Conversational AI for ecommerce use cases, results & how to start

Last holiday season, AI and AI agents drove 20% of all retail sales, about $262 billion, according to Salesforce holiday data. Shoppers are already typing questions into chat windows and expecting a real answer back. Conversational AI for ecommerce is software that talks with shoppers in plain language across chat, SMS, email, and voice, and then acts on what they asked, whether that's finding the right product, checking an order, or starting an exchange. For the broader definition, our glossary covers conversational AI in general.

The rest of this guide is practical. You'll see where conversational AI fits across the shopping journey, what results brands like KÜHL and Rothy's have seen, how a small team can start without a developer, and what to ask when you choose a platform.

How conversational AI for ecommerce differs from a chatbot

Most shoppers have met the old kind of chatbot. It matches keywords to a list of FAQ answers, and when a question falls outside the list, it loops or sends a link to the help center. That works for "what's your return window?" and falls apart the moment someone asks "can I swap the medium I ordered Tuesday for a large in the green?"

Conversational AI for ecommerce understands the question the way a person would, and it's connected to the systems that hold the answer. When a shopper asks about sizing, it checks the product catalog. When they ask where their package is, it looks up the order and reads the tracking status. When they want to cancel, it checks whether the order has shipped and cancels it on the spot if it hasn't, so the shopper leaves with the problem solved and nobody on your team has to step in.

It's also a narrower idea than conversational commerce, which is the wider practice of selling and serving customers through messaging channels. Conversational AI is the technology that lets you do conversational commerce at scale, without a person reading every message.

Conversational AI use cases across the shopping journey

The most useful way to think about conversational AI is to follow one shopper from the first question to the second purchase. Most guides split this into "sales use cases" and "support use cases," and shoppers don't experience it that way. To them it's one relationship with your brand, and the setups that work best treat it as one conversation.

Product discovery and guided selling

A shopper lands on your site and types "I need a waterproof jacket for hiking in the rain, under $150." A search bar returns 40 jackets sorted by relevance. Conversational AI for ecommerce product discovery asks a follow-up or two, like how cold it gets where they hike or whether they want something that packs down small, and then recommends three or four jackets that fit.

Those follow-up questions do the heavy lifting, because a shopper who says "a gift for my dad who golfs" hasn't given a search engine much to go on, but a good conversation can narrow it to a price range, a size, and a style in about four messages. That's also a natural point to suggest the gloves or the hat that pair with what they picked.

This is the closest thing an online store has to a good salesperson on the floor. It's also where smaller brands can compete with big retailers, because a shopper who gets a confident, specific recommendation from you has less reason to go comparison shopping on a marketplace.

Gladly trains its AI on 450 million retail conversations, so it starts with a working sense of how shoppers ask about fit, materials, and gifting before it learns your catalog.

Product questions on the product page

Once someone finds a product, the questions get specific. Does this run small? Is the lining wool or synthetic? Will this case fit the older model? Will it arrive by Friday if I order today?

Every one of those questions is a moment where a shopper either buys or leaves. When the answer takes 12 hours by email, most of them leave. When it takes 10 seconds in chat, a lot of them buy, and the ones who don't at least leave with a good impression of your brand.

Sizing and fit questions pay off twice. Answering them well before the purchase is one of the few levers you have on your return rate, because a shopper who picks the right size the first time never needs to send anything back. If your size chart is buried in a pop-up, conversational AI can pull the same information into the answer and add what the chart can't say, like "most customers size up in this style."

The answers are only as good as your product data. If your product descriptions skip the fabric weight or the inseam, the AI can't invent them, and it shouldn't try. Cleaning up the top 20 or 30 product pages before launch is usually a better use of an afternoon than any setting in the AI itself.

Cart and checkout help

Carts get abandoned for small, fixable reasons. A promo code doesn't work. The shopper can't tell if they've hit the free shipping threshold. They want to know if two items can ship together or whether a gift note is possible.

Conversational AI can answer those questions in the moment and keep the shopper moving. A shopper who's $12 short of free shipping will often add a pair of socks if someone points it out, which a static banner rarely manages. With Gladly, checkout happens in the thread, with no redirect, so the shopper doesn't have to leave the conversation to finish buying.

Order tracking and "where's my order?"

"Where's my order?" is the most common question most ecommerce support teams get, often shortened to WISMO. It's also the easiest to automate well, because the answer already lives in your order system.

