Ecommerce customer service is the work of getting online shoppers what they need before and after they buy: an answer about a product, the status of an order, or a change to it, such as a return, refund, exchange, cancellation or replacement. It is done well when that outcome actually happens (the refund is submitted, the courier has the corrected address, the replacement is created) with as few contacts as possible, not when a reply is sent.
That is also what people mean by ecommerce customer support. This guide is for the person who runs it at a consumer brand with a high volume of order, delivery and return work across WhatsApp, Instagram, email and phone. It breaks the work into jobs, shows what each one needs, and sets out how to split it between your team and AI.
In short
- Ecommerce support is a set of jobs about orders. The facts each job needs live in your store, warehouse, courier and payment systems, not in the support tool.
- A job is done when the outcome is recorded in the system that owns it, not when someone replies.
- Split the work three ways: answering, acting under a rule, and owning exceptions. AI can take the first two where your rules are clear; people own the third.
- Measure per order and per job: contacts per 100 orders, time to remedy and repeat contacts, alongside response time.
What makes ecommerce customer service different
Most service advice is written for any business. Four things change when the customer bought online:
- Almost every contact is about an order. The customer wants something to happen to a specific order, not general help.
- The facts sit in other systems. Order state lives in the store or order management system, stock and packing in the warehouse, the latest scan with the courier, and the payment and refund record with the payment gateway. Support has to read them, and often write to them.
- The work often finishes outside support. The courier re-attempts the delivery, the warehouse ships the replacement, the bank settles the refund. Support can start all of these and still has to follow them through.
- Volume follows orders, and much of it is avoidable. Contacts rise with every sale event, and many of them come from customers who weren't told something: that the order was delayed, or that a delivery attempt failed.
The jobs, and what "done" means for each
Here are the jobs that make up most ecommerce support, with what each needs and where it usually gets stuck. The right answers vary by brand; the questions don't.
| Job | What the customer wants | What you have to check, and where | Done when | Where it gets stuck |
|---|---|---|---|---|
| Pre-purchase questions | The right product, size or model; whether it delivers to them and how they can pay | Catalogue, stock, delivery area, payment options | They have an answer before they leave the page | Answers that live in one person's head |
| Order status ("where is my order?", or WISMO) | Where the order is and when it will arrive | Order and fulfilment state in the store or order system; latest courier scan against the promised date | They know the status, and any delay is already being handled | Tracking that hasn't updated; promising a date nobody can source |
| Delivery problems (delay, failed attempt, return to origin) | The parcel delivered, or a clear next step | The courier's reason for the failed attempt, attempt count, address and phone number | The courier has corrected details and a reattempt is accepted, or the order is redispatched or refunded | The courier's own rules and systems |
| Order changes and cancellations | A new address or item, or no order at all | Whether the order is packed, dispatched or paid | The change or cancellation is made in the order system, or the customer knows why not and what happens instead | Orders already handed to the courier |
| Cash-on-delivery (COD) confirmation | For the brand: to know the order is wanted before it ships | The order, the address and the customer's reply | Confirmed, cancelled or switched to prepaid before dispatch | Customers who don't reply |
| Returns and exchanges | To send it back or swap it | The return window and exclusions for that item, the reason, a photo if your policy needs one, exchange stock | The return or exchange is created in the returns system and pickup is booked | Policy exceptions; pickup coverage; stock for the exchange |
| Refunds | Their money back, and proof it's on the way | The original payment and any refund already made | The refund is submitted against the original payment (or store credit issued) and the reference shared | Settlement time sits with the bank; duplicate refunds |
| Damaged, missing or wrong items | A product that works, or their money back | The delivered item, photo evidence of the damage and label, past claims on the account | A replacement or refund is created, or a person decides with the evidence already attached | Unclear photos; suspected abuse; warehouse investigations |
Brands that sell appliances, furniture or electronics usually have a ninth: warranty, installation and repairs, which bring a technician into the chain.
Three things the table makes visible:
- Most jobs end in an action, not an answer. A cancellation that is explained but not made is still open.
- Each job has its own evidence. A missing item needs different proof from a damaged one. Asking every customer for "photos and the order number" slows the easy cases and still misses what the hard ones need.
- Some jobs can't be closed by support alone. A refund can be submitted today and settle days later; a reattempt can be booked and still fail. Those need an owner until the other side finishes.
