An ecommerce AI agent, often sold as an AI shopping assistant, answers shoppers' questions about products, sizing, shipping and returns from your store's own pages and files, suggests products from your catalog, collects an email from the shopper who is not ready to buy, and hands the conversation to a person when your content runs out. Building one is mostly a content job, not a technical one. Most teams are live the same day; the five-day plan below spreads the work out so the agent is right on day one rather than merely live.
Why most store chatbots disappoint
The first generation of store chatbots ran on scripts. A shopper asked whether the linen shirt runs small, the bot offered three buttons that had nothing to do with linen, and the shopper left. Swapping the script for a language model fixed the loop and created a new problem: an agent that answers everything, including questions your store never documented, in the same confident tone.
The numbers in this category do not help. Vendor pages promise conversion multiples and a share of queries "automated," usually with no denominator. Automated out of what: every chat, or only the ones the bot chose to count? A rate without a denominator is a guess wearing a number. The figure that matters is the share of evaluated conversations resolved without a person, and it depends far more on your content than on the vendor.
Wonderchat's support team sees one failure most often, and it is quieter than a made-up answer. It is retrieval pulling an archived policy and answering from it correctly: last season's returns window, quoted during this season's sale. The citation is real, the answer is wrong, and nobody notices for weeks.
What a good ecommerce AI agent does
Five capabilities separate an agent that sells from one that frustrates. The first two decide whether shoppers can trust it at all.
It answers only from your content and shows the source. Product pages, size guides, shipping and returns policies, catalogs, and FAQ answers. Each reply should link to the page it came from, so you can check it and the shopper can too.
It says so when it does not know. "That is not covered on the site, so here is a person who can help" beats a plausible guess about a fabric blend.
It helps shoppers choose. Product discovery is where most store bots are weakest. A good agent asks one or two questions ("for running or for walking?") and suggests products from the catalog, with links.
It hands over cleanly. When a shopper asks for a person, or the agent cannot help, the conversation goes to your team with the full transcript, by email or as a ticket in your help desk.
It knows the brand. A persona ("warm and brief" or "precise and formal"), a welcome message, suggested questions, and hard limits on topics it will not touch.
Order status is the capability to check most carefully. Whether an agent can read live order data depends on how it connects to your store platform. Ask any vendor to show you a real order lookup on your own store before you pay.
The large retailers show where this is heading. Amazon's Rufus answers product research and comparison questions inside the Amazon app, and Walmart runs its own shopping assistant. McKinsey calls the next step agentic commerce, AI agents shopping on a consumer's behalf, and estimates it could orchestrate up to $1 trillion of US B2C retail revenue by 2030. Those are the shopper's agents, not yours. Both kinds read the same thing, though: your product pages and policies. Clear, complete content serves the agent on your site and the one in the shopper's pocket.
The five-day plan to build your agent
The plan below is platform-neutral, with notes on how each step works in Wonderchat, which runs an AI Customer Service Agent for support questions and an AI Sales Agent for product questions and lead capture. Five days is generous. The point of spreading it out is to test before shoppers do.
Day 1: Collect the content the agent will answer from
Gather everything a good store assistant would have read:
Product description pages and the full catalog (PDF or DOCX is fine).
Shipping, returns and exchange policies, including international rules.
Size guides, care instructions and product comparison pages.
Your FAQ and help center articles, plus any policy documents the agent should answer from.
Point the agent at your store URL so it reads the site, then upload the files. Now the step most stores skip: list the questions your inbox gets most often and check that the content actually answers each one. " If your size guide does not say whether the shirt runs small, no agent can. Write the missing pages now.
Day 2: Set the persona and the boundaries
Write a short base prompt that sets tone and scope. For example: "You are a friendly, brief shopping assistant for Luxe Apparel. Help customers find the right outfit and answer questions about our materials, sizing and delivery. If a question is not covered by the store's content, say so and offer to connect them with the team."
Then set a welcome message and three or four suggested questions that match your top inbox topics. Decide what the agent must never discuss: competitor products, medical claims about skincare, discounts you have not published. Match the brand voice, playful or formal, but keep answers short. Shoppers on a product page want the answer, not a personality.
Day 3: Build the sales workflows
Answering questions is half the job. The other half is starting the right conversations. Three workflows earn their place on most stores:
Proactive help on product pages. After a shopper has spent a while on a product page, the agent opens with "Can I help you find the right size?" and two buttons: "Yes, help me" and "Just browsing." Custom triggers and dynamic greetings that open the chat on their own are on the Scale plan.
Email capture for later. When a shopper asks about restocks or upcoming offers, the agent asks for a name and email and passes them to your CRM.
Comparison and fit. A short sequence that asks what the shopper needs and compares two or three products from the catalog side by side.
Keep each sequence short. A shopper who has to answer a questionnaire before getting help has already opened a competitor's tab.
