---
title: "Conversational AI in eCommerce: How brands sell and serve through two-way messaging"
description: "Two-way AI agents on WhatsApp, SMS, and in-app answer questions and place orders. See how conversational AI in eCommerce works across the buying journey."
canonical_url: "https://www.braze.com/resources/articles/conversational-ai-ecommerce"
language: "en-US"
datePublished: "2026-10-28T18:59:00.000Z"
dateModified: "2026-10-28T18:59:00.000Z"
publisher: "Braze"
authors:
  - "Team Braze"
---

# Conversational AI in eCommerce: How brands sell and serve through two-way messaging

Conversational AI in eCommerce is software that understands natural-language messages from shoppers and responds, recommends, or completes a task in a two-way conversation, across the messaging channels a brand owns.

Conversational AI only works when it is connected to real customer and catalog data and can hand off to a person when needed. Without that, it's just a scripted bot with better grammar.

The use cases below cover the buying journey from discovery to reorder, the owned channels each one runs on, and what has to be connected behind them.

**TL;DR**

- Conversational AI in eCommerce is software that understands natural-language messages from shoppers across the customer lifecycle and responds, recommends, or completes a task in a two-way conversation, across the messaging channels a brand owns.
- The difference from a scripted bot is architectural: AI interprets intent from free text, remembers information across a whole conversation, and can act on what it finds.
- Owned channels like WhatsApp Business messaging, SMS, RCS, and in-app are where conversations write back to your own customer profiles, so what a customer says in one place informs everything that follows.
- Four things decide whether conversational AI works: a unified customer profile, live catalog and inventory data, clear guardrails, and a clean handoff to a person.

## What is conversational AI in eCommerce?

Conversational [AI in eCommerce](https://www.braze.com/resources/articles/ai-in-ecommerce) is software that understands natural-language messages from shoppers and responds, recommends, or completes a task in a two-way conversation, across the messaging channels a brand owns.

It's the technology underneath conversational commerce, and what it does differently from the broader practice of [conversational marketing](https://www.braze.com/resources/articles/conversational-marketing), is interpret consumer intent from free text. A shopper doesn't have to use the right keyword or pick from a menu for the message to be understood.

Conversational AI also remembers what's already been said. For example, if a shopper mentions a size early in a thread, that size still applies when they ask about delivery a few messages later.

And it completes the task as well as describing it, like looking up an order, changing a size, or placing a reorder.

Conversational AI in [eCommerce and retail](https://www.braze.com/solutions/retail-and-ecommerce) can run across the entire buyer journey. The information it learns is first-party customer data, being given by the shopper with permission. Before purchase, a shopper describes what they want and gets relevant options back. At purchase, questions about sizing or stock are answered before they become an abandoned cart. After purchase, order status, delivery changes, and returns run through the same conversation. And when it's time to buy again, a reorder prompt becomes an exchange that ends in a completed order.

## Conversational AI vs. scripted bots: what changed

A scripted bot runs on a decision tree built in advance. The team maps out the questions the brand expects to get, writes a response for each one, and lists the keywords that should trigger them. So "Where's my order" reaches the tracking reply because that exact phrase is on the list. Yet "Has my stuff shipped yet" might not reach anything at all, and when a message falls outside the tree, the shopper gets sent a fallback message that might look something like, "Sorry, I didn't catch that, try rephrasing," or a link to the help center.

Conversational AI agents for eCommerce read the message rather than matching it against a list, so the phrasing matters far less. "I need the next size up" and "can I swap it for a 10" carry the same intent, and the agent is able to understand that and respond accordingly. Generative AI is what makes the reply possible, since the agent composes an answer based on what it understands rather than returning something written in advance. It also remembers what's already been said, so if an ideal size or color was mentioned earlier on, it will remember that later when the shopper asks a related question.

One of the biggest differences is that conversational AI for eCommerce can connect to catalogs and customer data, and then take action on what it finds. It can check whether an item is currently in stock, place the order, take payment, and arrange delivery, all without the shopper having to leave the conversation and go to the website. If a request goes beyond what the agent is permitted to handle, it gets passed to a person with the whole conversation attached, so the shopper doesn't have to explain themselves twice.

All of this relies on the agent having access to that data. Without it, it's just a bot deflecting a question. An agent can resolve a query, and that's where the value is. Turning a query or a support request into a completed purchase shows up in conversion rate and repurchase rate, saves the support team time, and gives the shopper a better customer experience.

