What shoppers ask
| Before they buy | After they buy |
|---|---|
| Do you have this in size M? | When will my order arrive? |
| Which moisturiser is good for dry skin? | Can I change my delivery address? |
| Do you ship to Malaysia, and how much? | Can I return a sale item? |
| Is it back in stock soon? | My parcel arrived damaged |
The first column wins sales; the second keeps customers. Both have answers in information you already have: your catalogue, your stock and order data, and your policies.
Before they buy: products that fit
- Describe each product the way a shopper would ask about it: who it's for, what it's good at, sizes and materials.
- Recommend only close matches, with a short reason why, and link to the product page to buy.
- Use the catalogue's own name, price and link, and never invent one: if the catalogue and the website disagree, fix one of them.
- Don't push a product on every reply; suggest one when it clearly helps.
Stock and prices that change
A catalogue is right for descriptions; stock changes by the hour. If your stock lives in a database, a read-only connection lets the agent check it at the moment of the question, so “is it in stock?” is answered from now, not from last week's import.
Use a database login that can only read, and choose exactly which tables the agent may look at.
After they buy: “where's my order?”
The most common question after a purchase can be answered from the order itself, if it's done safely:
- Look an order up only by something the customer typed, such as their order number.
- Show its status, items and delivery, never who it belongs to or anyone else's orders.
- Never list customers, totals or samples of records to someone who isn't signed in.
- Put these limits in code, not in the AI's instructions, so no clever question gets around them.
Returns and problems
Return windows, sale items and delivery times are policy questions, so the agent can answer them from your policy pages. A damaged parcel or a refund request is different: it needs a decision.
- Step 1
The agent gathers the details
Order number, what happened, a photo if there is one.
- Step 2
It opens a case
With everything a person needs to decide, instead of the customer explaining it all again.
- Step 3
A person decides
And approves the reply before the customer hears back.
An AI agent shouldn't issue refunds or promise them on its own. Cases and approvals
Selling on WhatsApp: the rules
- Restricted goods: WhatsApp's Business Messaging Policy prohibits buying, selling or promoting goods such as alcohol and tobacco, drugs, medical and healthcare products, weapons, live animals, adult products and real-money gambling.
- Exceptions: alcohol and over-the-counter medicines are allowed in some listed countries, on the WhatsApp Business Platform only, and never to people under 18.
- Order updates, such as shipping and delivery, go out as utility templates when the customer hasn't written in the last 24 hours.
- Offers, new arrivals and abandoned-cart nudges are marketing templates, sent only to customers who opted in to hear from you.
WhatsApp Business Messaging Policy · Meta: getting opt-in · WhatsApp template messages
Setting it up in HeyKiko
- Step 1
Add your catalogue
Each product with a name, a description, a price, a link, a photo and a category, or import a CSV.
- Step 2
Teach it your policies
Delivery, returns and FAQ pages from your website, or as documents.
- Step 3
Connect your database, if you have one
A read-only MySQL or MariaDB login and the tables for stock and orders.
- Step 4
Turn on cases
For damaged orders and refund requests, with a person approving each reply.
HeyKiko doesn't take payments, keep stock counts, or sync with Shopify or WooCommerce: customers buy on your own site through each product's link, and live stock comes from your database.
Sources
Checked on 5 November 2026. We make HeyKiko, which this guide uses as its example.
