An AI agent for an online store is the first reply on WhatsApp: it reads what the shopper wrote, looks up the products in your catalog, drafts an answer with the real price and stock, and either waits for your team’s approval or sends it inside the rules the owner sets. In WACRM this runs on your own WhatsApp Business number, in the same inbox your team already uses. This playbook belongs to the WACRM guides for every business that sells online; the product page is WACRM AI agents.
What does a WhatsApp AI agent do for an online store?
It takes the repetitive first reply off your team. Most chats to a shop are the same few questions: is this in stock, how much is it in my size, which colour is left, where is my order. An agent answers those from your data in seconds, at 10 pm or on a holiday, and leaves the judgement calls to people.
In WACRM the agent works in four steps:
- A customer writes to your WhatsApp number and the message lands in the shared inbox.
- The agent proposes a reply, drafted from the conversation and the products in your workspace.
- Your team approves it in one tap, or the owner has turned on auto-send for replies inside the 24-hour window.
- The reply goes out from your number, and the conversation stays in the inbox for anyone to pick up.
Customer
Hi! Do you still have the linen shirt in M? How much is it?
AI agent · Proposed reply · Catalog: Linen shirt / M
Yes, the linen shirt in M is in stock at the price shown in your catalog. The S and L are in stock too. Shall I send you the link to order?
✓ Approved by your team · sent ✓✓
The same flow serves more than fashion. A cosmetics shop gets “is this safe for sensitive skin” questions, an electronics seller gets “does this charger fit my phone”, a bakery that takes online orders gets “can I collect on Saturday”. The agent answers from what you have written about each product; the rest goes to a person.
Is an AI agent allowed on WhatsApp?
Yes, when the AI serves your own customers about your own business. Meta changed the WhatsApp Business Solution terms so that, from 15 January 2026, AI model providers may not distribute general-purpose AI assistants on WhatsApp; as TechCrunch reported, a business running a bot for customer service is not barred.
Note
The rule that matters day to day: WhatsApp’s Business Policy allows automation when you respond inside the 24-hour window, but you must also offer prompt, clear and direct escalation paths. A shop agent therefore needs a way back to a person, which is covered below.
Meta also runs its own AI for business messaging, which is a separate product. A WACRM AI agent is your agent: it works only on conversations with your customers, on your number, under your settings.
How does the agent answer price and stock questions without inventing them?
It looks the product up first and only then writes. An AI model on its own will happily guess a price; a shop cannot afford that, because a wrong price in a chat becomes a dispute at checkout.
Where the product facts come from
WACRM keeps a product list per project: products you add by hand, and products synced from your store. Each row
carries a name, SKU, price, stock and its variants. WooCommerce exposes the same fields through its
REST API (for example sku, regular_price and
stock_quantity), and Shopify stores each size or colour combination as its own
variant with its own price and SKU. When a customer’s
question names a product or a SKU, the agent matches it against that list, including every variant of the matched
product, and drafts the answer from those facts.
What happens when a number doesn’t match
Before a product answer is used, WACRM checks every money amount in it against the prices stored for the matched
products. Formats such as 1,299.00, 1.299,00 and 1 299 count as the same amount; 12.99 does not. If the draft
mentions an amount that is not in your catalog, the draft is replaced with a safe line that names the matching
products, leaves the price out and says a teammate will confirm. When nothing in the catalog matches at all, the
agent says it could not find the item instead of making one up.
Two practical consequences for your catalog:
- Keep product names the way customers say them. “Linen shirt” matches a question; an internal code alone may not.
- Keep stock current in your store. The agent can only be as right as the stock number it reads.
How does it recommend products and help close the order?
It recommends from what is actually available, then hands the buyer a clear next step. The agent does not place orders by itself; the order is created at your store’s checkout or by a teammate in WACRM.
Recommending a variant that is in stock
Because the agent reads every variant of a matched product, a “do you have it in M” question can be answered with the sizes and colours that are in stock, each with its stored price. That is the most common recommendation a shop makes in chat, and it is the one that saves the sale when the first choice is gone.