## Conversational AI eCommerce use cases across the buying journey

The five use cases below follow a shopper from first question to repeat order. Each one covers what the shopper is trying to do, what the agent needs access to, the owned channel it usually runs on, and the kind of brand it suits.

Four of these are written as scenarios, built from elements of real campaigns from Braze customers, to illustrate how each could play out for an eCommerce brand. The fifth is a named example with a published result.

### Use case #1: Guided discovery and product questions

#### Shopper goal

Plenty of shoppers arrive with an idea rather than a product. They want something for a dinner party, or something warm enough for a trip they're taking, but they don't know which of 400 items on a category page fits. Conversations allow them to describe what they want, in their own words, and keep asking questions to narrow the options.

#### What the AI needs to know

Access to the product catalog data, with enough attribute detail to answer on: ingredients, materials, fit, what an item is actually for. It also needs whatever the brand already knows about the shopper, and past orders tend to be the most useful thing in that profile. A meal delivery service can ask what someone feels like eating, then match the reply against both the current menu and what that customer has ordered before, so the suggestions come back shaped by their taste rather than by what's selling well overall. [BrazeAI™](https://www.braze.com/product/brazeai) does the work of deciding which items to put forward.

#### Channel

[WhatsApp Business messaging](https://www.braze.com/product/whatsapp), where the thread carries images and product cards alongside text and stays open, so a conversation that starts on Tuesday can be picked up on Thursday. In-app messaging does the same job when the shopper is already logged in.

**Best for:** brands with large or varied catalogs, where the hard part of shopping is narrowing down rather than checking out.

### Use case #2: Size, fit, and stock questions before purchase

#### Shopper goal

Someone has found the thing they want and stalled over a detail. Does this run small? Is the 10 back in yet? Will it arrive before the weekend? On a website, answering any of those questions means hunting around for it, either by digging through a size guide, scrolling reviews to see what other people said about the fit, or checking a delivery page that may not be specific to their address. It's enough friction that plenty of shoppers give up and leave the cart where it is.

#### What the AI needs to know

Stock levels for the specific item and size, updated as they change rather than once a day. An agent that tells a shopper something is available when the item sold out that morning does more damage than one that says nothing at all.

It also needs the product details to answer questions about fit, and those differ by category. For clothing, that means measurements and cut. For furniture, it's dimensions. For anything technical, it's what the product works with.

If the brand holds returns data, the agent should have that too. The reasons customers give when they send something back are the clearest signal of how an item really fits, as opposed to how the product page describes it.

#### Channel

SMS and RCS conversations, since these questions tend to arrive at the point of purchase and both channels reach a shopper without requiring an app. RCS carries branded product cards with SMS fallback where the handset doesn't support it. In-app and web messaging suit the same questions asked mid-session, where the agent can see what the shopper is looking at.

**Best for:** brands where fit, sizing, or availability is the main reason a cart gets abandoned, such as apparel and footwear most obviously, but any category with variants that behave differently.

### Use case #3: Order status and delivery changes

#### Shopper goal

A customer wants to know where their order is, or they need to change something about it. That might be checking how far away a delivery is, chasing a parcel that never turned up, moving the delivery to a day they'll be home, or correcting an address they got wrong at checkout. Normally that means finding the tracking page, which tells them the status but won't let them do anything about it, then completing a support form, or calling and being held in a queue. In a conversation, they can ask where the order is and move the delivery to a different day without leaving the thread.

#### What the AI needs to know

The current status of the order, matching whatever the warehouse or carrier is showing. If the agent is working from yesterday's information, it will tell the shopper something that is out of date.

It also needs to know what it's allowed to change. Rescheduling a delivery, redirecting to a neighbor, or updating an address are all routine requests, but each one has a point after which the parcel is with the carrier and the brand can't move it. The agent has to know where that cutoff falls, so it doesn't promise something that can no longer happen.

#### Channel

SMS, mostly. Delivery updates already go out by text, the shopper has the thread on their phone, and replying to it is the shortest path to getting a question answered. RCS does the same with richer formatting where it's supported. In-app messaging works for brands whose customers are already in the app, where the agent can show the order alongside the conversation.

**Best for:** brands shipping physical goods on any scale, and particularly those where delivery timing matters, like groceries, perishables, anything bought for a specific date.

