This guide follows the WhatsApp chats an online store gets every day — who writes, what they ask, how an AI Agent answers and where the chat ends. Names, products and prices in the examples are made up. For the same idea in other industries, see AI Agents for Real Estate Sales and AI Agents for Education and Coaching.
AI Agents for E-commerce and D2C: Quick Answer
The short version:
What it does: an AI Agent answers shoppers on WhatsApp before and after the purchase: it recommends a product from your own catalogue, answers size, delivery, COD and return questions, and hands the chat to your team when a decision is needed.
Where it helps most: “where is my order?”, questions that stall a purchase, return requests, and replies to cart reminders — above all on sale days, when the inbox fills up.
What stays with people: refunds, replacements, address changes after dispatch, discounts and anything the agent has no approved answer for.
In WABB: a knowledge database, an AI Agent with instructions, boundaries and actions (run a flow, tag the chat, keep values), and a fallback to your team. Order data reaches WABB through Shopify events, not a live lookup.
Where Online Store Chats Break
It is 11 pm on the first night of a sale. A shopper has sneakers in her cart, is unsure about the size and wants to know whether cash on delivery works, so she writes to the store’s WhatsApp number. The reply comes the next afternoon, and by then she has ordered elsewhere. This is an illustration, not a case study.
A store sells a product through a series of small questions before the purchase and a series of small problems after it. Six things go wrong again and again.
“Where is my order?” fills the inbox
After dispatch one question keeps coming back. Each time, a person stops what they are doing to open the order and paste a tracking link.
Pre-purchase questions go unanswered
Which size, which shade, is it fragrance-free, is cash on delivery available. A shopper who cannot get an answer in the moment may leave.
Returns become long threads
Order number, photo, reason, replacement or refund — collected one message at a time by whoever is online.
Cart reminders are one-way
A reminder with a link cannot answer the question that stopped the purchase. The shopper has to find another way to ask.
Sale days overwhelm the team
Chats can arrive faster than people can read them. Messages that are about to become orders wait next to the ones about parcels.
Order data and the chat live apart
The store knows the order; the inbox knows the conversation. Support switches between the two, and the shopper repeats the order number.
Basic automation already covers the notifications — see WhatsApp + Shopify: How to Automate Orders, Support and Sales. The gap is the conversation that starts when the shopper replies.
What an AI Agent Can Actually Do for an Online Store
An AI Agent reads a free-text message, uses your own catalogue and policies, and decides what to ask or do next. In a store chat that comes down to seven jobs:
| Job | The shopper says | The agent |
|---|---|---|
| Recommend a product | “I have oily skin, budget under ₹1,000.” | Asks one or two questions, matches the answers to products in your catalogue and explains the choice in a sentence. |
| Answer size, delivery and COD questions | “Is cash on delivery available? Does it run small?” | Answers from your size guide and delivery policy. If the answer is not there, it says so instead of guessing. |
| Show order status | “Where is my order?” | Replies with the tracking link and status your Shopify event stored for this customer. If nothing is stored, it hands the chat over and does not invent a date. |
| Start a return or exchange | “The cap arrived broken.” | Explains the policy, collects the order number and the reason, tags the chat and passes it to support. It never promises a refund. |
| Continue after a cart reminder | A reply to the reminder: “Is size 8 in stock?” | Answers within the open window, points back to the checkout link and says plainly what it cannot see, such as live stock. |
| Qualify a gift or bulk enquiry | “I need 40 gift boxes for a wedding.” | Asks quantity, date and budget, then moves the chat to a sales person with the answers attached. |
| Hand over with context | “I want to talk to someone.” | Assigns the chat to a person, who sees everything the shopper has already said. |
The list leaves out refunds, discounts, address changes after dispatch and payments. Those are decisions or actions in your store, not answers — see How It Works in WABB for who takes them.
Four Use Cases: From the First Question to the Handover
Each case starts with a problem the store already has, shows the chat that solves it and ends with what the team receives. The conversations are illustrations, not transcripts of real shoppers. Read them for what the agent does between the lines: it asks only what it needs, says what it cannot see, and ends on a clear next step.
A product recommendation before the purchase
Skincare, fashion, home and lifestyle brands
The problem. A shopper cannot tell which product fits and does not want to read ten product pages. Someone on the team ends up answering the same “which one should I take?” again and again.
Store assistant
AI Agent · WhatsApp
Hi, I have oily skin and get breakouts. Need something under ₹1,000
Hi Meera! Happy to help. Will you use it morning and night, and does fragrance bother your skin?
Both times. No fragrance please
Two fragrance-free options from our range: the Gel Cleanser at ₹449 and the Niacinamide 5% Serum at ₹599. Together that is ₹1,048, a little over your budget. Which matters more to you, cleansing or a leave-on treatment?
Serum. Is it ok for sensitive skin?