A good setup asks for an order number or email, looks up the order, reads the fulfillment and tracking status, and gives the shopper a direct answer with the carrier link. With the Shopify integration, Gladly can look up orders by order number or email address and pull shipment details like the estimated delivery date and tracking links.

Outdoor apparel brand KÜHL saw a 44% reduction in WISMO emails after moving those questions to AI. Their team used to come in on Monday to more than 100 weekend emails, and now it's about 40.

Returns, exchanges, and saving the sale

Returns are where most conversational AI setups come up short, and where the biggest money is hiding.

When a customer asks to return a jacket that doesn't fit, a basic bot sends a return label, a refund follows, and the sale is gone. When the AI that sold the jacket also handles the return, it already knows what they bought, when, and in what size. It can offer the next size up and start an exchange in the same thread, and the customer ends up with a jacket that fits while you keep the revenue.

The math is simple enough to run on your own numbers. Say you process 200 fit-related returns a month at an average order value of $120. If a well-timed exchange offer turns one in five of those into an exchange, that's 40 orders, or $4,800 a month, you keep as revenue. Your rate will be different, and it's worth tracking from the first week.

Gladly describes the approach this way on its conversational AI for sales page: "The same AI that just sold the jacket handles the exchange, the 'where is it,' and next month's sizing question."

Order changes belong here too, because a shopper who can fix a mistake quickly is far less likely to return the order later. Gladly can update a shipping address or cancel an unshipped Shopify order, with an optional refund and restock. Allbirds uses Gladly to resolve 87% of "cancel order" conversations, and Bombas reaches an 88% resolution rate on repair and damage orders.

Meeting shoppers on the channel they already use

Shoppers don't pick a channel based on what's convenient for your support team. Some text and some email, and a surprising number still pick up the phone. If your AI only lives in the chat widget, everyone who reaches out another way waits in a queue for a person.

Gladly runs the same AI across chat, SMS, email, and voice, and keeps each customer's conversations in one timeline. Rothy's moved chat from 2% of its conversations to 41% after switching, which is a good sign of what happens when a channel starts giving fast, useful answers. As Rothy's Head of CX put it, "It's really impactful that we can have all channels inbound into the same conversation timeline."

From support conversation to the next purchase

A customer asking about care instructions is thinking about the product. A customer whose issue just got fixed quickly is in a good mood. Both are sales conversations, even though they arrive in the support queue.

At Rothy's, more than 20% of customers made a purchase after connecting with a team member. KÜHL saw 120% more revenue per call once its team had time to talk about products, where before they'd been racing to shrink the queue. When AI takes the repetitive questions, the people on your team get time for the conversations that build loyalty and lead to the next order.

What results actually look like

Big claims about conversational AI are easy to find. A lot of them trace back to vendor roundups quoting other vendor roundups, with no study behind the number. When you see a stat like "AI resolves 80% of inquiries," ask who measured it, across how many conversations, and what counted as resolved. A conversation where the customer gave up and left can look a lot like a resolution in a dashboard. Here's what the primary data says, with the caveats attached.

Salesforce found that retailers with their own AI agents grew sales 6.2% year over year during the 2025 holidays, compared with 3.9% for those without, and that shoppers used retailers' AI for customer service 126% more during the holiday rush than in the two months before.

Rep AI's analysis of more than 17 million shopping sessions found that 12.3% of shoppers who engage with AI chat make a purchase, compared with 3.1% of those who don't. Read that one carefully, because shoppers who start a chat were already more interested than the average visitor, so part of that gap comes from who chooses to chat in the first place. It's still a strong signal that the conversation is where your buyers show up.

Gladly customers see results on both the service side and the revenue side:

• Rothy's resolves 31% of conversations with AI at an AI CSAT above 93%, with a small team and no in-house engineers.

• KÜHL reports a 79% email resolution rate, 63% on SMS, and 59% on chat, alongside the 44% drop in WISMO emails.

• Cosabella resolves 48% of its conversations with Gladly.

• KÜHL's team brings in 120% more revenue per call.

The numbers you'll get depend on your question mix and how well your AI is connected to your order and product data. To model your own, try the CX ROI calculator. If you're still deciding whether AI is worth the investment at all, our guide to the ROI of AI in commerce walks through the math.