### Use case #4: Reorders and replenishment by text

#### Shopper goal

Buying the same thing again should be the easiest purchase a customer ever makes, and quite often it isn't. A reminder arrives to say they're due for a refill, they tap through to the site, log in, find the product again, and go through checkout for something they’ve already bought several times before. Every one of those steps is a chance to put it off until later, and with consumables, putting it off usually means picking it up somewhere else instead.

Replying to a text is a much shorter route. A straightforward “yes” gets the order placed in one message, and a customer who wants something slightly different, like the same product in a larger size, a different strength, or another flavor, can say so in the same thread and have the agent sort it out from there.

Subscriptions have a similar problem. A notice that the next delivery is coming up is genuinely useful, but only if the customer can act on it there and then. Someone going away for a trip wants to push their order back rather than cancel it. Someone with three unopened boxes in a cupboard wants to skip the next one. Both of those are small adjustments if they can be made in the thread, and both are a common reason people cancel a subscription altogether when the only way to make them is through an account page.

#### What the AI needs to know

Purchase history, so the agent knows what the customer bought, when they bought it, and roughly how often they come back for more. For a subscription, AI also needs the schedule: what is due to go out next, when it is due, and how far that date can reasonably be moved.

It needs the catalog too, in enough detail to work out which product someone means when they describe it loosely rather than by name.

And it needs to be able to place the order or change it, rather than simply getting it ready. A conversation that ends by sending the customer to a checkout page re-adds the step that was causing the problem in the first place.

#### Channel

SMS. Reorder prompts and delivery notices already go out by text, so the thread is somewhere the customer is used to hearing from the brand, and the whole exchange can be done in a few taps without opening anything else.

#### Aeroflow Health: a text that finishes the job

Aeroflow Health is a healthcare-focused consumer goods provider. The brand’s Mom & Baby division helps pregnant and nursing mothers find medical devices, such as breast pumps, that are covered by their insurance.

The team was already running an SMS reordering journey. Customers due for a repurchase received a text asking whether they wanted the same size again, and for anyone happy with that, it worked well. Anyone who wanted something different had to go to the website and start a new purchase. The size details customers sent back came in so many different formats that the team ended up processing those orders manually.

Using AI agents, a customer who wants a different size is now asked what size she would like, in the same thread. The agent reads the reply, which might be something as loose as "next size up please," matches it against Aeroflow's catalog, confirms the product with her, and places the order. As Brian Munn, senior marketing automation manager at Aeroflow Health, explained, the agent could "decipher all this data into one standardized format that we could automate."

That journey produced an [8.8% lift in repurchase conversions over business as usual](https://www.braze.com/customers/aeroflow-health-case-study).

**Best for:** brands selling consumables, subscriptions, or anything bought on a regular cycle, where what the customer wants each time is broadly the same but not always identical.

### Use case #5: Returns and exchanges started in a conversation

#### Shopper goal

Something has arrived and it isn't right. The jacket is too small, or the color looks different in daylight than it did on screen. What the customer wants next is usually simple: send this one back, or swap it for the right one.

Getting there is often the least enjoyable part of buying from a brand. There's a returns portal to find and an order number to dig out, then a list of reasons that don't quite describe the problem, and a label to print at home. For an exchange it's worse, because many brands don't offer one. The customer returns the item, waits for the refund, and buys the right size separately, which means the brand loses the sale and only gets it back if the customer bothers to come back.

In a conversation, the customer says what's wrong and the agent takes it from there, within whatever the brand allows it to do. Some requests will be passed to a person. Some customers will be offered a replacement without needing to return the original. Others will have to send the item back before a refund is issued. Either way, the customer gets an answer in the conversation thread rather than working it out from a help page.

#### What the AI needs to know

What the customer ordered and when, so it can work out which item they mean without asking for an order number they'd have to go and find.

The returns policy as it applies to that specific order, which usually comes down to whether the window is still open and whether the item is eligible at all. Policies vary by product and by market, and an agent that gets this wrong either costs the brand money or tells the customer something that gets reversed later, creating a poor experience.

For an exchange, it also needs access to what’s in stock, so it can check if the replacement is available before offering it.

#### Channel

WhatsApp Business messaging, where the thread can include a return label or a drop-off location alongside the conversation, and where the customer can send a photo if the item arrived damaged. In-app messaging works the same way for brands whose customers are signed in, with the order history already in view.

**Best for:** brands where returns are a normal part of how people shop rather than a sign something went wrong, like clothing and footwear, and where an exchange keeps revenue the business would otherwise refund.

