Conversational commerce: how AI chats turn into sales

It's 9 PM and a message lands in your WhatsApp inbox: "Do you have this jacket in medium?" One possible reply says yes, it's in stock, and the chat ends there. Another says yes, sends the medium with its price, and offers a payment link.

Only the second reply is on its way to a sale.

Brands get messages like this every day, and plenty still send the first kind of reply. The shopper then has to leave the chat, find the product again, and check out alone.

That's where ready buyers drift off. Conversational commerce closes that gap by finishing the sale inside the conversation the shopper started. Below, we'll cover how a chat becomes a sale step by step, which guardrails an AI agent needs, where AI should hand over to your team, and how to measure both the agent and what actually closes.

What is conversational commerce?

Conversational commerce is a way of selling where the whole purchase happens inside a messaging conversation. A shopper asks a question on WhatsApp, Instagram, Messenger, or live chat, gets a product recommendation, and pays from the same thread. An AI agent handles the routine replies, and your team steps in when a person is needed.

It sits alongside your online store rather than replacing it. Your catalog and checkout still exist, and the conversation becomes one more place people can buy.

Conversational commerce describes the channel: a shopper and a brand interact in a messaging conversation. Agentic commerce describes what an AI agent can do, such as make decisions and take permitted actions across connected tools. A scripted chatbot follows set prompts; an agent can act within clear guardrails. The two can overlap: an AI agent can power a conversational commerce experience.

For the technology behind the replies, see our guide to conversational AI for business.

Why sell in chat instead of sending shoppers to checkout?

Selling in chat reaches the shopper at the moment of intent, in the app they already chose. Every step between "I want this" and "I've paid" is a chance to lose them, and a website checkout adds several: finding the product again, creating an account, typing in card details, and remembering the discount code.

Those steps add up. Baymard Institute's benchmark of 50 studies puts average online cart abandonment at 70.22%, a long-run figure built from studies going back to 2006.

Shoppers also lean toward brands they can message. In a 2025 Kantar survey of 11,056 adults across 22 markets, commissioned by Meta, 72.4% said they're more likely to buy from a brand that offers messaging.

Selling in the thread removes most of the steps that cost you the sale. The product, the answer, and the payment link arrive in one place, minutes apart.

How does a chat become a sale?

A chat becomes a sale in 8 steps: capture, instant reply, needs discovery, qualification, recommendation, routing to a person, follow-up, and conversion. Connected AI agents can also automate routine commerce tasks, such as checking order status or guiding shoppers through return questions. Payment, sensitive account changes, and other high-stakes requests should follow clear handoff rules. AI and automation handle fast, repetitive tasks at any hour, while your team steps in when judgment or personal support is needed. The table shows who does what at each step.

How do you capture shoppers and reply instantly?

Capture and instant reply decide whether the conversation happens at all. Every message from an ad, a website chat, or a QR code in store should get an answer within seconds, day or night, because a shopper who waits until morning has often bought elsewhere. An AI agent handles this first touch on every channel at once.

For website chats, label the page, campaign, or entry point that brought each shopper in. That context helps your team understand what the shopper saw and compare which sources lead to orders.

How does an AI agent work out what a shopper needs?

The agent finds out what the shopper needs by asking the questions a good salesperson would: what it's for, which size, what budget, and how soon. It also works out how ready that shopper is to buy, and passes that along with the conversation when your team takes over.

Pro Tip: Write down the 5 questions your best salesperson always asks before recommending anything, and give them to the agent as its discovery script.

How should the agent recommend products?

Recommendations should come from live product data, so the agent quotes the real price, size, and availability rather than an answer someone typed into an FAQ months ago. If you sell on Shopify, our guide to the Shopify AI agent covers the setup.

How do follow-up and payment close the sale?

Follow-up brings back shoppers who went quiet, with an automated reminder about the item they looked at or the cart they left. Conversion is the final step, where the shopper pays from the thread. Your team builds the cart and sends a payment link, so the buyer never has to hunt for the product on your website.

What does an AI agent need to sell, not just answer?

An AI sales agent needs three things beyond fluent replies before it can sell: reliable product and order data, a way for shoppers to pay from the conversation, and clear rules for when to hand over to a person. Connect the agent to your live catalog so it can quote accurate prices, sizes, and stock, then define which actions it can take and which require approval. Test its answers against common shopping questions before launch. When a shopper is ready to buy, send a payment link to checkout in the conversation; route payment issues, sensitive requests, and other exceptions to your team. Without these connections and rules, an agent can describe a product well but still send the buyer elsewhere to finish.

• Live product data. Connect your catalog inventory, and order systems through supported integrations or APIs so the agent can quote current prices, sizes, and stock. Check how often data syncs, set permissions for what the agent can read or change, and test common product and order questions before launch. If your sales team uses a CRM, connect it to give them customer and conversation context during handoff.

• Payment in the thread. Your team sends a payment link that opens a checkout with the cart already filled in, so the shopper pays in the same place they asked.

• Handoff rules. Decide which topics, signals, or completed tasks end the AI session and bring in your team, with the full conversation attached so the shopper doesn't repeat themselves.

What guardrails should AI agents have?

AI agents should have guardrails on what they can say, what they can do on their own, and when they must hand over to a person. Set these before launch, so the agent stays accurate and never takes an action your team hasn't approved.

• Approved sources only. The agent answers from your live catalog, policies, and help content, and says it doesn't know rather than guessing a price, size, or delivery date.

• Clear action permissions. List what it can do alone, like recommending products, checking order status, or booking a slot, and what needs a person's approval, like discounts, refunds, and order changes.

