Open the inbox of almost any online store and the pattern is the same. A long tail of interesting problems, and on top of it a thick layer of the same four questions: where is my parcel, when will it arrive, can I send this back, and do you have it in blue. That layer is what a chatbot for ecommerce is genuinely good at, and it is where you should start.
What follows is a practical order of operations: which queries to automate first, which ones to leave to a person, and how to tell the difference before you have annoyed a customer finding out.
What an ecommerce chatbot is, and what it is not
An ecommerce chatbot is an assistant in the corner of your shop that answers customer queries in plain language. The useful ones do not run on a script. They read the question, look through your own material — shipping page, returns terms, product descriptions, FAQ — and answer from what they find. When they find nothing relevant, they say so and pass the conversation to a human agent.
What it is not is a salesperson. Chatbots that open with a pop-up asking whether you would like to hear about today's offer are an interruption dressed as help. The version that earns its place is quiet until someone asks something, then answers accurately and fast.
It is also not a replacement for your support team. It is a filter. Everything predictable goes through it; everything that needs judgement lands with a person who now has time to deal with it properly.
Three kinds of ecommerce chatbots
The word covers three fairly different things, and comparing them like for like is how shops end up disappointed.
- Rule-based bots — a menu of buttons and a decision tree. Predictable, cheap, and useless the moment a customer types something you did not anticipate.
- Retrieval-based AI chatbots — they read the question and answer from your own documents. This is what most e-commerce brands actually want for customer support.
- AI agents — an AI agent also acts: fetching an order, checking stock, starting a return. Powerful, and dependent on being connected to the systems that hold that information.
Modern ecommerce chatbots blur the middle two, and that is fine. The distinction that matters is whether the thing answers from your content or from a script you have to maintain by hand.
The five queries worth automating first
Pick the questions that arrive daily, have one correct answer, and can be answered from something you have already written. In most ecommerce businesses that comes down to five.
1. Order tracking
"Where is my order" is the single most common message a webshop receives, and it is pure repetition. Order tracking is worth automating first because the answer already exists in your system — the customer just cannot see it. Look for a chatbot that can verify identity before it reveals anything: an order number plus a postcode or house number is the usual bar. Reading out a delivery address to whoever types a guessed number is not a feature.
2. Delivery times and shipping costs
Cut-off times, weekend handling, free shipping thresholds, deliveries outside your home country. These change with the season and customers ask about them constantly, often on the product page, right before deciding. A correct answer here does more for conversion than any amount of banner copy.
3. Returns and exchanges
How long is the window, who pays the postage, what happens with sale items, how long a refund takes. Returns questions are emotionally loaded and factually simple — the perfect thing to automate, as long as the policy the chatbot reads is the current one.
4. Stock and availability
Whether something is in stock, when it comes back, whether a different size exists. If your bot can reach live inventory this is a real-time answer; if not, it can still point at the product page and offer to notify a colleague. Both beat silence.
5. Size, fit and compatibility
Does this fit my model, does this run small, is this cable the right one. These sit in your product descriptions and manuals already. An AI chatbot that has read them can answer variations no FAQ page would ever cover, which is exactly where a scripted bot falls over.
What to leave to a person
Automating the wrong thing costs more than automating nothing. Keep these with your support team:
- Complaints — a damaged parcel or a late birthday present needs a human tone, not a policy quotation.
- Anything involving money — refunds outside policy, goodwill gestures, disputed charges.
- Edge cases in warranty or repair — where the right answer depends on facts nobody has written down.
- Anything the bot is unsure about — an honest "let me get a colleague" protects customer satisfaction far better than a confident guess.
That last point is the whole design principle. A chatbot's value is capped by how well it fails, which is why the handover matters as much as the answers. We go through the mechanics in how a conversation should move to a real person.
Do product recommendations work?
Sometimes, and later than most vendors suggest. An AI shopping assistant that can recommend products from your catalogue is a genuine use case — "I need a rain jacket for cycling, under 150 euros" is a query a search box handles badly. But it only pays off once your product data is clean and your descriptions actually describe things. If your catalogue is thin, product recommendations will be thin too.