One order, followed end to end
A customer orders an air fryer on cash on delivery. Dispatch slips by two days, the first delivery attempt fails because her phone number has a typo, and when it arrives the basket is cracked. Here is the same order handled two ways.
| What happens | Reply-driven support | Resolution-driven support |
|---|---|---|
| Order placed (COD) | Nothing is sent. She isn't sure it went through. | A WhatsApp message asks her to confirm the COD order. She taps Confirm. |
| Dispatch is delayed | She asks "where is my order?" on WhatsApp, then again on Instagram two days later. Two agents check the store admin and both reply "processing". | She gets a message with the new dispatch date before she thinks to ask. |
| Delivery attempt fails | The courier marks her unreachable. She calls to chase; the phone agent emails the logistics team and closes the call. | She is told the attempt failed, corrects her number in the chat and picks a reattempt day. The courier gets the correction; she is told the reattempt is booked. |
| Arrives damaged | She emails photos. The agent asks for her order number, then emails the warehouse for a replacement and closes the ticket. | She replies to the delivery message with a photo. It is matched to the delivered item, checked against the replacement policy, and she confirms a replacement, which is created in the warehouse system. |
| Replacement ships | A week later she messages to ask where it is. Nobody has an open ticket to check. | She gets the shipping update, and a rating request after delivery. |
| Contacts she had to start | Five, on four channels, each a new ticket | One: the damage report |
The products, courier and policy are identical. What changed is that the brand told her things before she asked, each job had the data and permission to finish, and the replacement stayed owned until it shipped.
Channels: which job goes where
Start from the channels your customers already use, then decide what each is for. For many consumer brands that starts with WhatsApp for customer service, then Instagram, email, live chat and phone. Keeping them on one conversation per issue is what omnichannel customer service means; the table covers what each channel is good at.
| Channel | Good for | Watch for |
|---|---|---|
| Order updates, photos and documents, confirmations, and actions the customer approves in the chat | Outside 24 hours from the customer's last message, WhatsApp only lets you send pre-approved templates, so delay and delivery alerts need templates | |
| Instagram and Facebook Messenger | Questions before purchase; complaints raised in public or in DMs | Moving an order issue into a private thread without losing its history |
| Long issues, invoices and documents, formal complaints | Slow back-and-forth; one issue splitting into several threads | |
| Live chat | Questions while someone is browsing or checking out | Answers have to arrive while they are still on the page, from the catalogue and delivery data |
| Phone | Urgent or emotional issues, high-value orders, customers who won't type | Call outcomes that never reach the ticket |
| SMS | Delivery notifications in markets where customers expect them | Opt-in and sender rules vary by country; check them before sending updates |
| Order-tracking page and help centre | Status checks and policy questions the customer can answer alone | Saying something different from what agents and the AI say |
The cheapest contact is the one you prevent. Send your own messages at the moments customers otherwise chase: order confirmed, COD confirmation, shipped, delayed (with the new date), out for delivery, delivery failed (with a way to fix it) and delivered. On WhatsApp, post-purchase automation also lets the customer reply to the same message to act on it, such as correcting a phone number after a failed delivery.
Who does what: answer, act, own
The question for a support leader isn't "bot or human". It is which kind of work each job contains. There are three.
| Kind of work | Examples | Best done by | What it needs |
|---|---|---|---|
| Answer | Product and policy questions; order status | An AI agent, from your knowledge and live order data | Accurate, current policy and product knowledge; read access to order and courier data |
| Act | Cancel an eligible order; change an address before dispatch; create a return; submit an eligible refund; book a reattempt; create a replacement | An AI agent under rules that must pass, or a human agent from the ticket | Written rules (eligibility, required evidence, approvals); write access to the system that owns the record; customer confirmation before sensitive actions |
| Own | Policy exceptions, unclear evidence, repeat claims, angry customers, anything another team must finish | A named human owner | A ticket that stays open until the other team is done, with a deadline |
An AI agent for customer service earns its place on the first two rows, and only when it follows the same rules your team does. Use this to decide what to give it first:
| Give it to the AI first when… | Keep it with people when… |
|---|---|
| The job is frequent | The decision needs judgement your rules don't capture, such as a goodwill gesture |
| Eligibility can be decided from data and a written rule | The evidence needs a human eye |
| The system that owns the record accepts the action through an API | The customer is angry or threatening to go public |
| A wrong outcome is small or reversible | The value or risk is high, such as repeat claims on one account |
| The customer can confirm the outcome in the conversation | The fix depends on another team |
Order status, address changes before dispatch, eligible cancellations and returns inside the window usually qualify first. Goodwill refunds and fraud calls usually don't. Whatever the AI can't finish should reach a person with the checks already done: the order, the evidence, the rule that failed and what's left to do. The broader case for customer service automation is the same: automate the work whose rules you can write down.
