Ecommerce

Can you automate ecommerce customer support entirely? Most of it. Here is where the line sits, and how to build it

By Babar Lal  ·   ·  11 min read

It is 11pm and the founder of a clothing store is still in the inbox. Where is my order. Can I change my size. My discount code did not work. Do you ship to Canada. Where is my order, again.

Buried under them is one message that matters. A jacket arrived torn, and the customer sent a photo two days ago. Nobody has replied yet, because the other forty messages came first.

Most of those forty could have been answered without anyone touching them. The torn jacket is the one that needed a person.

The same support inbox two ways. On the left, the founder answers at 11pm: where is my order, can I change my size, my discount code failed, and one torn jacket waits two days. On the right, an AI agent that can see orders and policies answers the order status question with live tracking, swaps the size on an unshipped order, and passes the torn jacket to a person with photos and the order attached.

The short answer

You cannot automate all of it, and you should not try. You can automate most of the routine questions, so a person only sees the damaged parcels, the angry customers and anything outside your policy.

QuestionShort answer
Can it be fully automated?No. Routine questions can be; damage, disputes and exceptions need a person
What to automate firstWhere is my order, then returns, then simple order changes
What it needsRead access to orders, tracking, policies and products, plus clear limits on what it may change
Cost rangeShopify Inbox is free; AI agents from about $1 per resolved conversation; a custom build has a setup fee and a monthly fee
What to do firstSort last month's tickets into types and count each one

What customers actually ask

Every store is different, so count your own. Gorgias, a helpdesk vendor, says order status questions are 18% of requests on average in its data. It does not say how big the data set is or when it was collected. The same handful of types tend to fill the inbox.

Ticket typeCan a machine close it?What it needs
Where is my orderYesThe order, the tracking number and the carrier's latest scan
Return or exchange inside policyYesYour return rules and a returns portal or app
Change address, size or cancel before it shipsYes, if unshippedThe fulfilment status, and permission to edit unshipped orders
Product questions: size, material, stockYesYour catalog, size charts and live stock
Shipping times, countries, policiesYesYour written policies
Discount code did not workPartlyThe code's rules; a person decides on exceptions
Damaged, wrong or missing itemPartlyPhotos and the order collected automatically; a person decides
Refund outside policy, chargeback, complaintNoA person, with the whole thread and order attached
Three columns. Fully automated: where is my order, returns and exchanges inside policy, changes to unshipped orders, product, size and stock questions, and shipping and policy questions. Automated, then a person decides: discount code problems, and damaged, wrong or missing items, where the system collects photos and the order first. Always a person: refunds outside policy, chargebacks and disputes, complaints and angry customers, and anything the system is unsure about.

The three layers that do the work

Layer one is self-serve. A tracking page with real delivery dates, a returns portal and a clear FAQ answer many questions before they are asked. Our posts on where-is-my-order automation and returns automation cover these two in depth.

Layer two is an AI agent with access to your store. It answers in chat, email, WhatsApp or text. It reads the order before it replies, so it can say "your parcel was scanned in Memphis this morning" instead of "check your email". It can also do a few safe things, like changing the size on an order that has not shipped.

Layer three is a person. Everything the agent is unsure about, or is not allowed to do, lands in one queue with a short summary, the order and the conversation attached. Nobody has to ask the customer to repeat themselves.

How to set it up, step by step

  1. Count last month's tickets by type. Start with the biggest type the table above marks as yes.
  2. Write down your policies so a machine could follow them. Return window, shipping times, what counts as damaged, and who pays.
  3. Connect the agent to your orders, tracking, catalog and stock. Read access first.
  4. Decide what it may change. A good first list is: send tracking, start a return inside policy, and edit or cancel an order that has not shipped.
  5. Set the hand-off rules. Damage, refunds outside policy, chargebacks, angry messages and anything it is unsure about go to a person, with a summary.
  6. Say it is an assistant, and make "talk to a person" easy to find.
  7. Read a sample of conversations every week for the first month. Fix wrong answers in the policy text, not by patching the bot.
Six steps for one support message: a message arrives by chat, email, WhatsApp or text; the agent reads the order, tracking and policies; it sorts the message into a type; it answers routine questions from live data; it takes only allowed actions, such as editing an unshipped order or starting a return in policy; and anything else goes to a person with a summary and the order attached. The rule: the agent only says what your written policy says.

