A shopper lands on your store at 11pm, finds a jacket they like, and has one question before they buy: does it run small, and can it arrive by Friday? Nobody is awake to answer. They leave, and most of them never come back. An AI chatbot for ecommerce exists for that exact moment: a chat assistant that lives on your store, knows your catalogue and your policies, answers product and order questions instantly, and hands the complicated cases to a human — at any hour, without a ticket queue.
I build AI chatbots for online stores through my development services, including WooCommerce and Shopify builds where the chat is part of the store, not bolted on. This guide covers what an ecommerce chatbot actually does on a live shopping session, where it makes money and where it wastes it, what it costs in 2026, and how to deploy one without annoying the shoppers you are trying to keep. For the broader picture, start with my AI chatbot development overview and my chatbot cost guide.
What an AI Chatbot Does on a Live Shopping Session
A shopper opens the chat and types like a person, not a search box: "a red dress under 300", "do you have this in size 10?", "where is my order?". A good ecommerce chatbot does five things from there, based on rules the store sets in advance.
First, it does product discovery. It searches the live catalogue from a plain description — colour, size, price range, use case — and shows matching products as cards with photo, price, and availability, instead of making the shopper filter through fifty results. It can compare two products, explain the difference in plain language, and suggest the right size or variant from the store's own size guides. Second, it answers order questions with real data: where the parcel is, when it arrives, and what happens if it is late — the "where is my order" (WISMO) questions that eat most of a small store's support time. Third, it handles returns, exchanges, and cancellations: starting the return, generating the label or instructions, and confirming the refund timeline from the store's policy. Fourth, it helps at checkout: delivery options and costs, payment methods, discount codes, and the nudge that recovers an abandoned cart while the shopper is still deciding. Fifth, it escalates: a damaged item, an angry customer, a bulk or wholesale enquiry, or anyone who asks twice for a person goes to a human, with the full conversation attached.
The difference between an AI chatbot and the old rule-based pop-up is that the shopper talks normally. Nobody has to click through "track order" and "returns" menus, and nobody gets a canned answer to a question the store could have answered from its own data. The chatbot reads the catalogue, the order system, and the policy pages — and says so when it does not know.
Sales Mode and Support Mode Are Two Different Jobs
The best ecommerce chatbots switch between two modes without the shopper noticing, and a store should buy or build for both. In sales mode the chatbot is a shop assistant: it recommends products, explains differences, suggests the matching accessory, and makes it easy to add to the cart without leaving the conversation. In support mode it is a service desk: it tracks orders, explains returns, and fixes the small problems that would otherwise become tickets.
Most off-the-shelf tools are strong at one mode and weak at the other. Support-first tools (the helpdesk family) are excellent at order tracking and returns but treat product recommendations as an afterthought. Sales-first tools are excellent at guided selling and quizzes but cannot touch an order or a return. Before you buy, decide which mode your store needs most. A store drowning in "where is my order" emails needs support mode first. A store with a big catalogue and confused shoppers needs sales mode first. A store with both problems needs a tool — or a custom build — that does both, because a chatbot that can sell but cannot help after the sale creates the very tickets it was bought to remove.
Why Stores Lose Sales They Never See
Most stores underestimate the cost of unanswered pre-purchase questions because the loss leaves no record. Analytics shows the exit; it does not show the question that caused it. The questions cluster around the same handful of doubts: sizing and fit, delivery time, whether an item is actually in stock, the returns policy, and whether a discount code works. Every one of them is answerable from data the store already has — and every unanswered one is a sale that went to a competitor whose chat did answer.
The honest test is simple: read your last fifty support emails and chats, and count how many were questions the store's own pages could have answered. If that number is most of them, a chatbot is not a nice-to-have — it is the cheapest support hire you will ever make. Vendor case studies will happily quote deflection rates of 80 or 90 percent; treat those as vendor numbers. Your own inbox is the only figure that matters for the buying decision. My WhatsApp chatbot guide covers the same triage logic for messaging channels, and it applies just as well to the on-site chat.
The second leak is subtler. Even stores that answer quickly often answer badly: a copy-paste policy paragraph for a sizing question, a tracking link with no explanation, a returns answer that contradicts the checkout page. A chatbot that answers from one approved knowledge base gives the same correct answer every time, at 3am as well as 3pm. Consistency is the product here, not novelty.
Shopify vs WooCommerce: The Setup Realities
The platform decides how much of the store the chatbot can actually touch, so check integration before price. On Shopify, most chatbot tools connect natively: they read the product catalogue, stock, and order data, show product cards in the chat, and let the shopper add to the cart without leaving the conversation. On WooCommerce, the same tools usually connect through a plugin or API, and the depth varies — product and order data is generally available, but cart actions inside the chat and live stock checks depend on the specific tool and setup.