How to start small without an enterprise budget or a dev team

A lot of conversational AI advice assumes you have a data team, a customer data platform, and six months. Most ecommerce brands have none of those, and they don't need them to get started.

Pick one high-volume question

Look at last month's conversations and find the question you answer most. If your help desk doesn't tag conversations by topic, read through the last 200 and tally them by hand. It takes an afternoon, and the answer is rarely a surprise: for most online stores, it's order status. For apparel and footwear brands, sizing is often a close second. Start there.

One question, answered well, gives you a clean before-and-after. Within a few weeks you'll know whether volume dropped, whether customers were happy with the answers, and where the AI needed a person.

Connect the data the AI needs

Conversational AI is only as good as what it can see. To answer order questions, it needs your order system, which for many brands means Shopify. To answer product questions, it needs your catalog. To answer policy questions, it needs your actual return and shipping policies written down in one place, with the exceptions your team already knows by heart.

That's usually a shorter list than people expect. Gladly connects to Shopify out of the box, and CX teams can configure and test AI behavior in plain language, with no engineering tickets.

Design the handoff before you launch

Decide ahead of time which conversations should go to a person. Refund disputes, damaged items, and anything involving a VIP customer are common picks. Then make sure the person sees the full history when they pick the conversation up, because nothing frustrates a customer faster than explaining the problem to a bot and then explaining it again to a human.

Rothy's got this right. In the words of their Head of CX, "The customer doesn't feel that the handoff between AI and agents is clunky."

Measure a few things from day one

Pick your numbers before launch so you can tell whether it's working. Resolution rate tells you how many conversations the AI closed without a person. CSAT on AI conversations tells you whether customers were happy with how it closed them. Handoff rate and handoff reasons show you what to fix next. If the AI is selling, track revenue from conversations too, since that's the number that turns a support project into a growth project.

Expand one use case at a time

Once order status runs smoothly, add sizing. Then cancellations and address changes, and after that exchanges. Each new use case builds on the data connections you already have, so the second and third go faster than the first.

Why conversational AI projects stall

Conversational AI projects rarely fail on launch day. They stall a few weeks in, for reasons you can usually see coming.

A sales bot and a support bot that don't share context

Plenty of brands buy a shopping assistant for the product page and a separate support bot for the help center. The shopper doesn't know or care which bot they're talking to. When they come back with a question about the order the first bot helped them place, the second bot has no idea who they are, and they start over from scratch.

Answers that aren't connected to real data

AI that isn't connected to your order system and your real policies will guess. It'll tell a customer their return window is 60 days when yours is 30, or promise a delivery date it can't see. Every wrong answer creates a second conversation, usually an angrier one, and sometimes a public review.

Trying to automate everything at once

It's tempting to switch on every use case in week one. When 15 different kinds of questions go live at the same time and a few of them go wrong, it's hard to tell which ones, and your team loses trust in the AI before it's had a chance to prove itself. Brands that start with one or two question types, get them right, and then add more tend to end up with broader coverage a few months in than brands that tried to do it all on day one.

Handoffs that lose the history

If the conversation history doesn't travel with the customer to your team, you've automated the easy part and made the hard part worse. Your team starts every escalated conversation by asking questions the customer already answered.

Launch and forget

Gladly puts it bluntly on its sales page: "Most AI rollouts take six months to ship and plateau six weeks after." Your catalog changes, your policies change, and new questions show up every season. Someone on your team should review AI conversations every week, fix the answers that missed, and add the questions that keep coming up. For a small team, 30 minutes on a Friday is often enough.

How to choose a conversational AI platform for ecommerce

The demos all look good. These are the questions that separate a platform that'll work for your store from one that'll stall:

• Does it cover selling and support in one conversation? A shopper who asks for a recommendation today and an exchange next week should be talking to the same AI, with the same memory.

• Can it see order and customer history? Ask to watch it look up a real order during the demo.

• Can it take action? Answering "can I exchange this?" is easy. Starting the exchange, canceling an unshipped order, or updating an address is the part that saves your team time.

• How does the handoff work? Ask what your team sees when a conversation reaches them. The answer should be the full history.

• Which channels does it cover? Your customers use chat, SMS, email, and phone. One AI across all of them beats a separate tool for each.

• Does it report on revenue as well as resolution? If you only measure how many conversations the AI closed, you'll miss the sales it made or saved.

• How long until it's live, and who has to build it? For a small team, "no engineering required" matters more than any feature on this list.