01What can be automated in the order flow

In a WooCommerce store, much of the daily work is repeating the same steps: confirming an order, saying it is ready or shipped, arranging a pickup, an installation or an appointment, and chasing orders that get stuck. These tasks are simple, but they eat hours and cause mistakes when done by hand.

  • Order confirmation on WhatsApp, in addition to email.
  • Status updates: being prepared, ready for pickup, shipped.
  • A Google Calendar event when the order includes a pickup, a scheduled delivery or a service.
  • A reminder before the appointment, with the address and what to bring.
  • An internal alert when an order does not add up: incomplete address, pending payment, uncertain stock.

There is no need to do everything at once. The first step is the notification that saves the most calls or messages. How to pick the process is in the guide on automating orders, bookings and CRM with AI.

Fig. 1 · The architecture: webhooks, a middle service and two APIs

02The architecture: webhooks, a middle service and two APIs

WooCommerce lets you create webhooks from its settings or its REST API. Each webhook has a topic, such as order.created or order.updated, a delivery URL and a secret. Every delivery carries the X-WC-Webhook-Signature header, an HMAC-SHA256 hash of the body signed with that secret, so the receiver can check that the request really comes from the store.

The middle service receives the webhook, verifies the signature against the raw body, reads the order and decides: which WhatsApp template applies, whether an appointment has to be created and whether anything needs review. It can be an n8n or Make workflow, or a small custom service. The difference is covered in the guide on connecting AI with n8n, Make and existing tools.

PieceWhat it doesWhat to watch
WooCommerce webhookReports new or updated ordersIts own secret and signature check
Middle serviceDecides what to send and what to reviewA log of every run and retries without duplicates
WhatsApp Cloud APISends template messagesApproved templates and opt-in
Google Calendar APIChecks slots and creates eventsMinimum permission and a dedicated calendar

03WhatsApp Business: templates, the 24-hour window and opt-in

With the WhatsApp Business Cloud API, a business can only start a conversation with an approved template message. Templates are categorised as marketing, utility or authentication; an order confirmation or a shipping update is a utility template. When the customer writes, a 24-hour customer service window opens, during which replies do not need a template. Outside that window, only templates can be sent.

Since 1 July 2025, Meta charges per delivered template message, with rates based on the category and the recipient’s country. Rates change, so I check Meta’s current table on every project.

The WhatsApp Business policy requires prior opt-in and a clear way to opt out. In practice, an unticked checkbox is added at checkout, the time and method of consent are stored, and consent for order updates is kept separate from consent for marketing. The policy also asks that, when replies are automated, there is a clear path to a human.

04Google Calendar: appointments, pickups and availability

When an order includes an appointment (an in-store pickup, an installation, a session), the service first checks availability with the Google Calendar API freeBusy request and, if the slot is free, creates the event with events.insert on the agreed calendar. The event stores the order number so it is not duplicated if WooCommerce resends the webhook.

Google recommends requesting only the scopes you need: checking availability works with a read-only or free/busy scope, while creating events needs write access to that calendar. A dedicated calendar is used, not someone’s personal one.

Fig. 2 · Where AI helps, and where it does not

05Where AI helps, and where it does not

Order notifications do not need AI: they are rules. AI helps with what comes next: understanding a customer’s free-text reply (“can I pick it up on Thursday afternoon?”), proposing a slot, summarising an issue for the team or sorting incoming messages. Always with a notice that the customer is talking to an assistant, and a handoff to a person.

Editing an order, granting a discount, refunding money or writing to a customer outside the agreed templates are decisions I always leave to a person. The assistant only prepares the proposal, with the order context attached, so someone on the team can confirm or discard it. That pattern is in the guides on internal AI agents with human approval and on the AI assistant for WhatsApp.

06Testing, logging and launch

Before anything goes live with real customers, the flow is tested with test orders and edge cases: a phone number without a country code, an order without opt-in, an appointment in a busy slot, a repeated webhook. Then it runs in parallel for a while, sending notifications only to the team. Every message is logged with its order, template and result.

Once it is live, I go through the send log with the team: which notifications cut down on calls, which orders end up in the review queue and which message is worth adding. How I handle each phase is explained in my AI integration service.