The article
If you run an online store, you know the questions by heart: where is my order, how do I send this back, does it come in a size 10, will it arrive by Friday. They arrive at night and on weekends, and each one waits for a person who could be doing something else. An AI chatbot development service for ecommerce should take those questions off your team without ever guessing about an order, a size or a refund.
I have been building software for 18 years, and since 2017 my team of 70+ specialists has delivered 150+ projects. This page covers a store assistant: what it handles before and after the purchase, how it checks who is asking before it opens an order, how it plugs into the store you already run, and a table to hand to your support lead.
A store assistant in short
| What | How it works |
|---|---|
| Order questions | Status and tracking from your store or warehouse system, after the bot checks that the order belongs to the person asking |
| Returns and exchanges | Started within your written policy |
| Product questions | Answered from your catalog: sizes, stock and delivery times as they are at that moment |
| Help before the purchase | Suggestions from your catalog and stock, a link to the product, and the cart filled when the shopper asks |
| Hard cases | Passed to your support team with the whole conversation |
| Your store | One of the big platforms or code of your own, connected through its API |
| Price | An individual estimate; the first call and the rough estimate are free |
What an AI chatbot development service for ecommerce covers
It covers two jobs: support after the purchase and help before it. After the purchase, the assistant takes what fills a support inbox:
- Where is my order: the status and the tracking number from your system.
- Returns and exchanges, started within your policy.
- Product questions, answered from your catalog.
- Sizes, stock and delivery times, always current, because they are read from your data at the moment of the question.
- A person from your team when a case needs one.
Before the purchase, shoppers ask what fits, what goes with what and what arrives by Friday. The assistant answers from your catalog and your stock, links to the product and adds it to the cart when asked.
The assistant is part of our ecommerce development services for stores in the USA and Europe. The team that connects it to your store also builds stores and keeps them running, so the chatbot is not the only part of your shop we understand.
Where is my order: the check before the answer
Before the bot says a word about an order, it has to know the order belongs to the person asking. An order carries a name, a home address, a phone number and a list of what was bought, and an order number on its own proves little: it travels in emails and screenshots. There are three common ways to check, and we choose one with you in the scope:
- The shopper is signed in to your store, and the bot sees only the orders of that account.
- The order number plus the email or phone on the order, and both have to match.
- A one-time code sent to the email or phone on file, before anything that changes the order.
Only after the check does the bot pull the status and the tracking number from your store or warehouse system, never from the model's memory.
In October 2026 I read four chatbot pages from development companies that Google showed me for this search. One looks orders up by the customer's email or order number; none says how the bot makes sure the person asking is the person who ordered. Four pages are a small sample, so treat it as a question for your vendor, not as a statistic.
Returns and exchanges within your policy
The bot starts a return only where your written policy allows it, and passes everything else to a person. Return policies are written for people, who fill the gaps with judgment; a bot needs the gaps closed first. Before it can act, your policy has to answer:
- How many days a customer has, counted from the order, the shipment or the delivery.
- Which items are final sale.
- What condition an item has to be in, and who decides.
- Who pays for the return shipping.
- Refund, store credit or exchange, and when each applies.
- How an exchange works when the new size is out of stock.
Whatever the policy does not decide goes to your team: a damaged item, a missing parcel, a request outside the window, a customer who mentions a chargeback. Writing these answers down helps your human agents too.
Before the purchase: questions that turn into orders
A shopper who asks a question before buying is close to buying, and an answer at that moment keeps them on your page instead of in a search for the answer somewhere else. The assistant answers what fits, what goes with what and what arrives by Friday, from your catalog and your stock, and links to the product. When the shopper asks, it adds the item to the cart.
Two limits keep it honest. It never invents a product, a stock level, a delivery date or a discount: when the data is not there, it says so. And it is only as precise as your product data. If half of your products have no size chart, half of the fit questions go to a person, so size charts and product attributes can be the first job of the project, before the bot.
Built into the store you have
The assistant works with the store you run now, whether it sits on one of the big platforms or on code of your own. Through the platform's API it reads the catalog and the orders and passes cases to your help desk, so your team keeps working where it works today.
The catalog is indexed for search, so answers come from your product data, not from memory, and they change when the data does. To start, we need:
- Access to the store's API for the catalog and the orders.
- Your shipping, return and exchange policies, as they are today.
- The help desk or inbox where handovers should land.
- A month of real customer questions, so the test set starts from your inbox, not from our imagination.
- Someone who can tell a right answer from a wrong one.
Keeping the store itself working, from the checkout to the payment provider, is ecommerce website maintenance. A new store built around how you sell is custom web development.
