AI product recommendation on WhatsApp works when the agent asks two or three short questions about the buyer's need, then suggests one or two SKUs from your own catalogue instead of dumping a price list. For an online store, this turns the usual 'which one is good?' message into a sale without a staff member typing the same advice again.
- A good recommendation agent asks before it suggests: use, budget and one deal-breaker are usually enough.
- It should recommend one main SKU and one alternative, never a wall of links.
- The agent can only recommend well if your catalogue has real attributes, not just names and prices.
- Stock, price and promo changes must come from one source, or the agent will recommend things you cannot ship.
- Hand over to a person for bulk orders, custom requests and anything the catalogue does not answer.
- Check results by reading the chats where the buyer ignored the suggestion.
What is AI product recommendation on WhatsApp?
AI product recommendation on WhatsApp is an AI agent that reads what a buyer says, asks the missing questions, and replies with the SKU from your catalogue that fits. It does the job of the shop assistant who says 'what are you using it for?' before pointing at a shelf.
Most online stores already get this message many times a day: 'Which one better?', 'Got something cheaper?', 'Suitable for my mum?'. A person can answer each one well. The trouble is volume. At 10pm, after a Facebook ad has run all day, the same three questions arrive twenty times, and replies get shorter and slower.
In an AI Agent Office, this is the job of a dedicated sales-floor employee. It does not greet, book or chase payments. It only recommends, and it passes the buyer on once a choice is made.
Why should the AI ask questions before it recommends?
The AI should ask first because a recommendation made without knowing the need is a guess, and buyers can tell. A price list sent in reply to 'which one good?' puts the work back on the buyer, who then often leaves.
The rule that works in practice is: ask no more than three questions, one at a time, and make each one easy to answer. A buyer on a phone will tap out a number or a word, but will not write a paragraph.
Hi, which air fryer is good ah
Happy to help. How many people do you usually cook for?
3 of us, sometimes 4
Got it. Do you want something compact for the counter, or a bigger one that can do a whole chicken?
bigger one, my kitchen ok
Then I suggest the 6L dual-basket model. It fits a family of 4 and you can cook two items at once. If you want to spend less, the 5L single-basket is the next step down, but you cook one item at a time. Which do you prefer?
Illustrative example.
Notice what the agent did not do. It did not list eleven models. It named one main choice, gave the reason in the buyer's own words (family of 4, big kitchen) and offered one alternative with an honest trade-off.
Which questions should the agent ask for each type of product?
The questions depend on what decides the purchase in that category, so list them per category before you switch the agent on. The table shows how different they are.
| Category | Question 1 | Question 2 | Deal-breaker to check |
|---|---|---|---|
| Kitchen appliances | How many people at home? | Counter space or storage? | Voltage and plug type, warranty |
| School bags and stationery | Which school year or age? | Trolley or backpack? | Weight limit, school colour rules |
| Sneakers and apparel | Usage: sport, work or casual? | Which size do you usually wear? | Size run, exchange rules |
| Baby products | Baby's age or weight? | Budget range? | Safety rating, size limits |
| Gifts | Who is it for and what occasion? | Budget range? | Delivery date, gift wrap |
Store owners often skip the last column. It matters most. A recommendation that ignores the deal-breaker creates an exchange, a refund request and an unhappy review.
What does the catalogue need before an AI can recommend well?
The catalogue needs attributes that a buyer actually cares about, written in plain words, not just a product name, a photo and a price. The AI cannot recommend 'good for small kitchens' if no row says how big the item is.
- Plain attributes: size, capacity, material, who it suits, who it does not suit.
- A reason to choose it: one honest sentence for each SKU, such as 'lighter than the 6L model, but one basket only'.
- Variants kept together: colour and size under one parent product, so the agent does not recommend a size that is sold out.
- Status: in stock, low stock, pre-order, discontinued.
WhatsApp has its own catalogue feature, and many stores also keep products on Shopee, a website or Shopify. Whichever you use, choose one place as the master. If prices are edited in three places, the AI will quote the old one in one of them. ChatsHero, a Malaysian AI agent platform, can work with a Shopify store and with Google Sheets, so a shop with a simple spreadsheet can start there.
