Skip to main content

Sales

Agentic commerce on WhatsApp: how to let an AI agent sell for you, and keep the customer

Shoppers now start their research with AI assistants, and IDC expects most customer interactions to run through agents within a few years. Here is what agentic commerce means for a business that sells in chat, why messaging keeps both the convenience and the customer data, and how to set up an AI agent that sells on WhatsApp, with a Gulf example.

MonoChat Team Updated: 8 min read

On this page
  1. What the research says about where shopping is going
  2. Three places an AI agent can sell
  3. What a selling AI agent needs
  4. How to set it up in MonoChat
  5. A Gulf example: from question to paid order
  6. What to measure
  7. Rules to respect on WhatsApp
  8. Meta’s own agent, or your own?
  9. Start small
  10. Sources

Agentic commerce is selling where an AI agent does real work in the purchase: it understands what the customer wants, searches your products, compares them, answers the questions that block the decision and moves the order to payment. The question for a business is no longer whether shoppers will use AI agents, but where those agents will talk to them, and who keeps the conversation when they do. For a business that already sells in chat, the most practical answer is an AI agent that sells on its own WhatsApp number, grounded in its own catalog, paid through its own payment account, with a person one message away.

This guide explains what agentic commerce means in practice, why messaging is the channel that keeps both the convenience and the customer, and how to set up a selling AI agent on WhatsApp in MonoChat, with an example from a shop in the Gulf.

What the research says about where shopping is going

In a guest article on the WhatsApp Business blog, published on 12 August 2026 and sponsored by Meta, IDC research director Roger Beharry Lall cites IDC’s own figures:

  • IDC predicts that by 2029, 64% of customer interactions will run through agents and answer engines rather than a brand’s own site (IDC FutureScape, October 2025).
  • In an IDC consumer survey (June 2026), 63% of consumers said they begin product and service research with an AI assistant rather than a search engine, and 87% said AI plays a role in their purchase decisions.
  • IDC expects 27% of brands to launch “concierge” agents on their own properties by the end of 2026.

The article’s argument is a trade-off every seller will recognise. General AI assistants are convenient for the shopper, but the transaction and its data stay inside the assistant’s platform. A brand’s website keeps the context and the data, but has limited reach and more friction. Messaging channels, the article argues, can offer both: the shopper’s convenience and the business’s context. The numbers are IDC’s and the article is sponsored by Meta, so read them as one research firm’s forecast, not as a measured market. The direction they point in is still useful.

Three places an AI agent can sell

1. Someone else’s assistant. ChatGPT, Gemini, Claude or Perplexity recommends products from many stores. You can make your catalog easier to recommend (our guide to how AI shopping assistants choose products covers it), but you do not own the conversation, and you rarely learn what the shopper asked.

2. Your website. A chat widget on your store sees the cart and the page, and the data is yours. It only reaches people who already came to the site.

3. Your messaging channels. WhatsApp, Instagram DMs, Messenger, TikTok and web chat. The customer starts or continues the conversation in an app they already use, the agent has your catalog and order data through tools, and every message, order and payment stays in your inbox. When the agent cannot help, a person on your team takes over in the same thread.

The third is the one most businesses can set up this week, and it is the subject of the rest of this guide.

What a selling AI agent needs

An AI agent that sells is a language model with a job description, tools and limits. Without tools it can only talk; with the wrong tools it can promise things you cannot deliver. Five pieces matter:

  1. Catalog tools. Search products, read variants, prices and stock from your real data, every time. The agent recommends only what a lookup returned.
  2. Knowledge. Delivery areas and times, return rules, size guides, warranty terms: the answers that turn “maybe” into “yes”.
  3. A way to close. A checkout link with the items already in the cart, or a payment link for the exact amount from your own payment provider.
  4. Memory of the customer. Earlier orders, tags and profile fields, so a returning customer is not treated as a stranger.
  5. A handover rule. When the order is large or unusual, when the customer is unhappy, or when they ask for a person, the agent stops and your team continues.

How to set it up in MonoChat

MonoChat’s AI assistants answer on WhatsApp, Instagram, Facebook Messenger, TikTok, Telegram and web chat from one inbox, with the model you choose. Here is the order we recommend.

Step 1: Connect the channel. Connect your WhatsApp number through the Cloud API, and the other channels you sell on. MonoChat is an official Meta Business Partner; your WhatsApp Business account stays yours and Meta bills message fees to you directly.

Step 2: Give the agent your catalog. Two routes:

  • With Shop Assistant from the MonoChat Store, customers ask for a product, size or color and get product cards with photo, price and Buy and Details buttons. Buy opens your shop’s own checkout with the items already in the cart. It connects to Shopify, or to your own platform through its HTTP API or a CSV file, and also tracks orders, handles return requests and recovers abandoned carts.
  • With AI function tools, the agent calls your own systems, for example to search inventory or fetch an order, with schemas, permissions and logs. In a flow, an AI Agent node can also use the tools of MCP servers you connect to MonoChat.

Step 3: Add the knowledge. Put your delivery, returns and warranty answers in the AI knowledge base, so the agent answers from your text, not from guesses.

Step 4: Connect a payment system. Under Settings › Payment › Payment Systems, connect the provider you already use. MonoChat’s payment integrations include Stripe and PayPal, Checkout.com, and in the Gulf N-Genius, Telr, PayTabs, Tap, HyperPay, Paymob, Ziina, Mamo, MyFatoorah, Tabby and Tamara. In a flow, the Send Payment Request node creates the payment link for the exact amount and sends it in the chat; the New Payment Received event continues the flow when the provider confirms the payment, for example with the order confirmation.