The sequence that works: get the support questions right first, watch the transcripts, and add recommendations when you can see people asking for them. Fixing the inbox has a payback you can measure in hours saved this week. Recommendations are an experiment.
How an AI chatbot for ecommerce works
Almost all of them share one design. Your documents are indexed. A customer asks something; the system retrieves the passages most likely to be relevant and passes them to a language model along with the question. The model answers using that material and nothing else. Natural language processing handles the messy phrasing customers actually use — "cant return, past 30 days??" gets the same answer as a neatly typed sentence.
Two things follow from this. First, the assistant is only as good as what you feed it, which is why training a chatbot on your own data is the part that deserves your afternoon. Second, updates are instant: change your returns page, and the next answer changes with it. There is no retraining step.
It also means the system knows when it retrieved nothing useful. An ai-powered assistant built honestly uses that as a stop signal rather than an invitation to improvise.
Connecting it to your ecommerce platform
Installation is usually one script tag. On Shopify that goes into theme.liquid before the closing head tag; on WooCommerce a header-scripts plugin does the job; a custom store takes it directly. Nothing about this requires a developer, and our setup documentation covers the common platforms step by step.
Deeper integration is a separate decision. Reading order status from your existing ecommerce platform gives much better answers, but it also means the chatbot handles customer data, so check what is stored, for how long, and where. Ask that question before you switch it on, not after.
Notifications deserve a moment's thought too. When the assistant hands a conversation over, someone has to notice. A message into Slack or Teams gets picked up; a badge on a dashboard nobody has open does not. Small ecommerce businesses lose more conversations to unnoticed handovers than to bad answers.
Questions ecommerce managers ask
How much does a chatbot for ecommerce cost?
It varies more than it should. Some chatbot platforms charge per resolved conversation, which makes a good month expensive; others charge a flat fee and meter the AI. Veyra's plan is free and you connect your own OpenAI key, so the model usage is billed to you directly by OpenAI instead of being marked up. Whichever route you take, work out the cost of a busy month, not an average one. Our notes on what a free chatbot really includes cover the usual catches.
Will a chatbot increase sales?
It can, indirectly, and be wary of anyone quoting you a precise figure. The honest mechanism is this: shoppers who get a delivery or sizing answer at the moment of doubt are less likely to abandon the basket. That is a real effect on the customer journey, but it depends on your traffic and your answers, so measure it in your own shop rather than trusting a case study.
Which channels should it cover?
Your website first — that is where the customer already is, product in front of them. Add Facebook Messenger or WhatsApp only if your customers already message you there. Extra channels multiply the places you have to watch, and an unanswered message on a channel you forgot about is worse than no channel at all.
How do I know whether it is working?
Read the transcripts. Not the dashboard — the transcripts. Count how many conversations finished without a person, how many needed a handover, and how many ended with the customer repeating themselves. That third number is the one that tells you something is wrong. Automation metrics look good long before customers agree, which is why we argue for reading the raw conversations in our piece on what to automate and what to leave alone.
Does it work in more than one language?
Usually yes, and better than you would expect. A model that answers from your Dutch returns policy can reply in English or German without you writing a second version, because the translation happens at answer time. Two cautions: check the tone of the translated replies yourself before you trust them, and remember that anything legally worded — warranty terms, right of withdrawal — should be answered from a page you have actually had translated, not paraphrased on the fly.
Do I still need live chat?
Most shops end up with both: the bot in front handling volume, a person behind it for everything else. The comparison is set out in chatbot vs live chat, including the case where a small shop is genuinely better off with just one.
A sane way to implement a chatbot
Start narrow. Load your shipping policy, returns policy and top twenty FAQs. Switch on handover from day one. Let it run for two weeks and read every conversation, however tedious that sounds — the gaps in your documentation will be obvious within the first fifty.
Then widen it: product manuals, size guides, care instructions. Then, if it is worth it, order lookup. Each step should be justified by something you saw in a real transcript rather than by a feature list.
The shops that get the most out of an ecommerce chatbot are not the ones that automate the most. They are the ones that automate the boring half accurately and route the rest to a person quickly. If you want the wider view of where this fits, our pillar guide to the customer service chatbot covers the whole picture, and the buyer's checklist for a website chatbot is the shorter version to take into a vendor call.