The product page says it is fragrance-free and suggests a patch test first. I cannot promise how your skin will react. Here is the serum: yourstore.com/niacinamide-serum. Shall I send the cleanser link too?
Just the serum, thanks
- Asked
- Routine, fragrance sensitivity and which of two items matters more.
- Remembered
- Oily, breakout-prone skin, no fragrance, a budget of ₹1,000.
The result. The shopper gets two fitting options and a link within a minute, and nobody on the team had to type the explanation.
“Where is my order?” and an address change
Any store that ships parcels
The problem. A parcel is late, and the shopper wants to change the delivery address because they are travelling. Support has to open the order, find the link and then explain what can and cannot be changed.
Store assistant
AI Agent · WhatsApp
Order #1048 was supposed to come yesterday. Where is it??
Sorry about the wait, Priya. For order #1048 I have this tracking link: yourstore.com/track/1048. I do not see the carrier’s live scan, so I cannot give you today’s location or promise a delivery date. The link shows the latest status.
It says out for delivery but I am travelling. Can you change the address?
Once an order has shipped I cannot change the address myself, and I would rather not guess what the carrier allows. I will pass this to our support team. What address should I give them, and who will receive the parcel?
Flat 4B, Palm Residency, Pune. My brother Karan will take it
Thank you. I have noted order #1048, the new address and Karan as the receiver. Support will check with the carrier and reply here.
- Asked
- The new address and who will receive the parcel.
- Remembered
- Order number and tracking link, which came from the Shopify event saved on this contact — not from a lookup.
The result. Support receives the order number, the address request and the name of the person who will receive the parcel in one message.
A damaged item and a return request
Stores with returns and exchanges
The problem. A shopper writes that an item arrived damaged. The team needs the order number, a photo and a choice between a replacement and a refund, and collects them across several messages.
Store assistant
AI Agent · WhatsApp
my order arrived with a broken cap, I want to return it
I am sorry about that, Rohan. Our policy covers damaged items reported within 7 days of delivery. Could you send your order number and a photo of the damage?
#1102 — photo attached
Thank you. Would you prefer a replacement or a refund? I cannot approve either myself, but our team will decide with your order and photo in front of them.
replacement
Noted: replacement for order #1102. Support will confirm here. Please keep the packaging until they do.
- Asked
- Order number, photo, replacement or refund.
- Remembered
- Damaged cap, order #1102. The photo stays in the chat for the team; the agent does not judge the damage.
The result. The return request reaches support complete: order number, reason, the photo in the chat and the shopper’s preference.
A cart reminder that turns into a conversation
Stores with cash on delivery or cart reminders
The problem. A shopper left sneakers in the cart because of one open question, size or payment. A reminder with a link does not answer it.
Store assistant
AI Agent · WhatsApp
Hi Sana, you left the Classic Runner (size 7) in your cart. Need help with the size or delivery? Just reply here.
is COD available? and I am between 7 and 8
Yes. Cash on delivery is available for orders up to ₹3,000, and this pair is ₹2,499. The size guide says to go up half a size if you are between sizes in this model, so size 8. Want to continue with that?
ok 8, is it in stock?
I cannot see live stock in this chat, so I will not promise it. The product page lists the sizes available. If size 8 is shown, you can order from the same link: yourstore.com/classic-runner. If it is not, tell me and I will ask the team.
it shows 8
Then you are all set — pick cash on delivery at checkout. Write here if anything looks off.
- Asked
- Nothing extra: the shopper’s two questions were enough.
- Remembered
- Size 7 in the cart, half a size up, cash on delivery.
The result. The question that stopped the purchase is answered, and the shopper goes back to the checkout link without a call.
Many of these chats begin with an ad. If yours do, the Click-to-WhatsApp ad examples include catalogue and offer ads with the first reply for each.
How It Works in WABB: Knowledge, Agent, Store Events, Handoff
The four use cases run on six pieces set up in AI Lab; the step-by-step guide to building the agent shows every screen. Two things come first: the WhatsApp number has to be connected to the WhatsApp Business Platform — the API requirements checklist lists what to have ready, including the website Meta checks — and AI Starter has to be active under Settings → Subscriptions, with credits in Billing → Usage.
| Piece | What it does for a store | Where in WABB |
|---|---|---|
| Knowledge | Catalogue, size guides, delivery, COD and return policies, approved offers. Upload files (PDF, DOCX, PPTX, TXT, MD; up to 10 MB) or add approved answers as question groups. | AI Lab → Knowledge databases |
| Instructions and actions | How the agent behaves in repeatable situations. An Action inside an instruction can run a flow, tag the chat or keep a value. | Logic |
| Boundaries | What it must not do: invent stock, discounts or dates, promise refunds. | Logic |
| Fallback and handoff | A retry limit, then a transfer to a named person, or a flow you choose. | Fallback & Launch |
| Store events | Order, shipping and cart events from Shopify arrive through Shopify Flow or webhooks and can save values on the contact. | Webhooks, set up as in the Shopify guide |
| Review | Questions the agent could not answer, so you can add the answer. | AI Lab → Unanswered |
A first set of files for the knowledge database. A change applies to new conversations only, so update it before a sale, not during one.