Turning the chat into an order
Once the shopper is ready, there are two paths. They follow your checkout link and pay in your store, or a teammate opens Orders in WACRM Workspace, picks the contact, adds the product lines and creates the order. From there, order automation can send the order confirmation on WhatsApp from an approved template, so the customer gets the confirmation in the same chat where they asked the question.
| What the shopper asks | What the agent uses | Who sends the reply |
|---|---|---|
| “Is it in stock in M?” | Synced product, its variants and stock | Teammate approves, or auto-send inside the 24-hour window |
| “How much is it?” | Stored price of the matched product or variant | Teammate approves (catalog answers are always proposed) |
| “Can you hold one for me?” | Nothing to look up: a person decides | Agent hands the chat to your team |
| “Where is my order?” | Order status on the order record | Order automation template, or a teammate |
| “I want to talk to a person” | The customer’s own words | Agent hands the chat to your team at once |
How each row is sent follows WACRM’s AI agent and order automation settings; see the AI reply drafts guide for the exact auto-send rules.
When does the agent hand the chat to a person?
As soon as a person is the better answer. WACRM checks these signals before the agent spends anything on a draft, and again before anything is sent:
- The customer asks for a person. Phrases such as “talk to a human” are recognised in English, Vietnamese, Spanish, Portuguese, Indonesian, Hindi and Arabic.
- The agent repeats itself. After three proposed sends to one chat within five minutes, the agent stops itself.
- The wallet runs out of credit. The agent stops instead of going negative.
- The action needs a human decision, such as a free-text message outside the 24-hour window.
1 · Event
A customer message arrives on your number.
2 · Checks
AI paused? Asked for a person? Repeat limit reached?
3 · Draft
The agent drafts from your catalog; prices are checked.
4 · Result
Approved and sent, or handed to your team with a note.
When the agent hands over, AI stops for that conversation, the chat stays with the teammate already handling it or goes to the shared queue, and a private note explains why, with the agent’s unsent draft and the last few messages. Nobody has to scroll back to work out what the customer wanted. Staff can also take over at any moment simply by replying.
What does an AI reply cost on WhatsApp?
There are two separate costs: Meta’s message charge and the AI usage in WACRM.
- Meta. Replies inside the 24-hour customer-service window are service messages. Per Meta’s pricing page, from 1 October 2026 the first 1,000 service messages per business phone number per month are free (as of the check on 9 November 2026); a payment method is needed beyond that. Outside the window you need an approved template, which is billed by category. The message pricing guide has the per-market rates.
- WACRM. AI drafts are paid from your wallet balance. With default prices, drafting a reply costs 1 AI token and an automatic send is a second token, so a fully automatic reply is 2 tokens. When the wallet cannot cover a draft, the agent stops instead of running up a bill.
Because the agent answers inside the window the customer opened, most shop replies are service messages: free up to 1,000 per number per month, then billed at the service rate. Marketing follow-ups such as cart reminders are a different matter; see the abandoned-cart guide before you automate one.
How do you roll out an AI agent safely?
Start narrow, watch, then widen. A good first month for a shop looks like this:
- Clean the catalog. Customer-facing names, current stock, prices on every variant.
- Run approval-first. Every reply waits for a teammate. Read what the agent proposes for a week or two and note where it hesitates.
- Write the hard answers yourself. Returns, delivery times and payment options belong in your product and workspace details, so the agent draws on your wording instead of guessing.
- Turn on auto-send for in-window replies once the drafts look right. Product answers drafted from the catalog stay proposals.
- Keep the pause switch in reach. One switch in Messaging → Automations stops every AI action for the project, including a reply that was approved but has not gone out yet.
- Review the handoff notes weekly. They show the questions the agent should not take, and the gaps in your product details.
The AI reply drafts guide walks through the pause switch and auto-send step by step.
Does it work with WooCommerce, Shopify and your phone app?
Yes. WACRM connects to WooCommerce and Shopify, so the products the agent reads come from the store you already run, and the online shops solution shows how the inbox, order updates and AI fit together for a seller.
You do not have to give up the WhatsApp Business app either. With coexistence, which comes with the plans that include it, the same number stays active in the phone app while it also runs on the WhatsApp Business Platform through WACRM; see keep your phone app for what that means for your team. And if you already work with an AI assistant of your own, the WACRM MCP server lets an MCP-compatible client call WACRM through the open Model Context Protocol.
Put an AI agent on your shop’s WhatsApp
An AI agent earns its place when it answers the routine questions with real data and steps aside for everything else. See what WACRM AI agents can take on for your store, or let our team connect your store and set up your first order templates with Setup Commerce.
