How to route order enquiries in e-commerce support calls

The short answer

Most calls to an e-commerce support line come down to three questions: "where is my order?", "when will it arrive?" and "how do I return it?". On these calls an AI voice agent can do two things: answer policy questions from an approved source, and answer order-specific questions — after identifying the customer — either with data from the system or by collecting the details and handing over to staff.

The core principle: the agent does not guess an order status or a return condition. Status comes from the system, rules from the published policy. If neither is available, the agent says so honestly and hands the call to a person. This article shows the flow by call type, the identification rule and the boundary for returns calls.

Types of e-commerce support calls

• Order status and delivery time
• Changing the delivery address or time
• Returns and exchanges
• Payment: money taken but no order visible; when will the refund arrive
• Product questions: size, compatibility, stock
• Damaged or wrong item — complaint

The first three types are usually most of the calls, and many repeat in the same form. Payment and complaint calls are more sensitive and need a person.

Identification: finding the order

To talk about a specific order, the agent has to find the customer and the order. The simplest route: if the calling number matches the number on the order, the order is found; if not, the order number is asked for. Before revealing personal data, a second check is recommended — for example, confirming the name on the order or the delivery city.

Status and delivery calls

If the agent is connected to the order system or the courier's tracking data, it can give the status: "Your order was handed to the courier yesterday and is expected today between 14:00 and 18:00." If not, the agent must not make up a status; it should tell the customer where the tracking link is, or hand the enquiry to staff.

If delivery is late, the agent does not guess why. If the system records a reason for the delay, it says so; if not, it records the customer's enquiry and hands it to the responsible member of staff.

Address and time changes

Whether an address can be changed depends on the order stage: possible if it has not left the warehouse, difficult if it is with the courier. The agent can state the rule, but only makes the change if the system allows it. Otherwise the request is recorded and the customer is told honestly: "I've noted your request; a colleague will check whether it's possible and let you know."

Returns calls: the rule comes from the source

The return window, the conditions and who pays for return shipping must be written in the published policy, and the agent should answer only from that text. If the policy differs by category (hygiene products cannot be returned, for example), the difference should be written separately in the knowledge base.

1. 1State the ruleWindow, conditions, process — from the published policy.
2. 2Check eligibilityOrder date, whether the item was opened, the packaging.
3. 3Next stepA link, a branch or a member of staff to start the return.
4. 4ExceptionA request outside the rules — to staff; the decision is a person's.

Product and stock questions

"Is this shirt available in size M?", "will this adapter work with my phone?" — these are the calls closest to a sale. The agent can give the product description and size chart from the catalogue; stock information is only accurate in real time when there is a system connection. On compatibility questions, if the agent is not sure it should say so — a wrong "it's compatible" later turns into a return.

Payments and complaints: to a person

"Money has left my card but there's no order", "the refund hasn't arrived", "a damaged item arrived" — on these calls the customer is worried and the issue involves money. The agent collects brief details (order, date, problem) and hands over to staff. Card details are not asked for. On a complaint the agent promises no compensation.

Sale days

For a few days after a big sale, status and delivery calls jump sharply. This is when the agent adds the most value — the same question arrives hundreds of times. Before the sale, add a sale-specific answer about delivery times to the knowledge base and test the scenario: honest information such as "sale orders may take longer than usual to arrive" sets the customer's expectations correctly.

Illustrative example: a clothing store

Not a real customer case. More than half the calls to an online clothing store are about status and returns. The store launched the agent in two stages: first only the rules (return window, size exchange process, delivery times) with no system integration; then, with a connection to the order system, status questions.

In the first stage the agent tells a customer asking about status that "the tracking link was sent by text message" and, if they want, passes the enquiry to staff with the order number. In the second stage the agent gives the status itself and flags delayed orders to staff separately.

What to measure

• The split by call type and the share the agent completed
• The share of orders found — how well identification works
• Returns answers that match policy — via sample checks
• Waiting time and abandonment on sale days
• Repeat calls about the same order — the answer was not enough

Common mistakes

• Guessing a status without system data
• Stating return conditions in a vague sentence
• Not reading the order number back
• Revealing address and name after weak identification
• Promising compensation for a payment problem

Limits

A real-time status answer needs integration with order and delivery systems. Consumer rights and return rules are regulated by country; check your policy's legal compliance with a lawyer. Digit errors are possible in speech recognition, so the order number must always be confirmed.

An e-commerce line with Vexvon

In Vexvon AI Call Center return, delivery and sizing rules are written once in the knowledge base shared with the chatbot — the same answer works in calls and chats. One of the examples on the page is a store: the agent answers "how much is this model, is it in stock?" from the catalogue. To get status from the order system, you can give the agent your own API as a tool. After the call, the order number, question and outcome fields are written to the customer card, and a difficult call is passed to a live member of staff.

The same question in chat is covered in order status chatbot, and multilingual calls in the multilingual voice agent.

First step

Read your returns policy and turn it into a form that can be spoken on the phone: three or four short sentences. The agent's first stage starts with exactly those sentences. More articles are in the inbound calls by industry section; to discuss a scenario for your e-commerce line, get in touch.