The rules

This is a plain summary, not legal advice.

You are responsible for what your bot says. In Moffatt v. Air Canada, 2024 BCCRT 149, a chatbot gave a customer the wrong refund policy. The tribunal said "It makes no difference whether the information comes from a static page or a chatbot," and ordered the airline to pay. Keep your agent to your written policy, and keep that policy up to date.

Shipping delays have rules in the US. The FTC's Mail, Internet, or Telephone Order Merchandise Rule says you need a reasonable basis for the shipping time you state, or 30 days if you state none. If you cannot ship on time, you must ask the customer to agree to the delay and tell them they can cancel for a full refund. For most payment methods, the refund is due within seven working days of the cancellation. If your agent answers "where is my order", it should follow those rules when an order is late.

Customers want a way out. In a Gartner survey of 5,728 customers in December 2023, 64% said they would prefer companies did not use AI for service. Their top worry was that it would get harder to reach a person. An agent that answers fast and hands off easily is the answer to that worry. One that traps people proves it right.

Your options compared

Prices come from each provider's own page, checked in October 2026.

OptionTypical costGood atTrade-offs
Shopify InboxFreeChat on your store, with an AI that learns from your catalog, policies and store contentLives inside Shopify; other channels and marketplaces sit elsewhere
Tidio with Lyro AIStarter $24.17 a month; Lyro from $32.50 a month for 50 AI conversationsSmall stores starting outAI conversations are capped by plan
GorgiasHelpdesk priced by tickets, not seats; AI Agent about $0.90 to $1 per resolved conversationEcommerce helpdesk built around Shopify ordersPer-resolution fees grow with volume
Intercom FinFrom $0.99 per Fin outcome; seats from $29 a monthA polished AI agent that can also sit on another helpdeskPer-outcome fees grow with volume
Custom buildSetup fee plus monthly fee; our builds start at $200One agent across Shopify, marketplaces, WhatsApp and your 3PL, with your own action limitsOnly worth it when an app cannot see all your channels or data

What owners and the industry are saying

I only list pages I could open. Reddit would not open for me. Vendors are labelled.

WhereWhoWhat they saidWhat it tells you
Shopify Community, Dec 2023A store owner"I really don't like AI chat bots...they don't actually solve the issue I have." Also: "A well organized FAQ page/section is much less frustrating."A bot that cannot see orders is worse than a good FAQ
Shopify Community, 2026Merchants and app makers"Most WISMO is an expectations problem."Fix the delivery estimate before you automate the answer
Intercom FinAI support vendorClaims a 76% average resolution rate across 12,000+ customersA vendor figure; your rate depends on your data and rules
Gartner, Jul 2024Research firm, 5,728 customers64% would prefer companies did not use AI for service; reaching a person is the top worryMake the hand-off obvious
BC Civil Resolution Tribunal, Feb 2024Tribunal decisionAir Canada was held responsible for its chatbot's wrong answerKeep the bot to written policy

A cost sum you can redo with your own numbers

Here is the sum for a store doing about 1,700 orders a month. Three of the numbers are my assumptions. Measure your own.

LineHow to get itExample
Tickets per monthYour helpdesk or inbox600 (assumed)
Minutes per ticketTime a sample6
Hourly costWage plus overheads$20
Support time now600 x 6 minutes = 60 hours, x $20$1,200 a month
Share the agent resolvesResolved by the agent, divided by all tickets60% (assumed), so 360
Agent cost360 x about $0.90 per resolutionAbout $324
Time saved360 x 6 minutes = 36 hours, x $20$720
Net per monthTime saved minus agent costAbout $396, plus replies in seconds instead of hours
Support automation on example numbers. Support time now: 600 tickets a month times 6 minutes times 20 dollars an hour is 1,200 dollars. Agent: 360 tickets resolved, which is 60 percent, times about 0.90 dollars is about 324 dollars, and saves 720 dollars of time. About 396 dollars a month better off, plus replies in seconds. The 600 tickets and 60 percent are marked as assumed.