• Safe payment and personal data. Shoppers pay through a secure payment link, never by typing card details into the chat, and account changes go to your team.

• Escalation triggers. The agent hands over when it's unsure, the shopper is upset, the request is unusual, or the shopper asks for a person.

• Visible conversations. Every chat sits in the shared inbox, so your team can see what the agent said and step in at any point.

How is a commerce agent different from a support chatbot?

A commerce agent and a support chatbot can use the same technology and messaging channels, but they have different goals. A support chatbot answers questions and resolves routine issues. A commerce agent helps move a shopper toward a purchase by using connected product data, recommending items, and handing the conversation to a person when needed. In this workflow, your team confirms the cart and sends the payment link; the agent does not complete the checkout on its own.

An AI sales agent can also support B2B sales, a separate workflow from retail shopping. When connected to approved tools, it may research prospects, qualify a lead, personalize outreach, follow up, schedule a meeting, and update CRM records. Set permissions and review rules for actions such as outreach and CRM changes, and make sure sales reps receive the conversation context when they take over.

When should a person take over the chat?

A person should take over when money changes hands, sensitive data is involved, a customer is upset, or an order is complex or high value. AI handles volume and speed well, but those moments carry real cost if a reply is even slightly wrong, and the shopper should never have to repeat what they already said.

Set the handover around these moments:

• Payment, account changes, and personal data. A person confirms the cart, sends the payment link, and handles anything sensitive.

• Upset customers. Make negative sentiment a handover trigger, so a person answers the frustrated shopper.

• Big or complex orders. High-value buyers get a person who can negotiate, bundle, or advise.

How do you measure conversational commerce?

Measure selling in chat by what the conversation closes: orders placed, bookings made, carts recovered, and revenue from payment links. A high count of chats that ended without a person says nothing about whether the channel sells.

Track these, most important first:

• Revenue from chat. Sales completed through payment links sent in conversations.

• Orders and bookings started in chat. The purchases and appointments a conversation led to.

• Recovered carts. Carts completed after a follow-up message.

• First reply time. How fast the AI agent and your team each answer.

• Handover rate. How often the AI session ends on a condition you set, such as passing the chat to a person.

Which KPIs measure AI agent performance?

The KPIs that measure AI agent performance are chat-to-order conversion, answer accuracy, resolution rate, handover quality, first reply time, and customer satisfaction. Read them together. An agent that replies in seconds but quotes the wrong price, or hands over without context, loses sales that a speed figure alone won't show.

Compare these with chats your team handles alone, so you can see where the agent helps and where a person still does better.

How SleekFlow turns conversations into sales

SleekFlow puts the AI agent and your team in one inbox, so a conversation can go from first question to paid order without anyone switching tools.

AgentFlow's Inbound AI Agent answers on WhatsApp, Instagram, Messenger, live chat, and more. It works out what each shopper wants and how ready they are to buy, then hands over based on topic, sentiment, or a finished task. For services, it can check availability, propose times, and confirm a booking that syncs to your calendar.

With the Shopify integration, the agent recommends products using live descriptions, prices, and availability from your store. Your team sees order history and abandoned carts beside each chat.

When the shopper is ready, your team builds the cart and sends a payment link through Shopify or Stripe, in the markets where payment links are available. A dashboard shows the sales attributed to links sent in chat.

NNIO, a Singapore electronics and home appliance brand, had enquiries split across four separate teams and campaign spikes that ran past midnight. It set up AgentFlow AI agents trained on its product information, with Flow Builder routing chats to available team members.

Results:

• 20% increase in completed checkouts

• 5% higher conversion rate

• 40% faster response time

Animes-Pro, a Hong Kong toy and anime retailer with 14 branches, had one staff member handling 80 to 200 inquiries a day across WhatsApp, Facebook, and Instagram. AgentFlow now reads each chat's intent and sends buying questions to sales, while staff check purchase history through Shopify inside the chat.

Results:

• Inquiry handling time cut by 30% to 50%

• E-commerce sales growing at least 10% a year

Where to start selling in chat

Start with the questions your shoppers already send you most, and make sure each one can end in a purchase. Connect your store so the agent quotes live product data, and set the moments when your team takes over. Then track revenue from chat from the first week. See how AgentFlow turns a real conversation into an order.

Want to outcompete your peers with SleekFlow's help?

Book your personalised demo with SleekFlow today and unlock the potential of seamless communication

Book a Demo

View Pricing

Frequently asked questions

How is conversational commerce different from e-commerce?

E-commerce sends shoppers through a website: browse, add to cart, and check out alone. Selling in chat runs the same purchase through a messaging conversation, where an AI agent or a person answers questions and recommends products, and your team sends a payment link in the thread. You can run both, with the chat catching shoppers who have a question before they buy.

Which channels work best for selling in chat?

The best channels are the ones your customers already message you on. Common choices are WhatsApp, Instagram, Facebook Messenger, and live chat on the website. The channel matters less than what sits behind it: an agent with live product data, a way to pay in the thread, and a team that can step in.

Can customers pay from a WhatsApp chat?

Yes, where payment links are available. Your team sends a payment link in the WhatsApp conversation, and the shopper opens a secure checkout with the cart already filled in, without searching for the product on your website. Availability depends on your payment provider and your country, so check which markets support payment links before you plan around them.

Is conversational commerce only for large brands?

No. A small team often gains the most, because one or two people can't answer every message at 9 PM. An AI agent covers the first reply and the routine questions, and the team spends its time on the shoppers who are ready to buy. Starting with your most common questions keeps the setup small.