All posts
AI Employee#AI for ecommerce#AI ecommerce#WISMO

AI for ecommerce: what an AI employee does for an online store, and what it should leave alone

A plain guide to AI in ecommerce: where it actually saves a store time (WISMO, returns, stock, product copy, campaigns), what Shopify Sidekick and agentic commerce cover, and which decisions should stay with a person.

Md Shohel· October 18, 2026· 13 min read
Listen to this post0:00 / 15:49
AI narration of this post, read in full

Run an online store for a few months and you learn that the work is not the work you expected. You thought it would be products, photos and marketing. Most days it is the inbox. Where is my order. Can I swap this for a medium. The tracking link says delivered and it is not here. Do you ship to Canada. The same handful of questions, each with an order number attached, all day.

That is the honest starting point for AI in ecommerce. The flashy demos are about AI shopping agents and chat checkouts, and some of that is real now. But the hours most store owners lose are in the ordinary jobs around the sale, and that is where an AI that can do work, rather than just talk, earns its keep.

This post covers what AI for ecommerce looks like in practice, the jobs worth handing over first, how the AI built into your store platform differs from an AI employee, what agentic commerce actually means for a store today, and the decisions that should never leave a person's hands. We make Sanaf, an AI employee, so we have a stake in the answer. The product appears near the end, and most of what comes before it would be true without us.

The short answer

AI for ecommerce is software that does the repetitive work of running an online store: answering order-status and shipping questions from the real order data, handling routine returns and exchanges inside your policy, watching stock and drafting reorders, writing and updating product descriptions, and drafting email and social campaigns. The useful kind works inside the apps you already use, such as your store platform, help desk and email tool, and asks a person before anything that costs money or cannot be undone, such as a refund.

How AI is used in ecommerce today

When people ask how AI is used in ecommerce, they usually get one of two answers. The first is the shopper side: recommendations, search, chat on the product page, and now AI shopping agents that find and buy things for people. The second is the operator side: the work behind the store that a person used to do by hand.

Both matter, but they are different problems with different buyers. The shopper side mostly belongs to your platform. Shopify, for example, now puts eligible stores in front of AI channels automatically, which we come back to below. The operator side belongs to you, and it is where a small team feels the difference on a Tuesday afternoon.

Here is the operator side, job by job.

1. WISMO: the "where is my order" inbox

WISMO is the support shorthand for "where is my order?", and in most stores it is the question that never stops. It is easy to answer and slow to answer at the same time. Each one means opening the help desk, finding the order in the store admin, checking the carrier, then writing back. Three minutes each, forty times a day.

This is the clearest win for AI in ecommerce customer service, with one condition: the AI must read the real order. A chatbot that answers from a generic script ("orders usually arrive in 5 to 7 days") makes things worse, because the customer whose parcel is stuck gets a confident wrong answer. An AI that looks up the order, reads the tracking status and replies with the actual state of that parcel is doing the job a person did. It should also notice the ones that are genuinely late and hand those to you as a short list, because a late parcel is a customer you might lose.

2. Returns and exchanges

A routine return takes three or four messages: the request, the policy check, the label, the confirmation. Most of them are inside your policy and need no judgment at all. The size swap within 30 days, unworn, is a rule, and rules are what software is good at.

So the AI can carry the routine ones from request to label. What it should not do is decide the exceptions. The customer who is a day past the window, the damaged-goods claim with a blurry photo, the third return from the same address this month: those go to a person with the case already written up.

3. Stock watching and reorders

Bestsellers run out at the worst time, usually because nobody was looking at the number that week. An AI can check inventory levels on a schedule, compare them to the reorder point you set, and draft the purchase order for the items that are low. You approve it, and the order goes out. We wrote more about this job in what an AI inventory manager does, and the job page is AI inventory management.

4. Product descriptions and listing updates

This is where most people first try AI in ecommerce, through an AI product description generator. Those tools are fine at a first draft. The work that actually piles up is the maintenance: the twelve listings that still say the old material, the size chart that changed, the seasonal collection that needs new copy by Friday. Handing that over means giving the AI your notes and your tone, letting it update the listings, and checking a sample before they go live.

5. Campaigns

Most store email is a small set of shapes: the new arrival, the restock, the sale, the abandoned cart, the win-back. The AI can draft the week's campaign from what is actually new or back in stock, in the email tool you already use, and leave it for you to approve and schedule. The same goes for the social posts that go with it.