Step 5: Write the job description. Tell the agent what it sells, how it talks, what it must look up before it answers, and when it must hand over. Keep it short and concrete:

You sell for Al Noor Home, a homeware shop in Dubai.
Answer in the customer's language (Arabic or English).
Recommend only products returned by the product search tool, with their current price and stock.
Ask at most two questions before recommending.
For orders above AED 5,000, delivery outside the UAE, or any complaint, hand over to the sales team.
Never promise a delivery date the delivery tool did not return.

Step 6: Test, then go live. Run the conversations your customers really have, including the awkward ones: a product you do not stock, a discount request, a customer who writes in two languages. Fix the instructions and the tools, not the individual answers. Our guide to testing an AI agent before launch has a checklist.

A Gulf example: from question to paid order

A shopper in Riyadh writes to a Dubai homeware shop that ships across the GCC:

Customer: مرحبا، عندكم طقم فناجين قهوة عربية لـ 12 شخص؟
Agent: أهلاً وسهلاً! عندنا طقمان متوفران الآن:
1) طقم «نخيل» بورسلين أبيض بحافة ذهبية، 12 فنجاناً مع دلة، 690 ريالاً
2) طقم «رمال» خزف مطفي، 12 فنجاناً، 420 ريالاً
هل تفضّل الطقم مع الدلة؟
Customer: The first one. Can I pay in instalments?
Agent: Yes. The "Nakheel" set with the dallah is SAR 690, and delivery to Riyadh takes 3 to 5 working days. I can send you a Tamara link so you can split the payment. Shall I?
Customer: Yes please.
Agent: Here is your payment link. Once Tamara confirms the payment, you'll get your order confirmation here.

Three things happened that a scripted chatbot cannot do well. The agent answered in the language the customer used, switched when the customer switched, and stayed with the facts its tools returned: the two sets in stock, their prices and the delivery time. It offered instalments because the shop connected Tamara, which takes SAR, AED, BHD, KWD and OMR payments. And it closed the sale with a link from the shop’s own Tamara account, so the payment and the order are matched without anyone checking by hand. The names and prices in this example are illustrative; your agent quotes only your own.

If the customer had asked for 40 sets for a wedding, the instruction above would have sent the conversation to the sales team, with the whole thread in front of them. Our guide to AI-to-human handoff covers the rules.

What to measure

An AI agent that sells should be judged like a salesperson, not like a FAQ:

  • Conversations that reach a recommendation, and how many questions it took.
  • Conversations that reach a checkout or payment link, and the share that pay.
  • Revenue from chat, by channel and by campaign.
  • Handover rate, and why: missing product data, policy questions, complaints.
  • Questions the agent could not answer, which are your next catalog and knowledge-base fixes.

Rules to respect on WhatsApp

Meta lets a business send free-form messages only within 24 hours of the customer’s last message; after that, only approved templates. A selling agent therefore works best on conversations the customer starts: from your site, your ads, your social profiles or a QR code in the shop. Follow-ups such as an order update or a payment reminder go out as templates. Meta’s policy for the WhatsApp Business Platform also bars general-purpose AI assistants on WhatsApp since 15 January 2026; an agent that sells your own products and supports your own customers is the kind of business use the platform is built for. Our guide to the WhatsApp AI chatbot policy explains the difference.

Meta’s own agent, or your own?

Meta has introduced its own AI agent for businesses on WhatsApp. If you are weighing it against an agent you run in your own inbox with your own tools and payment providers, our comparison of Meta Business AI and an AI agent platform sets out the questions to ask.

Start small

Pick one product line and one channel. Connect the catalog, one payment provider and the handover rule, and let the agent handle the first message of every conversation for two weeks while your team watches. Then widen it. Start on the free Basic plan at app.monochat.ai, and compare plans on the pricing page: Growth is $120 a month.

Sources

FAQ

Frequently asked questions

What is agentic commerce?

Agentic commerce is buying and selling where an AI agent does part of the work a shopper or a sales assistant used to do: it understands what the customer wants, searches products, compares them, answers questions and moves the purchase to payment, using tools rather than only generating text.

Why run a selling AI agent on WhatsApp rather than only on my website?

Because the conversation stays with you. A general AI assistant can recommend your product, but the conversation and its data stay on that assistant's platform. A website keeps the data but reaches only visitors who come to it. A messaging channel such as WhatsApp reaches the customer where they already talk to you, and every message, order and payment stays in your own inbox and systems.

Can an AI agent in MonoChat take the payment?

It can send a payment link from your own provider account. A flow's Send Payment Request node creates the link with a payment system you connected in MonoChat, such as Stripe, PayPal, Checkout.com, Tabby or Tamara, and the New Payment Received event moves the conversation on when the provider confirms the payment. The money goes to your provider account, never to MonoChat.

How do I stop the agent from inventing products or prices?

Give it tools that look up your real catalog, prices and stock, and instruct it to recommend only what a lookup returned. MonoChat's AI function tools let the agent call your systems with schemas, permissions and logs, and the Shop Assistant app answers from your Shopify store or your own product data.

When should the AI agent hand over to a person?

For large or unusual orders, complaints, negotiations, and whenever the customer asks for a person. In MonoChat the conversation goes to your team in the same inbox with the full history, so the customer does not repeat themselves.

What does it cost to start?

MonoChat's Basic plan is free and Growth is $120 a month. Meta bills WhatsApp message fees directly, with no MonoChat markup, and AI usage is billed as described on the pricing page.

First month of Growth on us

Run WhatsApp, Instagram and Messenger from one inbox

Start free with MonoChat: shared team inbox, AI assistants, templates and campaigns on official Meta APIs.

One code per business • 30 days to redeem • No card needed

$150 to get started.

Claim