After every discount campaign, an online store’s support team gets the same question in hundreds of forms: where is my order. Questions about returns and delivery follow. An AI customer support assistant takes on this repetitive part, by phone and in chat: it looks up the order in the store, says what stage it is at and explains the steps to follow.
Its first message says plainly that it is an automated assistant. It reads data, it does not make decisions: it does not approve refunds and does not grant compensation. When the conversation calls for a person, it hands it to a support agent together with a summary, so that nobody has to repeat everything from the beginning.
“Where’s my order?” and the questions that fill the working day
Support questions fall, broadly, into two groups. Some have an answer that already exists somewhere: on the order page, in the parcel tracking, in the returns policy. The others call for judgement: a product that arrived faulty, a lost parcel, an upset customer.
When the team is small, both groups wait in the same queue. The person with a missing parcel waits behind those who only wanted to know whether there are Saturday deliveries.
Order status, read from the store and from the courier
The assistant says nothing about an order until it knows who it is talking to. It asks for the order number and a second piece of information (the email address or phone number used for the order) and only then reads the data.
From the store it learns the stage of the order: placed, paid, being prepared, handed to the courier. From the courier company it learns where the parcel is, using the waybill number (AWB). It does not promise a delivery time; it says what the courier’s system shows.
General delivery questions belong here too: areas, methods, the cost shown in the store. The answers come from your information pages, kept up to date.
Returns: steps explained by the assistant, decision made by the agent
With returns, the assistant acts as a guide. It explains, from the store’s policy, how to submit the request, what condition the product must be in and how the parcel is sent back. It can open the request in the system and send the instructions by email.
It does not decide whether the return is accepted and does not approve the refund. Those stay with an agent, who sees the product, the history and the exceptions. Nor does the assistant promise when the money will be returned.
The returns policy must respect consumer rights, including the right of withdrawal from distance contracts. Its wording is checked with a lawyer; we take it over exactly as approved.
Handover to the agent, with a summary of the conversation
The most important moment for a support assistant is the one when it steps back. The handover rules are written out explicitly: the customer asks for an agent, a word such as “complaint” or “faulty” comes up, the assistant still does not understand after a second attempt.
During working hours, the conversation passes to the agent in the same chat window or by call transfer. Out of hours a ticket is opened, and the customer is told honestly when they will be contacted. The agent receives a note with the following.
- Who is asking: name, contact details, whether their identity has been verified.
- The order in question: number, products, stage, waybill number.
- What was asked and what the customer has already been told, so that the agent does not inadvertently contradict the answers given.
- Why it was handed over: an explicit request, a subject reserved for people or a failure to understand.
AI customer support assistant by phone and in chat: connections and stages
Orders are read from the store through an API, the official route by which one application requests data from another. WooCommerce, for example, has such an interface for orders. In chat, the assistant sits in a window on your website; on the phone, the same answers pass through speech recognition and speech synthesis. A language model words the answer, but only from the data it has read and from your texts.
Order data is personal data, so it falls under GDPR. The assistant is given access only to the fields it needs, and calls are recorded only if the caller is told beforehand.
- Analysing the enquiries: together we read the support conversations from the recent period and group them by subject.
- The source texts: the returns policy, delivery, payment, warranty, brought up to date and written clearly.
- Tests on test orders, then a start in chat, where mistakes are easier to see and correct; the phone comes second.
- Tracking: which questions are closed without an agent, what is handed over and why.
Frequently asked questions
Can the assistant approve a small refund on its own?
No. A refund is a decision involving money and exceptions, and a person makes it. The assistant gathers the details and prepares the request, so that the agent can deal with it quickly.
Does it work if the store is not on WooCommerce?
Yes, provided the platform allows another application to read the orders. Without access to orders, the assistant can answer only general questions.
How can I tell whether the assistant is helping rather than getting in the way?
You track how many conversations end without an agent, how many are handed over and for what reason. You read whole conversations from time to time, not just the figures.
This is a typical project description: it shows how we usually approach this kind of work and does not present a project carried out for a particular client. Every real project starts from your company’s situation, and the stages, timescales and price are agreed after the initial discussion.