Day 4: Connect the stack
Connect the agent to the tools that act on what it learns. In Wonderchat, email handover and the website widget work on every plan, including the free one. Captured leads sync to HubSpot from the Basic plan, and unresolved conversations can be handed into Zendesk or Freshdesk as tickets with the full transcript attached. The Shopify integration connects the agent to your store for order tracking and product search on the Basic plan, and Slack and WhatsApp work from Starter. Stores with a custom backend or unusual workflows can build on the API. For a small store team with real shipping, returns and sizing pages, Wonderchat fits: it answers from them with the source shown, looks up Shopify orders from Basic, and issues no refunds.
One rule for this day: decide the handover path before launch. Teams on Zendesk or Freshdesk keep them and, from Basic, the agent hands over into them; otherwise Wonderchat's own tickets, with the full chat attached, are on every plan.
Day 5: Go live, test, and read what it missed
Paste the widget snippet into your site header and go live on your highest-traffic pages first: product pages and the cart. Then try to break it. Ask about a product you discontinued, a returns case your policy does not cover, a question in another language (Wonderchat answers in 40+ languages). Test the handover with a question you know the content cannot answer.
After a week, read three things:
Resolution rate, with its denominator. Across all Wonderchat accounts, about 75% of evaluated conversations are resolved without a person, as a long-run average. Your number will depend on your content.
The top questions. They tell you which pages shoppers cannot find on their own.
The questions it could not answer. Wonderchat collects this list in the review inbox. It is the most useful report in the product, because each line is a page you have not written yet.
Then go back to Day 1 and fill the gaps.
Troubleshooting the three common problems
The agent gives a wrong answer
Click the source on the reply. More often than not, the agent quoted a page correctly and the page itself is wrong, outdated or ambiguous. Fix the page, remove the archived version, and re-sync. If there is no source, the platform let the model answer from general knowledge, and that is the setting to change.
Shoppers get frustrated when it cannot help
Make the way out obvious. A shopper who types "talk to a person" should reach one, or at least a ticket with a promised reply time, without arguing with the agent first. Trapping people in a bot loop costs more sales than the agent will ever make.
Nobody uses it
Do not wait for the shopper to start. Use the proactive prompts from Day 3 on product pages, the cart and the shipping page, where questions cluster. And check privacy before you scale: Wonderchat holds SOC 2 Type II, follows GDPR and is hosted in the EU, in Frankfurt, which matters if you sell to European shoppers.
What it costs
Wonderchat's Free plan includes 50 AI replies a month, refilled monthly, with no credit card, which is enough to test on a single product page. Paid plans start with Starter, which includes 500 AI replies a month. Basic, with 3,000 AI replies a month, adds live chat, HubSpot sync, Shopify order tracking and an analytics dashboard, which most stores will want once the agent is live. Scale, with 9,000 AI replies a month, adds page-aware answers, weekly auto-recrawl and hot, warm or cold lead scoring. There is no per-seat pricing. Full details are on the pricing page, and the guide to chatbot pricing compares how other vendors meter their bills.
When another tool is the better choice
If most of your inbox is actions on orders rather than questions, such as "cancel my order," "change my address" or "refund this item," a help desk built around Shopify order data, such as Gorgias, is the better first purchase. It lives in the order record, and that is where those requests get done. Wonderchat's strength is the answer drawn from your content, with the source attached, and a clean handover when the request needs a person. Large retailers that want the agent to complete multi-step transactions on its own, such as processing an exchange end to end, should look at Sierra or Decagon, which are built for that and sold to enterprise teams. The roundup of the best AI chatbots for Shopify stores compares the store-native options side by side.
Verdict
Build the agent on your content, not on the vendor's promise. If your product pages, size guides and policies answer the questions your inbox gets, any decent platform will look good; if they do not, none will. Pick a platform that shows the source on every reply, says so when it does not know, and keeps the list of what it missed. Then spend the five days on the content, not the widget color. For a store whose problem is questions before the sale, that is the whole job. For a store whose problem is order changes after the sale, start with the help desk.
Frequently asked questions
What is an ecommerce AI agent?
It is software on your store that answers shoppers' questions from your own product pages and policies, suggests products, captures emails and hands hard cases to your team. It should not make up answers, and it should hand refunds and order changes to a person.
How does an AI shopping assistant increase sales?
It answers the pre-sale question at the moment of doubt, such as sizing, delivery dates or compatibility, instead of leaving the shopper to email and wait. It also suggests the right product and collects an email from shoppers who are not ready to buy yet.
How do you stop the agent from giving wrong answers?
Use a platform that answers only from your content, shows the source on every reply, and says so when the content does not cover a question. Then remove outdated pages, because an archived policy quoted correctly is still the wrong answer.
How long does it take to build one?
Most teams are live the same day: connect the store URL, upload policies and catalogs, paste the widget. Spending a week on testing, workflows and filling content gaps is what makes the agent good rather than merely live.
Will an AI shopping assistant replace the support team?
No. It takes the repeat questions about sizing, shipping and returns so your team handles exceptions, complaints and order changes. The handover to a person is part of the design, not a failure of it.
