AI customer service for e-commerce: order status, returns & refunds

AI customer service for e-commerce handles the questions that fill your support line every day: where is my order, how do I return this and when do I get my refund. An AI voice agent answers these by phone, and an AI WhatsApp agent answers them in chat, at any hour. Your team keeps the unusual cases that genuinely need a person.

Here is a practical look at what to automate, what to leave alone and how to roll it out without upsetting customers.

The same few questions, all day

Open any online store's support inbox and you will find a pattern. A large share of contacts are the same handful of questions, asked slightly differently. They feel urgent to the customer and routine to you.

• Where is my order, and when will it arrive?
• Can I change the delivery address or time?
• How do I return or exchange an item?
• Has my refund been issued, and how long will it take?
• Is this product in stock in my size or colour?

These are ideal for automation because the answer depends on clear rules and your order data, not on judgment.

Order tracking calls: answer in seconds

Order tracking calls are the biggest drain on many support teams. The customer wants a status, the agent looks it up and reads it out. A voice agent can do the same conversation without a queue, in the customer's own language and dialect.

When something is genuinely wrong, such as a delayed shipment or a missing item, the agent should not improvise. It should collect the details and pass the case to your team with the full context, so the customer explains the problem only once.

Returns and refunds without the back-and-forth

Returns policies are usually simple on paper and confusing in practice. The agent can explain your return window, which items qualify and what the next step is, then confirm what happens after the item comes back.

Be careful with money decisions. Let the agent explain policy and log the request, but keep exceptions, goodwill refunds and disputes with your team. That keeps tone and cost under your control.

WhatsApp customer support alongside the phone

In Egypt and the Gulf, many customers would rather message than call. A WhatsApp agent can answer the same questions in seconds, using the same information as the phone agent, so the answers match across channels.

It also helps after a call: Kalemna can send the tracking details or return instructions on WhatsApp the moment the call ends, so the customer has them in writing.

What to keep human

Damaged goods, suspected fraud, high-value orders, repeated failed deliveries and upset long-term customers deserve a real person. The agent's job is to recognise these early, collect the basics and pass them on with the full context.

A good rule of thumb: if the answer needs judgment or a decision about money, a person decides. If the answer is a fact or a policy, the agent can give it. Writing this rule down before launch saves a lot of arguments later.

How to roll it out

• Pull a week of support tickets and calls, and list the top ten questions.
• Write the real answer to each, including edge cases like cash on delivery or exchange-only items.
• Mark what the agent can do alone and what must go to a person.
• Test with your own staff calling as difficult customers.
• Launch, then read transcripts weekly and fill the gaps you find.

Mistakes to avoid

The common mistakes are feeding the agent out-of-date policies, hiding the human option and ignoring peak periods such as sales events and Ramadan shopping, when call volume jumps. Make it easy for customers to reach a person, and review the questions the agent could not answer.

To measure it, track how many contacts are resolved without your team, how many are handed over and what customers ask most. Kalemna's call insights show the most common questions and the busiest hours, which helps you plan staffing and fix confusing product pages.

Frequently asked questions