## Which channels conversational AI runs on, and why owned channels matter

Conversational AI can run anywhere a brand holds a two-way exchange with a customer, which includes the channels it already uses for delivery updates and order confirmations. It’s not limited to just a chat window on its own website.

Social channels like Instagram, Messenger, and Marketplace chat are great for first contact, as your brand is often found through social media first. But a conversation that happens there mostly stays there. Amazon and Meta each set rules on how long a business has to reply, what customer information it receives, and what it can keep. These platforms are good places to meet a customer for the first time, but what you learn there is harder to carry into the rest of your program.

Owned channels work the other way. Brands set the rules the agent operates under and keeps what the conversation produces, so a customer who mentions in a WhatsApp Business message that she's moved up a size has updated her own profile by saying so. The reorder prompt she gets next month offers the right product, and the agent handling her exchange in March already knows what she told it in January.

### WhatsApp Business messaging

WhatsApp is somewhere people already message friends and family, so a brand showing up there starts from a position of familiarity. It's also the richest of the channels, in that a customer can fill in a form, browse a set of products, and place an order without ever leaving the chat. It's a good fit for conversational AI because the channel supports open replies rather than only pre-written templates, so the agent's side of the conversation can be as flexible as the customer's. Our guide to [WhatsApp marketing](https://www.braze.com/resources/articles/whatsapp-marketing) covers the details, along with how to set a program up and get customers opted in.

### SMS and RCS

[SMS](https://www.braze.com/product/sms) reaches almost anyone with a phone. There's nothing to download and nothing to log into, which means fewer reasons for a customer not to answer. It's also already where most brands send delivery updates and reorder prompts. When someone replies, the conversation carries on from there rather than disappearing into an inbox. RCS runs on the same thread but carries more than plain text, so a brand can send product images and a recognizable sender name alongside the message, and it falls back to SMS where a handset or carrier doesn't support it.

### In-app and web messaging

[In-app messaging](https://www.braze.com/product/mobile-app-messaging) has one advantage the others don't, which is that the customer is already signed in. The agent knows who they are, what they've bought, and what they were looking at a moment ago, so a question like "Will this fit?" can be answered without anyone having to establish which product is being discussed. That makes it well suited to questions asked in the middle of shopping. Web messaging does the same for customers who browse on a laptop and have no reason to install an app.

## What conversational AI needs to work: data, catalog, guardrails and handoff

A working conversational AI setup needs four things working together to be successful. You need a unified view of the customer, accurate product and stock data, clear limits on what the agent can decide alone, and a way to pass a conversation to a person when it needs one. It involves systems that may not currently talk to each other and decisions that need sign-off from more than one team. These are the things to look at before you choose a platform, rather than six months in when a customer has been told the wrong thing.

### A unified customer profile

**What good looks like:** One profile per customer, holding order history, past conversations, and anything they've told the brand before, so the agent knows who it's talking to before the conversation starts. Unifying that data is the job of a [customer data platform](https://www.braze.com/product/braze-data-platform), which builds the profile from every source and updates it as the customer interacts.

**What happens without it:** The agent asks for an order number. Most customers read that as a sign the brand doesn't know them, and a fair number stop there and go to the website.

**What to ask the vendor:** Where does the customer profile live, and does a conversation update it?

### Live catalog and inventory data

**What good looks like:** Current stock at the level of the individual item and size, plus product detail specific enough to answer a question about fit or compatibility. An agent reads the same data a product page does, so anything vague there produces a vague answer.

**What happens without it:** The agent recommends something that sold out that morning. The brand then has to undo it, either by canceling the order or explaining itself to the customer.

**What to ask the vendor:** How does catalog and inventory data reach the conversation, and how often does it refresh?

### Guardrails and approval boundaries

**What good looks like:** Clear limits the business has set deliberately. Refunds up to a certain value, address changes after dispatch, a retention discount for someone about to cancel.

**What happens without it:** Set the limits too tight and the agent can't finish anything, which puts you back to deflection. Leave them undefined and you find out too late, after an agent has already promised something.

**What to ask the vendor:** Who controls what the agent is allowed to do without a person?

### A clean handoff to a person

**What good looks like:** The person picking up the conversation gets the thread, the order history, and whatever the agent had already established.

**What happens without it:** The customer explains everything again to someone new, having already explained it once. The added step and inconvenience makes a lot of people frustrated or they may leave the interaction altogether.