Want to see how this looks with your own business? WhatsApp us and chat with the ChatsHero AI.
What do Malaysian shoppers ask that a generic AI misses?
Malaysian buyers ask local questions that a generic assistant does not know, so the agent must be taught them from your own chats. Four come up again and again.
- 'Ada COD?' The agent must know your cash-on-delivery rules by area, and when you ask for a deposit first. See our guide to payment proof and COD.
- West or East Malaysia delivery. Shipping time and fee differ, and a recommendation that ignores a Sabah address wastes a day.
- Rojak language. 'Yang mana lagi worth?' and 'cheap cheap one got?' must be understood. Our post on how AI handles Bahasa rojak explains the idea. A buyer may ask for a cadangan produk (product suggestion) in Malay and switch to English one line later.
- Festive timing. Before Raya or CNY, buyers want to know the last order date for delivery, not just which item to buy.
What should the AI never do when recommending?
The AI should never invent a feature, promise a result the product does not claim, or push a SKU that is out of stock. Honest limits are what make a buyer trust the next suggestion.
- No made-up specifications. If the catalogue does not say whether a bag is waterproof, the agent says it will check with the team.
- No health or performance claims you cannot back up. For regulated products, read the rules first. Our post on chatbots in the supplement industry is a starting point.
- No pressure. One alternative is helpful. A fourth reminder that 'only 2 left' is not.
- No guessing size or fit when the buyer has not given measurements. Ask, or hand over.
The agent should also hand over cleanly. Bulk quantities, custom colours, corporate orders and complaints belong to a person, and the buyer should be told so in one line.
How do you know the recommendations are working?
You know the recommendations work when buyers pick the suggested SKU, ask fewer 'which one' follow-ups, and return fewer wrong-size or wrong-model items. Review it weekly with a short routine.
- Open ten chats where the buyer got a recommendation and did not buy.
- Mark the reason: price, stock, wrong product, no reply from the buyer, or a question the agent could not answer.
- Fix the cause in the catalogue first, then in the agent's instructions.
- Repeat next week with a fresh ten.
In our own sales data, 60 of 269 tracked ad leads stopped after the first message. A buyer who gets a useful question as the first reply is more likely to answer than one who gets a link and a price list. Treat that first reply as part of your recommendation work.
How does a ChatsHero AI Agent Office handle product recommendation?
In a ChatsHero AI Agent Office, product recommendation is a sales employee with one job: ask the buyer's need, match it to your catalogue and price list, and pass the buyer on. The receptionist role greets and sorts the enquiry first, the recommendation role suggests, and the follow-up role checks back if the buyer goes quiet.
It works from your own SOP, price list and FAQ, replies on WhatsApp, Messenger and Instagram, tags the lead from the chat, and hands over to staff for bulk or custom orders. The fastest way to see it is a free 30-minute Zoom demo using your own top ten buyer questions.
If you are still deciding between a plain store chatbot and an agent team, read our guide to the WhatsApp chatbot for Malaysian online stores, and the pillar explainer on what an AI Agent Office is.
Frequently Asked Questions
Can an AI recommend products on WhatsApp without a website?
Yes. The agent only needs a catalogue it can read, such as a spreadsheet or a product list with plain attributes, price and stock status. A website helps but is not required.
How many questions should the AI ask before recommending?
Usually two or three, one at a time. More than that feels like a form and buyers stop replying. Ask only what changes the recommendation.
Will the AI recommend items that are out of stock?
Not if stock status comes from one master source that the agent reads. If stock is edited in several places, mistakes happen, so keep one master list.
Can the AI recommend in Bahasa Malaysia, Chinese and English?
Yes, AI agents can reply in the buyer's language, including mixed Manglish. Test it with your real chats before going live.
What should still go to a human?
Bulk and custom orders, complaints, refund disputes and any question the catalogue cannot answer. The agent should tell the buyer plainly that a person will continue.
Does ChatsHero offer a free trial?
ChatsHero has no free trial because each setup is customised, but there is a free 30-minute Zoom demo.