The money is real but modest at this size. The bigger win is often speed. A customer who gets their tracking in ten seconds has less reason to open a chargeback, and the person on your team gets their evenings back.

When it backfires

No way to reach a person. A bot that loops "I can help with that" drives people to chargebacks and one-star reviews. Put the hand-off in every conversation.

A bot with no order access. If it can only say "check your email", it is a slower FAQ page. Connect it to orders and tracking, or do not launch it.

Making up policy. If the policy is not written down, the agent will fill the gap with something plausible. Moffatt v. Air Canada shows who pays for that.

Refunds with no limits. Let the agent start returns inside policy. Keep refunds outside policy, and anything above a set value, with a person.

Ignoring late orders. An agent that repeats a tracking link while an order is weeks late is not following the FTC's delay rules. Late orders should trigger a delay notice and the option to cancel.

What we build, and when we say no

If you get a few dozen messages a week on Shopify alone, turn on Shopify Inbox, write a proper FAQ and set up good shipping emails. You do not need us for that.

We build the next step for stores selling on several channels, or with an inbox that eats someone's day. One agent reads your orders, tracking, stock and policies across Shopify, Amazon, TikTok Shop and your 3PL. It answers on your site, email, WhatsApp and text, makes only the changes you allow, and passes everything else to a person with a summary. Our ecommerce automation work covers the whole order flow. Our AI automation service builds the agent, and our workflow automation service connects it to your systems. For the chat side on its own, see our post on AI website chat, and for messaging apps, WhatsApp and SMS support.

We say no when volume is low, when the policies are not written down yet, or when one app already sees every channel. New to all this? Our plain-English answers to 24 questions about AI automation is the place to start.

Questions people ask

Can I automate my store's customer support entirely? Not entirely, and you should not try. Order status, returns inside policy, simple order changes and product questions can be answered or done with no person involved. Damaged items, refunds outside policy, chargebacks and angry customers still need a person. The goal is that a person only sees those.

What should I automate first? Where is my order. It is usually the biggest single type of ticket, and the answer already sits in your store and the carrier's tracking. Gorgias, a helpdesk vendor, says order status questions make up 18% of requests on average in its data.

What does an AI support agent need to work well? Read access to your orders and tracking, your written policies, and your product catalog. Without order access it can only say check your email. It also needs clear limits on what it may change, and a fast way to pass a conversation to a person with the order attached.

Am I responsible for what my chatbot tells customers? Assume you are. In Moffatt v. Air Canada in 2024, a Canadian tribunal held the airline responsible for a wrong answer its chatbot gave, saying it makes no difference whether the information comes from a static page or a chatbot. Keep the bot to your written policies. This is not legal advice.

How much does ecommerce support automation cost? Shopify Inbox is free. In October 2026, Tidio listed its Lyro AI from $32.50 a month for 50 AI conversations, Gorgias charged about $0.90 to $1 per conversation its AI resolved, and Intercom listed Fin from $0.99 per outcome. A custom build has a setup fee and a monthly fee.

Will customers hate talking to a bot? Some will, if it blocks them. In a Gartner survey of 5,728 customers in December 2023, 64% said they would prefer companies did not use AI for service, and the top worry was that it would be harder to reach a person. Say it is an assistant and make the way to a person obvious.

Find out how much of your inbox a machine could close.

Bring last month's support tickets to the free audit. We sort them by type, count what could be closed without a person, check what your agent would need access to, and tell you whether Shopify Inbox, an app or a build is the right step. If a better FAQ will do, we say so.

Book the free audit →

Sources