In short
- AI does the groundwork: understanding the enquiry, assigning it, pulling data from your systems and drafting a reply.
- The safe start is drafts with approval. Replies are only sent automatically once results have been right for weeks.
- You do not need a chatbot. The biggest lever is often the ordinary email inbox.
- Data protection, a clear handover to people and, for direct customer contact, disclosure of AI belong in from the start.
What can be automated in enquiries
A customer enquiry almost always goes through the same steps. Each of them can be supported by AI:
- Bundle incoming messages: emails, contact form, shop messages and chat come together in one place.
- Read and classify: the AI recognises the request, urgency and language, for example a complaint, an appointment request or a request for a quote.
- Look up data: it finds the customer, order or delivery status in your CRM, ERP or calendar.
- Draft a reply: based on your templates, prices and rules, a draft is created in your company’s tone.
- Approve and send: a person checks, edits if necessary and sends. Standard cases can later go out automatically.
- Hand over: anything unusual, sensitive or emotional goes straight to the responsible person, with a summary.
Three levels of automation
| Level | What happens | Suitable for |
|---|---|---|
| Sorting | The AI assigns, flags urgent items and summarises | Getting started and full inboxes |
| Draft with approval | The AI writes the reply, a person checks and sends | Most enquiries |
| Automatic reply | Standard cases are answered without review | Frequent, clear-cut questions such as opening hours or delivery status |
Most businesses already get a long way with the first two levels. The third level only pays off once it is clear which cases really are always the same.
A typical process by example
A building services wholesaler, Monday morning, the inbox is full. A customer asks whether her order from last week will arrive before the weekend. The AI recognises a question about delivery status, finds the order through her email address, reads the status from the ERP and writes a draft with the delivery date and tracking number. The responsible colleague sees the draft in the mailbox, checks it in a few seconds and sends it.
Two emails further on, an installer complains about a wrong delivery. Here the AI does not write a reply but summarises the case, attaches the order and delivery note and passes it to the head of inside sales. This example describes a typical process, not a client project.
Channels: where enquiries arrive
- Email: the most important channel in most businesses. Connected to Microsoft 365, Gmail or other mailboxes through the usual interfaces.
- Contact form: structured fields make assignment easier. Only make fields mandatory where they are really needed.
- Website chat: an AI chatbot answers simple questions immediately and hands over the rest.
- WhatsApp and messengers: the most convenient route for many customers, with their own rules on data protection and consent.
- Phone: an AI assistant can take down requests and pass them on as tickets. That is an addition, not a substitute for being reachable.
What matters
- An up-to-date knowledge base: prices, delivery times, terms and frequent answers in one place, maintained by a responsible person.
- Clear handover rules: when does an enquiry go straight to a person, for example complaints, legal questions or cancellations?
- The right tone: examples of good replies from your business help the AI more than long instructions.
- Data protection: a data processing agreement, servers in the EU where possible, only the data needed for the reply.
- Disclosure: if an AI talks to customers directly, for example in a chat, they must be able to tell it is an AI. More on the transparency obligations.
- Measurement: response time, share of drafts sent without changes, and follow-up questions from customers.
What it delivers, honestly
The biggest gain is usually not the minute saved per email, but the faster first response and consistency: every enquiry is read, classified and answered with the right data, even on busy days and during holidays. Your team writes less routine and has more time for customers who really need a conversation.
Where AI reaches its limits: angry customers, complex complaints, negotiations and anything that needs tact. There it prepares at most. This boundary should be defined from the outset. How we implement such processes is described on our page on AI customer service.