Customer service for online shops: the complete 2026 guide

Suppose your Shopify shop does 400 orders a month and one in three orders produces an email. That is a good 130 support requests, and most of them are not complaints. They are questions: where is my parcel, can I change the address, does size M fit, how do I send this back. Each one takes a minute to answer. Together they are the reason you are still in your inbox at night instead of working on your range.

Customer service for an online shop is not a side task you squeeze in, and not a department you can only afford at 50 employees. It is the one moment a customer talks to you directly. According to Salesforce State of the Connected Customer (2022), 88 percent of customers say the experience a company provides is as important as its products. For a small shop that means: support is part of the product.

This guide is the starting point. It explains which requests come in, which channels you actually need, how fast you should reply, how to standardise answers, what automation can do today and where it has to stop. Every topic has its own deeper article. This is the map.

What customer service in an online shop actually has to do

In a physical store, a salesperson answers questions before the customer buys. In an online shop, the customer buys first and asks afterwards. That turns the job around. Most of your support happens after the purchase, in a window where the customer has already paid and has nothing in hand yet.

Three jobs are hidden in there:

• Removing uncertainty. The customer wants to know the order exists, is being processed and will arrive. Every gap in that chain produces an email.
• Solving problems. Wrong item, damaged parcel, return, refund. This is where it is decided whether the customer comes back.
• Providing information. Size, material, delivery time abroad, invoice. Much of it could be on the product page and isn't.

If you think of support as pure problem-solving, you miss the first two thirds. Most of the work does not come from mistakes but from missing information. And missing information is the part you can fix most easily.

The requests that actually come in

Open your inbox and sort the last 50 emails by topic. The result looks similar in almost every shop; only the order varies with the range:

1. Delivery status. “Where is my order?” in every variation. In support jargon, WISMO, and in most shops the biggest category. Why it happens and how to bring it down is in “Where is my order?” – reducing WISMO requests.
2. Returns and exchanges. How do I send it back, who pays the shipping, when do I get my money. The whole process is described in Returns process for online shops: step by step.
3. Changes after purchase. Wrong address, mixed-up item, cancel the order. Time-critical, because the window before dispatch is short.
4. Product questions before purchase. Size, compatibility, availability. The only requests that directly bring revenue.
5. Complaints. Damaged, wrong, late, disappointing. Few, but with the biggest effect on reviews.
6. Admin. Invoice, voucher not working, account, unsubscribe.

The full list with a standard reply for each type is in The 12 most common customer requests in e-commerce. The point of this exercise is not the list itself but the insight: about four of the six categories are recurring and predictable. Exactly those categories you can standardise, and later automate.

Channels: fewer is better

A small shop does not need to be reachable on seven channels. It needs to reply reliably on the channels it offers. An Instagram inbox nobody opens for three days is worse than no Instagram inbox.

Email is the foundation. It is asynchronous, documented, and the customer does not expect a reply in seconds. For most Shopify shops under 1,000 orders a month, email is the only channel that is really necessary, plus a contact form that lands in the same inbox.

Chat sounds modern and is a promise: whoever offers chat promises a reply within minutes. If you cannot keep that, leave it out. A chat that shows “We're not here right now, leave a message” is an email with a worse reputation.

Phone is rarely the customers' problem in online retail, but the merchant's. It interrupts, is not documented and cannot be standardised. If you offer it, offer it with fixed hours.

Social media and marketplaces are channels where customers address you whether you like it or not. One rule is enough here: check once a day, move anything substantive into the email process.

So the decision is not “which channels” but “which channels can I serve so that the response time holds”. Which brings us to response time.

Response time: what customers expect and what you should promise

The expectation is higher than most merchants think. In a survey by Toister Performance Solutions (2018) of more than 1,200 consumers, 89 percent said a reply to an email within one hour would meet their expectations. Reality looks different: the SuperOffice Customer Service Benchmark Report, which contacted 1,000 companies, found an average response time of 12 hours and 10 minutes, and 62 percent of companies did not reply at all.