What these five have in common

None of them need a brilliant model. They need access to the right apps, rules about what may happen without asking, and a record of what was done. That is a different product from a chat widget, and the difference is the most useful thing to understand before you buy anything.

Shopify AI, Sidekick and the AI already in your platform

If you are on Shopify, you already have an AI assistant. Shopify's help center describes Sidekick as "an AI-enabled commerce assistant in your Shopify admin" that can analyse data, manage orders, edit products, write content such as product descriptions and blog posts, and build small custom apps. It also presents changes for your review before applying them. We checked that page on 18 October 2026.

For a lot of store work, that is genuinely useful and you should use it. Ask it for last month's sales by channel, or to set up a discount, and you save real time.

The limit is where it lives. Sidekick works inside the Shopify admin, for the person who opens the Shopify admin. Your support inbox might be Gorgias or Zendesk. Your email is Klaviyo. Your labels come from Shippo. Your team talks in Slack. The work of running the store crosses all of those, and an assistant inside one of them can only do the part that happens there.

So it helps to sort ecommerce AI tools into three kinds:

Built into the platformA chatbot on the storeAn AI employee
ExampleShopify SidekickA product-page chat widgetSanaf
Who it works forThe person in the adminThe shopperThe store's team
Where it worksOne platformYour storefrontAcross the apps you connect
Typical jobReports, discounts, product editsPre-sale questionsWISMO, returns, stock, copy, campaigns
Starts work on its ownNo, you ask in the adminWhen a shopper writesYes, on a schedule or when something arrives

None of these replaces the others. A store can sensibly run all three. The question is which work is eating your week, and which kind of tool can reach it.

What is agentic commerce, and does it matter for your store?

Agentic commerce is the name for AI agents that shop on a person's behalf: they search for products, compare them and, increasingly, complete the purchase inside an AI assistant. It is the shopper side again, and it is moving fast.

Two open standards sit underneath it. OpenAI and Stripe announced the Agentic Commerce Protocol in September 2025 for buying inside ChatGPT. Google and Shopify announced the Universal Commerce Protocol in January 2026 for Google's AI Mode and Gemini. Shopify's own agentic storefronts page says the feature lets customers "discover and purchase your products in AI channels", names ChatGPT, Google AI Mode, Gemini, Microsoft Copilot and Meta, and says it is "active by default for eligible stores". On ChatGPT, the same page says, the customer finishes on your own store's checkout in an in-app browser or a new tab. We checked it on 18 October 2026.

What this means for an ordinary store owner is mostly a to-do list about your data. AI shopping agents read your catalog the way a search engine does, so clean titles, complete product attributes, accurate stock and a clear returns policy matter more than they did. A store whose listings are out of date will be described badly by an agent that trusts them.

That is also where the two sides meet. An AI shopping agent can bring an order in at 2am. Someone still has to answer the customer's "where is it?" on day four, process the exchange and notice that the item is now out of stock. Agentic commerce adds orders. It does not remove the work around them.

The decisions an AI should never make for a store

Any AI that touches customers and money needs written limits. These are the ones worth insisting on, whatever you buy.

Refunds and chargebacks wait for a person. The AI can prepare the case, with the order, the policy, the customer's history and a recommendation, so the decision takes you ten seconds. It should not move money on its own.

Policy exceptions are yours. The policy is what you have decided in advance. Bending it is a judgment about a customer, and a customer should get that judgment from a person.

Angry customers and public complaints go to a human. An AI can draft a calm reply, but a review threat or a damaged-goods dispute needs someone who can make a call about the customer.

Supplier orders are drafts until you say otherwise. Reordering is low risk when the numbers are right and expensive when they are not. Approve them yourself until you trust the pattern, then decide whether to let small, routine ones through.

And every action leaves a record. If the AI answered forty customers overnight, you should be able to read what it said to each of them, in plain language, without opening a dashboard you never visit.

How to use AI in ecommerce without making a mess

If you are starting from nothing, this order works for most stores.

  1. Pick the one job that eats the most hours. For most stores that is WISMO and shipping questions. Measure it roughly first: how many a day, and how long each one takes.
  2. Connect only what that job needs. For order questions, that is the store platform, the help desk or inbox, and the shipping tool. Not your bank, not your ad accounts.
  3. Run it with approval on for the first week. Let the AI draft every reply and send nothing until you have read a few dozen. You will find the gaps in your own policy quickly.
  4. Turn on the routine cases. Once the drafts are consistently right, let it send the plain status replies on its own and keep approval on for anything involving money.
  5. Add the next job. Returns usually come second, then stock, then product copy and campaigns.

