AI-Powered Customer Support
and Payment Automation

The owner was the support team, the sales team, and the payment processor. All at once. On WhatsApp.

The client ran a service business where the majority of customer interactions happened on WhatsApp. Enquiries came in throughout the day and into the evening. Each one required the owner to stop what they were doing, read the message, understand what the customer was asking, explain the relevant service, collect the necessary details, generate a payment link manually in Paystack, paste it into the chat, and then wait to confirm payment before sending a receipt.

This was not a small overhead. The owner was fielding upward of 20 conversations a day, each requiring multiple back-and-forth messages before reaching a payment. The cost was not just time. It was attention. Every interrupted work session to answer a WhatsApp message was a context switch that reduced the quality of the actual service being delivered.

The brief was to automate the entire pre-payment customer journey on WhatsApp so the owner only needed to get involved after payment was confirmed.

The AI needed to handle the full conversation, not just the FAQ layer.

The standard approach to this problem is a FAQ bot: the customer asks a question, the bot looks up the answer, the customer pays via a generic link. That approach fails at the edges, and service businesses live at the edges. Customers ask questions the FAQ does not cover. They negotiate. They need reassurance. They provide context that changes which service is appropriate.

I designed the AI to handle the full conversation dynamically rather than matching inputs to predefined responses. The prompt architecture gave the AI a complete brief on every service offered, the pricing structure, the information it needed to collect from each customer before generating a payment link, and clear escalation instructions for anything outside its scope. The AI handled qualification and service recommendation in the same conversation thread before ever touching the payment step.

The Paystack integration was the technically critical piece. Payment links in Paystack can be generated via API with specific amounts and metadata. I built the workflow so the AI, once it had confirmed the service and collected the customer's details, handed off to n8n to generate a unique payment link, pass it back into the WhatsApp thread, and then listen for the Paystack webhook confirming payment before triggering the confirmation emails.

One end-to-end workflow covering enquiry, qualification, payment, and confirmation.

• 01
  Conversational AI on WhatsApp

  Every inbound WhatsApp message fires a webhook to n8n. The message is passed to Claude AI alongside the full service catalogue, current pricing, and a structured prompt that instructs the AI to understand the customer's need, ask any necessary clarifying questions, recommend the appropriate service, and collect the required details before initiating payment. The AI maintains conversation context across the full thread so it never asks for information the customer has already provided.
• 02
  Dynamic Paystack payment link generation

  Once the AI confirms the service and has all required customer details, it signals n8n to generate the payment link. n8n calls the Paystack API with the exact service amount, the customer's name and email, and a transaction reference tied to the conversation. The generated link is passed back to Claude, which sends it to the customer in WhatsApp with a brief, friendly confirmation message explaining what they are paying for and what happens next.
• 03
  Payment confirmation and dual notification

  Paystack fires a webhook to n8n on successful payment. The workflow verifies the transaction reference matches the open conversation, then sends two emails simultaneously: a branded receipt to the customer confirming payment and outlining the next steps, and an internal notification to the business owner with the customer name, service purchased, amount paid, and all collected details. The owner sees a completed sale in their inbox rather than a pending conversation in WhatsApp.

The owner stopped being a bottleneck in their own business.

The most immediate change was that the owner stopped monitoring WhatsApp during working hours. Customers received instant, accurate responses regardless of when they messaged. The first payment completed through the system at 11pm on a Saturday, a transaction that would previously have waited until Monday morning.

The quality of the customer experience also improved. The AI never forgot to mention a relevant detail, never misquoted a price, and never left a customer waiting mid-conversation. Customers consistently remarked that the interaction felt responsive and professional, not automated, which was the intended effect of the conversational design.

For the owner, the change was in how they experienced their business at the end of each day. Instead of a WhatsApp inbox full of half-completed conversations and unpaid links, they had a clean email trail of completed transactions and a clear picture of the day's revenue without having been in the middle of it.