That is the good news in it: the bar is low. If you reliably reply within a few hours, you are better than most.

What you can do concretely:

• Name a promise you keep. “We reply within 24 hours on working days” is honest and perfectly fine for most customers. A broken two-hour promise is worse.
• Sort by urgency, not by arrival. An address change before dispatch is more urgent than an invoice request. Sorting decides more about perceived response time than the total number of emails.
• Separate the first reply from the solution. A customer who hears after ten minutes “I've got your email, I'll check with the warehouse and get back to you by tomorrow noon” is happy to wait. The same customer without that message writes again after 24 hours, annoyed.

Why repeat contacts are so expensive is shown by the Kundenservice Barometer 2024 by Armonia: 40 percent of respondents in Germany had contacted customer service more than once about the same request. Each of those contacts is a ticket you pay for twice. More on this in Response time in customer service: what customers expect, what is realistic.

Standardise before you automate

The most common shortcut merchants take: they buy a tool before they know what they actually want to answer. The tool then does quickly what used to be done slowly and wrong.

The order that works:

1. Write down the ten most common requests and formulate a reply for each that you would actually send. Not “Dear valued customer”, but your tone.
2. Check each reply for completeness. Does it contain what the customer would ask next? A returns reply without the refund timeline is guaranteed to produce a second email.
3. Note the rules behind it. From when is there goodwill, who pays return shipping, what happens with “delivered but not received”. These rules are more valuable than the templates, because they are what a new employee or an AI agent has to learn.
4. Templates as a starting point, not the end product. A template sent without adjustment is noticed by the customer. One sentence that refers to their order is enough.

Ready-made texts for the most important cases are in Customer service email templates: 15 templates for online shops. Take them as blanks and rewrite them in your tone.

When it goes wrong: complaints and the review that follows

A few percent of orders go wrong. Parcel damaged, wrong item, delivery too late for the birthday. These few cases shape your reputation more than the many that run smoothly, because only they lead to reviews with text.

According to Zendesk CX Trends 2025, a survey of around 5,100 consumers in 22 countries, 63 percent are willing to switch to a competitor after a single bad experience. The experience that counts is rarely the mistake itself. It is the reaction to it.

What has to come first in a complaint reply, in this order: confirm you have understood the problem; say what you are doing now; say by when. Only then explanations, if at all. Customers who feel understood accept almost any solution. Customers who first get an explanation of why the mistake was not your fault accept none. The full process is in Complaint management for online shops: de-escalation in 5 steps.

The moment between complaint and review is short. If you visibly react in that time, you often get no bad review at all, or one that praises the support. How to recognise that moment is in Preventing negative reviews: the support moment before.

Returns as part of support

In many shops, the return is a separate process with its own form, and support only hears about it when the customer asks where their money is. That is the most expensive variant, because it makes the customer wait twice: once for the parcel, once for the refund.

Treat the return as a support case with fixed stations: registration, label, goods received, inspection, refund. At every station the customer gets a short message. Once that is set up, you save a large share of the “has my parcel arrived?” emails that otherwise follow every return. And you immediately see which returns fall outside the norm, for instance when the same customer sends back worn goods for the third time.

At Humane this runs through a returns portal for end customers, in which the merchant stores rules and approvals, for example which refunds run automatically and which they approve themselves. The details are in the returns guide linked above.

Automating: what works today and what doesn't

Precision pays off here, because “automation” means three very different things.

Autoresponders and text blocks. “We have received your request.” That is not a reply, it is a receipt. It does not reduce ticket volume, it postpones it.

Chatbots with a decision tree. “Choose: 1 delivery status, 2 return, 3 other.” The customer clicks through, ends at a link to the carrier or at “Please contact support”. Usable for simple FAQ questions, frustrating for anything order-related, because the bot does not know the order.