Two questions to ask any vendor before you commit. First, does the chatbot read live data — price, stock, variants — at the moment of the answer, or from a catalogue snapshot that goes stale when you change a price? A snapshot that is a day old will happily sell a size you sold out of yesterday. Second, what happens when the shopper asks about something the catalogue does not cover — a custom order, a bulk discount, a product you have not listed yet? The right answer is a clean handoff to a human, not a confident guess. If a vendor cannot show you both behaviours on a live demo with your own products, that is your answer about the vendor.
For stores that need the chatbot wired into a custom stack — a headless storefront, a CRM, or an ERP behind the shop — the off-the-shelf tools usually stop short, and that is where a custom build earns its keep. My RAG chatbot implementation guide explains how a chatbot is connected to a store's real data instead of a generic model, which is the difference between answers you can trust and answers that sound plausible.
What an AI Chatbot for Ecommerce Costs in 2026
Pricing falls into two buckets, and the right one depends on how much of your store the chatbot must touch.
Off-the-shelf ecommerce chatbot tools charge a monthly subscription, usually tiered by conversations or resolved outcomes. As of late 2026, entry points run roughly from $25 to $50 a month for small-store tools (for example Tidio's Lyro AI from about $32.50 a month, Gorgias AI Agent plans from about $40 a month), rising to $99 a month and beyond for sales-focused agents with product discovery and cart recovery, and to per-outcome pricing for larger operations. Most offer a free plan or trial — use it with your real catalogue before you pay, because a demo with sample products hides every integration problem. Check the live pricing page before you buy; these numbers move every quarter.
A custom build costs more up front but is wired into your actual store: your catalogue, your stock rules, your shipping and returns policies, your order system, and your escalation list. You can see typical project pricing structures on my pricing page. The custom route makes sense when you have a large or frequently changing catalogue, a WooCommerce setup the SaaS tools integrate with poorly, or a checkout and post-purchase workflow that off-the-shelf tools keep getting wrong. For a small store with a simple catalogue, start with a subscription tool, measure deflection and conversion for a month, and let the numbers decide whether a build is worth it.
Whichever route you take, budget for the unglamorous parts: writing the product knowledge base, approving the returns and shipping answers word for word, connecting the order system properly, testing with real staff playing difficult shoppers — the one who changes their size three times, the one whose order is late, the one asking about a product you do not stock — and reviewing the first month of conversations. The software is the cheap part. The catalogue accuracy is the work that makes it safe.
The Answers a Chatbot Must Never Guess
Some answers a wrong guess turns into a refund, a chargeback, or a lost customer: "is this in stock?", "will it arrive by Friday?", "can I return this?", "does this contain nuts?" The chatbot may only answer from live store data and the approved policy pages. If the answer is not in the data, the chatbot must say so and offer a human — never guess, never improvise a delivery date or a stock level.
This is a data job, not a model job. Before launch, the store writes down the approved answers: shipping times and costs per region, the returns window and who pays return shipping, the sizing guidance per product type, and the discount rules. The chatbot answers from that list and live data, and nothing else. It also needs a clear rule for what it cannot do — change a price, promise a delivery date the carrier has not confirmed, or approve a refund above a set amount — and a clean handoff for each.
The same discipline applies to discount codes. A chatbot that invents a code, or applies one that expired, will cost you margin and trust in the same conversation. Codes come from the store's live promotion list, or the chatbot does not mention them at all.
How to Deploy Without Annoying Shoppers
Shoppers will forgive a chatbot that is fast, accurate, and honest about what it is. They will not forgive one that pretends to be human, pops up on every page, or traps them in a loop while their question goes unanswered. Five rules cover most of it.
First, do not ambush the shopper: trigger the chat after they have spent time on a product page, or when they show exit intent — not the second the homepage loads. Second, say it is AI in the first message, and make "speak to a person" work from anywhere in the conversation, instantly, without a lecture. Third, keep product recommendations to two or three options with a reason each; a wall of ten products is a search results page, not a recommendation. Fourth, log every conversation — what shoppers asked, what the chatbot recommended, what it got wrong — and fix the knowledge base weekly for the first month. A chatbot improves fastest in its first thirty days, and only if someone is actually reading the chats. Fifth, measure the right things: deflection of routine questions, conversion on chats the bot handled, and average order value on bot-assisted sessions — not raw chat volume, which goes up when the bot is failing too. And if your customers also call, remember the phone still matters: my chatbot vs voice agent guide helps decide which channel to automate first.
The Bottom Line
An AI chatbot for ecommerce pays for itself in the questions it answers while your team sleeps — the sizing question that would have been an abandoned cart, the order-tracking question that would have been a ticket, the product recommendation that would have been a lost sale to a competitor. It does not replace your support team. It takes the routine work off their hands, answers from your real catalogue and policies, and hands the complicated cases to a human with the context attached.
If your inbox is full of questions your own pages could answer, start by counting them for two weeks. That number tells you whether a chatbot is a nice-to-have or a leak worth fixing — and it gives you the baseline to judge any tool against. If you want one built around your catalogue, your order system, and your policies instead of a generic script, talk to me — I build chatbots for ecommerce stores, and the first step is always the same: read your real customer conversations before writing a line of code.