The team around it
With AI on answering and routine actions, the human team changes shape rather than disappearing:
- Front line by skill, not by channel. Group agents by the jobs they resolve (orders and delivery; returns, refunds and exchanges; escalations and social), so a return reaches the returns team whether it arrived on email or WhatsApp.
- Named owners in other teams. Someone in the warehouse, logistics and finance receives support's requests with a deadline, and the customer's ticket stays open until they finish.
- Quality and improvement. Someone reviews AI and human conversations every week and turns repeated failures into a rule, knowledge or integration change. AI quality assurance makes it practical to review every conversation instead of a sample.
- One policy owner. The head of support decides the rules the AI and the team both follow, so customers get the same outcome wherever they ask.
How to measure ecommerce customer support
First response time and SLA compliance still matter, but they measure replies. These measure outcomes, per order and per job:
| Measure | How to calculate it | What it tells you |
|---|---|---|
| Contacts per 100 orders, by job | Customer-started conversations about a job ÷ orders shipped in the period × 100 | Where your volume comes from, and whether a fix worked |
| Avoidable-contact share | Order-status contacts where the order was on time and no update was due ÷ all order-status contacts | How much volume your own silence creates |
| Time to remedy | From first contact to the remedy being recorded in the owning system (refund submitted, return or replacement created, reattempt accepted) | Whether customers get outcomes, not just replies |
| Repeat contact per order | Orders with another contact about the same issue within 7 days, on any channel ÷ orders with a contact | Whether the first resolution held |
| Action completion | Actions confirmed by the owning system ÷ actions attempted; then follow through (refunds settled, reattempts delivered) | Where integrations or other teams fail after support did its part |
| Resolved without a person | Conversations closed with no human involvement and no repeat contact on that order within 7 days ÷ all conversations for that job | Automation that holds up, job by job |
| CSAT by job, AI vs human | Post-resolution rating, split by job and by who resolved it | Which jobs frustrate customers, whatever the speed |
Set the 7-day window to suit your delivery times; what matters is keeping it the same over time. For the standard measures and how to calculate them, see our guide to customer service metrics.
Five signs your team is answering, not resolving
Pull last month's tickets and check:
- Order status is your biggest contact reason, and most of those orders were on time. Customers are chasing because nobody told them.
- Agents open three tools to answer one ticket. Store admin, courier portal and payment dashboard, then copy details into a reply.
- Refunds and replacements are "raised" and the ticket is closed. The request goes to another team by email or spreadsheet, and nobody owns it until it's done.
- One order produces contacts on several channels. The same customer, the same order, three tickets.
- Your bot answers policy questions but hands every return, cancellation or refund to a person. It explains the rules without being allowed to apply them.
Two or more usually means the team is fast at replies and slow at outcomes.
How to improve ecommerce customer support, in order
- Tag a month of contacts by job and calculate contacts per 100 orders for each. This is your baseline and your priority list.
- Remove avoidable contacts first. Send delay, failed-delivery and delivery messages, and make sure order status gets the same answer on every channel.
- Write each job's policy as rules. Eligibility, required evidence, approval limits and exceptions, in one place used by agents and AI alike.
- Connect the systems each job needs. Store, order management or warehouse, courier and payment gateway, so both AI and agents can check and act without leaving the conversation.
- Automate the answer and act work with clear rules, starting with the most frequent jobs, and make every handoff carry the checks already done.
- Give exceptions an owner and a clock. Track time to remedy and repeat contacts weekly, and re-forecast contacts by job before each sale event.
How Flowcall runs ecommerce customer support
Flowcall combines an AI Agent with an AI Helpdesk for these jobs. Flowcall's AI Agent turns your SOPs for returns, refunds, exchanges, cancellations, COD and damaged products into workflows it follows step by step, on WhatsApp, Instagram, Messenger, live chat and email. Your rules decide what it may do: required checks, eligibility conditions and approvals must pass before a refund, cancellation or replacement happens. It checks current order, delivery, payment and refund data in your connected systems (Shopify natively, and delivery tracking from logistics tools such as Shiprocket, Delhivery and ClickPost), then completes the work: refunds, cancellations, exchanges, address changes and replacements. Customers can send a photo of the damage in the conversation, and the AI assesses it and uses the result in the workflow. When it is waiting on a photo or an address, it sends reminders and picks up where it left off.