Where the assistant should appear in your store
The assistant helps most where a question comes up, not only in a bubble in the corner of the home page. Four places earn their keep:
- Product pages, where fit, stock and delivery questions come up before anything is in the cart.
- The cart, where a shopper asks what arrives by Friday before paying.
- The order status page and the order emails, where a link to ask about the order beats a reply to a no-reply address.
- Your app and the messengers your customers already use, with the same rules behind each of them.
Each place gets its own first line. A shopper on a product page and a buyer waiting for a parcel need different first words from the same assistant.
Answer, act or hand over: a table for your support lead
Forward this table to whoever runs your support. It is where I would start the scope of a store assistant, and your policies move the lines.
| The shopper writes | The assistant | What it reads | It hands over when |
|---|---|---|---|
| "Do you ship to Canada?" | Answers | Shipping policy | Never needed |
| "Where is my order?" | Answers after checking the order is theirs | Order or warehouse system | The parcel is late beyond your promise, or lost |
| "Can it come on Saturday instead?" | Offers a new delivery date if your rules allow | Order system, delivery rules | No date fits |
| "Does this run small?" | Answers from the size chart and product notes | Catalog | The product has no size data |
| "Is it in stock in blue?" | Answers and links the product | Live stock | Never needed |
| "Add the blue one to my cart" | Adds it | Store API | Never needed |
| "I want to return this" | Starts the return within the policy | Order system, return policy | Outside the window, or final sale |
| "It arrived broken" | Takes the order number and what happened | Order system | Always |
| "Can I get a discount?" | Offers only the codes you approved | Your list of codes | Any other request |
| A chargeback threat or a safety complaint | Does not argue or answer on the merits | Nothing | Always, at once |
The rows that end in "always" are the point of the table. A bot that knows what it must not decide is the one you can leave alone on a Saturday night, and every row becomes a question in the test set before launch.
Several brands, languages or warehouses
A store with several brands, languages or warehouses gets the rules and handovers each of them needs. A return policy that differs by country, a warehouse that ships only part of the catalog, a brand with its own support team: each becomes a rule and a route in the scope, and each gets its own questions in the test set. Each shopper gets answers in their own language.
One more thing changes at that size: who approves the rules. Name one owner per brand or market before the project starts, or the scope waits for a meeting nobody schedules.
How you will know it pays off
Count the same things before and after launch, or the result becomes a matter of opinion. For a store assistant I would count five:
- Order-status questions that still reach a person each week.
- Returns started without a person, and the ones the policy sent back to your team.
- Questions the bot could not answer because the product data was missing; each one is a fix to the catalog.
- Products added to the cart from a conversation.
- Handovers, with the reason for each.
Start counting the first one now, before any bot exists: it is the baseline the assistant will be measured against. The general signals of a working chatbot, and what to do about each, are on the AI chatbot development page.
Our work behind it
Our published cases do not include a store assistant yet, so here is the work this one is built from.
In the AI Call Center project in the UAE, a voice agent takes routine calls at any hour while operators keep the cases that need a person, with models hosted in the country so that no call leaves it. For a store it is the same split of work: the AI takes "where is my order", and your team takes the broken parcel.
Mutualz is a city app for restaurants and cafes in Germany, in German and English: diners order and pay from the phone and collect points and vouchers, and restaurants run menus, orders and offers from an admin panel. The hard part was that nothing could get lost: every order, every point and every user action is tracked from the app through the restaurant to the backend. That is the kind of order data a store assistant has to be able to trust.
Tarot Fans is a marketplace app in France with live video sessions, paid calls charged per minute, chat, a shop where experts sell readings, and a wallet. Money flows both ways, from clients who top up and pay to experts who earn and withdraw, and a wallet built on Stripe keeps every balance straight: a shop, money and live chat in one app.
What the price depends on
A store assistant gets its own estimate, because most of the work sits in the connections, not in the model. The estimate follows:
- Your platform. A platform with a public API is the usual case; code of your own may need an API first.
- How many systems. The store alone, or the store, a warehouse system and a help desk.
- Your product data. Size charts, attributes and stock in order, or a cleanup first.
- What the bot may do. Answers only, or returns, delivery dates and the cart too.
- Brands and languages. Each adds rules, routes and test questions.
- Conversations a month, which set the running cost.
The first call and the rough estimate are free, and the price is fixed in the contract before work starts.
Next steps
How we build every chatbot, with a 15-minute check for any demo, is on the AI chatbot development page. For AI beyond the chat window, such as sorting the support email that never reaches it, see AI automation; for search inside your store that answers in sentences, AI integration. Everything else we build with AI is under AI development.
Tell us about your store: the platform, the questions your team answers most and where hard cases should land. We reply with what the assistant could handle and a rough estimate, free.