**What to ask the vendor:** What gets passed to the person taking over, and does the customer have to repeat anything?

## How Braze approaches conversational AI in eCommerce

Braze runs conversational AI through the same platform that sends the rest of your messages. That means agents have access to the same data, channels, and controls as everything else in the program.

That starts with the customer conversation itself. [BrazeAI™ Agents](https://www.braze.com/product/ai-agents) hold two-way exchanges in the channels you already use, reading from and writing to the same customer profile as every other campaign, so a size change a customer mentions by text is available to the next email without anything being synced between two systems.

What the agent is allowed to do in that conversation is set by your team. The brand guardrails cover what it can promise, refund, or change without a person involved, the agent’s tone and voice, and operational limits.

Then there's where it can read and write to. Catalog and order data get to the conversation from wherever you already keep them, including a cloud data warehouse, so stock levels and order records are available to an agent without a separate integration project first.

## How to evaluate conversational AI for your eCommerce stack

Six things will tell you whether a given option fits what you're trying to do. Score platforms using this framework and you'll create a shortlist for yourself to take to the demo stage.

| # | Criterion | What to check | What to ask |
| --- | --- | --- | --- |
| 1 | Channels, and whether you own them | Whether the agent reaches customers where they already hear from you, or only where you've installed it | Which channels can the agent run on, and which of those do we own? |
| 2 | Customer data access | Whether the agent recognizes a returning customer, and what it can see about them | Where is the customer profile held, and does the conversation write back to it? |
| 3 | Catalog and order system connection | Whether the agent can act on what it finds, rather than describing what the customer should do next | How do catalog and order data get into the conversation, and how current are they? |
| 4 | Guardrail controls | How much of the agent's behavior your team can configure, and how quickly | Who decides what the agent can do unsupervised, and can we change it ourselves? |
| 5 | Handoff quality | Whether escalation continues the conversation or restarts it | What does the person taking over inherit, and will the customer need to repeat themselves? |
| 6 | Measurement | Whether reporting shows commercial outcomes or only volume handled | Can we see assisted conversions and repurchase rate, not just resolution counts? |

If your customers almost all arrive by browser and you have no app or messaging list, a single-channel website assistant covers what you need. Owned-channel AI agents are a better fit when there's a mobile app, an SMS or WhatsApp list, and enough volume for conversations to add up.

_See how eCommerce brands run two-way AI conversations on WhatsApp, SMS and in-app with Braze._

## Conversational AI in eCommerce FAQs

### What is conversational AI in eCommerce?

Conversational AI in eCommerce is software that understands natural-language messages from shoppers and responds, recommends, or completes a task in a two-way conversation, across the messaging channels a brand owns. It relies on connection to real customer profile and catalog data, and knows when to hand off to a person.

### How is conversational AI used in retail and eCommerce?

Across the whole buyer journey. Before purchase, shoppers describe what they want and get relevant options back. At purchase, sizing and stock questions are answered. After purchase, order status and reorders, delivery changes, and returns run through the same conversational thread.

### What is the difference between conversational AI and a chatbot?

The difference is in architecture. A scripted bot matches messages against keywords it already knows and follows a pre-scripted decision tree. Conversational AI interprets intent from free text, retains information across an entire conversation, and can act on what it finds, including checking stock, placing an order, or changing a delivery.

### Which messaging channels work for conversational AI in eCommerce?

Any channel supporting a two-way exchange. WhatsApp Business messaging carries the richest conversations, SMS reaches almost anyone with a phone, RCS adds branded formatting with SMS fallback, and in-app messaging brings signed-in context. Owned channels have the advantage that conversations write back to your own customer profiles.

### How do eCommerce brands measure conversational AI?

By resolution and revenue, not deflection. Track how many conversations reached a resolution, how many assisted a conversion, and the effect on repurchase rate. Deflection shows how much work the agent absorbed, not what it earned. To isolate the agent's impact, compare these figures against a control group or the journey it replaced.

### Can conversational AI take orders or reorders over SMS or WhatsApp?

Yes. An agent connected to your catalog and order system can confirm a product, place an order, and process a change inside the message thread, with no website visit. Braze customer Aeroflow Health's agents handle SMS reorders this way, interpreting requests like "next size up" and matching them to the catalog. Repurchase conversions rose 8.8% compared with its previous SMS journey.

---

*Published 2026-10-28T18:59:00.000Z · Updated 2026-10-28T18:59:00.000Z · canonical lives at https://www.braze.com/resources/articles/conversational-ai-ecommerce*