a deterministic event. The system knows what it needs to do. Automation principle #1 If the input and the required action are predictable, use rules and automation first. Then WhatsApp becomes the interaction layer The interesting part starts when the customer replies. A message coming from WhatsApp isn’t always structured like a database field. A customer might write: Example: “Yes, send the same whey again.” “I already paid. Please check.” “Can you send it tomorrow?” Those messages contain business intent, but the information isn’t necessarily formatted as a clean command. This is where an AI layer becomes useful. AI is used where the input becomes messy In the demonstrated workflow, an incoming WhatsApp message can be parsed, the relevant order can be identified, a prompt can be built and an AI-assisted response can be generated before the workflow continues. That distinction matters. I don’t want to call every webhook an AI agent. Most of the system is ordinary automation — and that is a good thing. Business task Best approach Why Find an order by phone Automation Deterministic lookup. Check an order status Automation A database value should decide the result. Send scheduled refill reminder Automation No AI reasoning is necessary. Understand an unstructured customer reply AI-assisted Natural language contains variable intent. Generate a natural response AI-assisted AI can turn structured information into a human-readable reply. Update order status Automation The final state should be controlled by business logic. Handle unusual or risky situations Human review Not every decision should be delegated to a model. How the payment workflow fits into the system Payment is another point where the workflow branches. The demonstrated automation includes a payment approval path that reads order data, identifies the relevant order, makes a decision and then follows an approved or rejected branch. There is also a WhatsApp path for payment-related customer messages. The workflow can identify whether a message contains a payment screenshot, build the relevant AI prompt and continue into the payment stage. This is where business automation becomes more than a chatbot. The customer isn’t simply talking to an AI. The conversation is connected to an actual business record and an operational process. The architecture behind the build The original workflow contains several separate automation paths rather than one giant chain. That makes sense because a reminder, payment approval, booking confirmation and order reset are different business events. Simplified architecture of the refill automation WhatsApp Customer messages Webhooks Business events n8n Workflow orchestration Order Data Customer & order records AI Layer Unstructured message handling Business Actions Payment, booking, delivery What the n8n workflow is doing The workflow screenshot shows several independent automation paths. The exact implementation can evolve, but the architecture can be understood in a few practical layers. 1. Reminder workflow A reminder webhook enters