On WhatsApp, Flowcall can send order confirmed, shipped, delayed and delivered messages automatically, and confirm COD orders, letting customers accept or cancel and offering an incentive to switch to prepaid (for Shopify stores).
When a person is needed, the AI creates a ticket with the conversation, a summary, the customer and order details and what's left to do, in one inbox for every channel. When a customer switches channel about the same issue, it stays on one ticket with its history. Agents see orders, delivery, payments and refunds on the ticket, process a refund or replacement from it, and open child tickets for the warehouse, logistics or finance while the customer's ticket stays open and owned. Automated QA reviews AI and human conversations, and the requests the AI couldn't resolve are grouped into the workflows, knowledge or integrations worth adding next.
Where it stops. Flowcall doesn't support SMS or RCS. It isn't a phone carrier or dialer: calls come through providers such as Exotel, Ameyo, Ozonetel or Contaque. Its knowledge base powers the AI and isn't a public help centre. A successful action records the immediate result; it doesn't prove the courier, warehouse or bank has finished its part. And Flowcall can't fix a late courier or a warehouse backlog: it can tell customers, act within your rules and keep the dependency owned until it's resolved.
If you'd like to see one of your own jobs, returns for example, running on your channels and order data, book a demo.
Related guides
- By job: order tracking automation · returns, refunds and cancellations · COD confirmation automation · warranty and after-sales support
- Choosing tools: ecommerce help desk software · AI customer service for ecommerce · conversational AI for ecommerce · Flowcall for Shopify
- Comparisons: Flowcall vs Freshdesk · Flowcall vs Zendesk · Gorgias alternatives
- By industry: appliances customer support · beauty brand customer support · apparel customer support · food and beverage customer support · health and wellness customer support · luggage and accessories customer support
Frequently Asked Questions
What is ecommerce customer service?
Ecommerce customer service is the help an online store gives shoppers before and after they buy: answering product questions, tracking orders, fixing delivery problems, and handling changes, cancellations, returns, refunds and replacements. It is the same thing as ecommerce customer support. It is done well when the outcome happens in the order, courier or payment system, not when a reply is sent.
What support services are available for ecommerce?
The usual set is pre-purchase help, order tracking, delivery-problem resolution, order changes and cancellations, returns and exchanges, refunds, damaged or wrong item claims and, for durable goods, warranty and repairs. A brand can deliver them with an in-house team, an outsourced team, an AI agent, or a mix, usually across WhatsApp, social DMs, email, live chat and phone.
What does an ecommerce customer support agent do?
An agent answers customers, checks the order, courier and payment systems, and completes what the policy allows: a return, a refund, a replacement or an address change. When another team has to act, they hand the work over with the details and keep the customer's issue open until it's done. Where AI handles the routine jobs, agents spend more of their time on exceptions and upset customers.
What are the most common ecommerce customer service issues?
Order status, delivery problems, returns and exchanges, refunds, cancellations, and damaged, missing or wrong items, plus pre-purchase questions. Which is largest depends on your category, delivery network and payment mix, so tag a month of contacts to find your own order.
How big should an ecommerce customer support team be?
Work it out from your own numbers: orders per month × contacts per order × the share that needs a person × average handle time. For example, 100,000 orders at 0.2 contacts per order is 20,000 contacts; if 40% need a person at 8 minutes each, that is about 1,070 hours of handling a month, before you add cover for shifts, breaks and sale peaks. These figures are illustrative; your own contact rate and automation share change the answer more than anything else.
Should you outsource ecommerce customer support?
Outsourcing adds people and hours, which helps with 24/7 cover and sale peaks. It doesn't add system access, written rules or ownership of exceptions, so the jobs still need the same checks and actions. If you outsource, keep the policy, quality review and escalations in-house.
What is the 80/20 rule in ecommerce?
The 80/20 rule (the Pareto principle) is the observation that a small share of causes produces most of an effect, such as a few products driving most sales. In support, the useful question is whether a handful of contact reasons make up most of your volume. Tag a month of contacts by job to find out, and fix the biggest first.