AI agents with access to order data. The agent reads the email, understands what it is about, pulls the order from Shopify, checks the shipment status with the carrier and writes a reply that fits this case. This is the category that actually reduces ticket volume, because the reply closes the ticket instead of passing it on.

This is how it works at Humane: the agent categorises every incoming email, pulls order and customer data from Shopify and the live status from the carrier, and writes a reply in the shop's tone. At the start, every reply lands as a draft in the dashboard. The merchant reads, corrects, approves. Every correction becomes a rule. When a category runs cleanly for weeks, the merchant switches on automatic sending for exactly that category, not for others. What the agent cannot assign with confidence stays a draft. What can be automated in practice and what cannot is the subject of the guide AI in customer service: what actually works in 2026.

Honestly, there are cases no system should take over:

• A complaint in which the customer is already writing for the second time.
• Anything that needs a goodwill decision outside your rules.
• Requests where the order cannot be clearly identified.
• Requests with a legal undertone: lawyer, consumer protection, police report.
• Anything you are seeing for the first time.

A good agent recognises these cases and puts them in front of you. A bad one answers them with a text block. That difference is the real buying decision, not the price.

A calculation with realistic assumptions

Suppose your shop gets 300 requests a month. Suppose you need four minutes per request on average, including looking things up in Shopify and with the carrier. That is 20 hours a month, half a working week, for replies that are mostly identical in content.

Suppose further that two thirds of those requests are delivery status, returns questions and standard changes that can be standardised with clean templates and rules. If you bring those 200 requests down to one minute per case (read, adjust template, send), you save 10 hours a month. No tool, just preparation.

If an agent prepares those 200 requests and you only approve, the effort per case drops to a few seconds. What remains are the 100 requests that actually need your attention, and for those you then have time. That is the real gain: not less support, but support in the right place.

For cost comparison: at Humane, the Starter plan includes 100 tickets, then €0.49 per ticket, all features in every plan. Whether that pays off for your shop depends mainly on what an hour costs you, and you can repeat the calculation above with your own numbers.

The tools you really need

For a Shopify shop under 1,000 orders a month, a short list is enough:

• One shared inbox where all requests land, whatever the channel. As soon as more than one person replies, you need an assignment of who has which case.
• Categories. Every email gets one: delivery, return, change, product question, complaint, admin. Without categories you don't know what to work on. How to set that up is in Tidying up the support inbox: categorising, prioritising, triage.
• Templates with rules, as described above.
• A tracking page in your own shop, so that delivery questions don't arise in the first place.
• A returns process with status emails.
• Optional: an agent that prepares or answers the recurring categories.

What you don't need: seven channels, a call centre, one tool per task. A comparison of the common options for Shopify is in Shopify customer service apps compared 2026.

How you measure progress

Four numbers are enough, and you can collect them without a tool if you count once a month:

1. Requests per 100 orders. The most important number. If it falls, you have closed information gaps. If it rises, a new gap has opened somewhere, often after a change of carrier or a new product category.
2. Time to first reply. Median, not average, otherwise one weekend distorts everything.
3. Share of requests resolved with one reply. Every second email about the same case is a sign that something was missing in the first reply.
4. Share of categories. If delivery questions make up half, you work on the shipping email. If returns questions dominate, on the returns process.

You don't need more numbers at the start. If these four are right, the rest usually is too.

What you can do now

Don't start with the tool, start with the inbox. Sort the last 50 emails by category and see which category is biggest. Fix the information source behind it: the shipping email, the returns page, the product description. Write a reply in your tone for the ten most common cases and note the rules you decide by. Name a response time you can keep for sure. And only when that is in place, think about which categories you want to hand off, to a helper, to a tool or to an agent that knows your rules.

If you want to see what your requests look like categorised and answered before you change anything: try it free for 14 days, no credit card, and in the first week everything stays a draft.