The mistake to avoid is the opposite order: a chatbot on the storefront on day one, answering from a script, with nobody reading what it says. That is how stores end up with screenshots of their own bot promising a refund they never offered.

Where Sanaf fits

Sanaf is an AI employee. It lives in your team's Slack or Microsoft Teams, and it works inside the apps you connect. For an online store, that typically means Shopify, a help desk such as Gorgias or Zendesk, Klaviyo for email, Shippo for labels and Stripe for payments, all of which are in the connection catalog.

Day to day, it does the jobs in the first half of this post. It answers order-status and shipping questions from the real order and tells you which ones are actually late. It handles routine returns and exchanges inside your policy and flags the exceptions. It watches stock and drafts reorders at your threshold for you to approve. It keeps product listings current from your notes and drafts the week's email and social campaigns. You can ask it things in a sentence, such as "answer the overnight order emails and tell me which ones are late", and it reports back in the same thread.

The limits are the ones above, and they are on by default. Refunds, chargebacks and policy exceptions wait for your approval, with the case prepared. Anything it cannot undo asks first. Every job leaves a work log anyone on the team can read. The online stores page lays out what it takes on and what it hands back, and hiring an AI employee shows what the first day looks like.

On price, you start with $120 of free credit and no card, and that free credit lasts 30 days. After that you buy credits from $20 when you want them, and credit you buy never expires. You pay for the work done, not per seat, so adding a teammate to the channel costs nothing. The pricing page has the details.

If the inbox is where your week goes, start there. Hand over the order-status questions, read what comes back for a week, and decide from what you see.

Questions people ask about AI for ecommerce

What is AI for ecommerce?

AI for ecommerce is software that helps run an online store: answering customer questions from order data, handling routine returns, watching stock, writing product descriptions, drafting campaigns, and on the shopper side, recommendations and AI shopping agents. The most useful kind for a store's team works inside its existing apps and asks before anything involving money.

How is AI used in ecommerce?

On the shopper side, AI powers product recommendations, on-site search, storefront chat and AI shopping agents that find and buy products. On the operator side, it answers order-status questions, processes routine returns, monitors inventory, drafts reorders, updates product listings and drafts marketing emails and social posts.

What does WISMO mean?

WISMO stands for "where is my order?" It is the support term for order-status and shipping questions, which are usually the most common message an online store receives. AI handles them well when it looks up the real order and tracking status instead of answering from a script.

What is Shopify Sidekick?

Shopify Sidekick is the AI assistant built into the Shopify admin. Shopify's help center says it can analyse store data, manage orders, edit products, write content such as product descriptions, and build simple custom apps, and that it shows changes for your review before applying them. It works inside Shopify rather than across your other apps.

What is agentic commerce?

Agentic commerce is shopping done by AI agents on a person's behalf: searching, comparing and completing purchases inside assistants such as ChatGPT, Gemini or Copilot. It runs on open standards including the Agentic Commerce Protocol (OpenAI and Stripe) and the Universal Commerce Protocol (Google and Shopify). For a store, the main job is keeping product data, stock and policies accurate.

Can an AI agent for ecommerce issue refunds?

It can, technically, if you give it that access, but it should not do so on its own. A sensible setup has the AI prepare the refund case, with the order, the policy and a recommendation, and wait for a person to approve it. Sanaf holds refunds, chargebacks and policy exceptions behind approval by default.

Is an AI chatbot the same as an AI employee for an online store?

No. A chatbot answers shoppers on your storefront, usually from a knowledge base. An AI employee works for your team across the apps you connect, takes on jobs such as the support inbox, returns and stock checks, starts work on a schedule, and keeps a record of what it did.

How much does AI for an online store cost?

It varies widely. Platform assistants are often included in the store plan, chat widgets usually charge per conversation or per month, and many AI tools charge per seat. Sanaf starts with $120 of free credit that lasts 30 days and no card, then credits from $20 that never expire once bought, priced on the work done rather than per seat.

Get the next one by email

We write when we have learned something worth passing on, so nothing arrives unless it is worth your time.

Unsubscribe in one click, any time. Privacy policy