Catalogue: each product with what it is for, key details such as ingredients or sizes, the price and the link.
Size and use guides: the charts and notes the team already sends by hand.
Delivery, COD and return policies: timelines as ranges, the COD limit and the return window.
Approved offers: only offers you want quoted as written. Anything else goes to a person.
Instructions with actions
An instruction can carry an action from the menu below. That is how the return chat in case 3 is tagged and how the agent keeps the budget a shopper mentioned. Each instruction is limited to 500 characters.
Boundaries and the handoff to your team
Boundaries are where a store agent earns trust. Each card in the demo agent’s list names something a shopper may push for — a date, a discount, a refund — and tells the agent not to invent or promise it.
When the agent still cannot help, the Fallback & Launch step takes over: after the retry limit, the chat is transferred to a person you name.
What the agent knows about an order
WABB’s agent has no action that queries Shopify live, and it cannot refund, edit an order or take payment. It can use what you have already stored. Shopify sends an event such as “shipped” through Shopify Flow or a webhook, and a flow can save the tracking link on the contact.
In an instruction, Get Variable Value lists system fields such as name, phone and e-mail, plus your user variables — the fields stored on a contact — so the value the flow saved can be read. Test with a test order that it arrives, and add a boundary for the empty case: no stored value means a handoff, never a guessed date. The Shopify setup guide explains both methods, and the integration page lists what is connected.
What the agent may send, and when
Meta allows messages only to people who have opted in, and a business can start a conversation only with an approved template. In Meta’s own example the cart reminder is a marketing template and the order confirmation a utility one. The agent’s answers are free-form, so they fit the 24-hour window a shopper’s message opens; outside it, use a template message. For routine requests such as “Track my order”, put a fixed menu of up to three buttons from the chatbot builder in front of the agent.
Sources: WhatsApp Business Messaging Policy (23 September 2026) and Pricing on the WhatsApp Business Platform (30 September 2026), checked 14 October 2026.
Business Impact: What You Are Aiming For
Results depend on your products, your traffic and how well the agent is set up, so there are no percentages here. These are the changes a store should look for:
Fewer parcel chats by hand
Where-is-my-order messages with a stored tracking link are answered without a person opening the order, and people handle the exceptions.
Answers while the shopper is deciding
Size, delivery and COD questions get a reply in the moment instead of the next morning.
Cleaner returns
Support receives requests with the order number, reason and preference already collected.
Cart reminders that can answer
The shopper’s reply to a reminder gets a real answer and a way back to checkout.
A clearer view of what shoppers ask
The Unanswered list shows which questions your product pages and policies do not yet answer.
How to tell it is working
Write down your own numbers for a normal month, then track four things every week. The support automation use case covers the team inbox side.
Shopper messages answered within a minute, on sale days too
Order-status chats closed without a person opening the order
Return requests that reached support with order number and reason
Unanswered questions per week, and how many you closed
The agent answers first. Your team handles the exceptions.
FAQ
Common Questions
Can an AI Agent check the order status in my store?
Not by querying the store live: WABB’s agent has no action that calls Shopify. It can share what a Shopify event has already saved for the customer, such as a tracking link, and it hands the chat over when nothing is stored. Test that the value you save reaches the agent before you rely on it. See How It Works in WABB.
Can it recover abandoned carts?
It can handle the conversation after a reminder. The reminder itself is a template message sent by an automation, needs the shopper’s opt-in and is a marketing message in Meta’s pricing. Shopify Flow, which sends the cart event, is not on the Basic plan. See Four Use Cases: From the First Question to the Handover and the cart recovery use case.
Can the agent take payments or place orders?
No. It answers questions and sends the product or checkout link; payment happens in your store. It is also told never to ask for card details in the chat. See How It Works in WABB.
Does it process refunds and returns?
It explains the policy, collects the order number and the reason, tags the chat and passes it to your team. The decision and the refund stay with people. See Four Use Cases: From the First Question to the Handover.
Does it only work with Shopify?
The documented path here is Shopify, through Shopify Flow or webhooks. For another platform, check whether it can send events to a webhook and ask the WABB team what fits. See How It Works in WABB and the Shopify integration.
Do I pay for every message the agent sends?
Meta charges per delivered message by category and market, and does not charge for messages from users. Replies within the open 24-hour window and the free entry point window follow the rules on Meta’s pricing page, and the agent also uses AI credits in WABB. See How It Works in WABB and the plans.