AI chatbots and customer service: what actually works

A badly designed assistant drives customers away faster than an answering machine. Here are the real conditions for one to serve the business.

Inès Neifar

Founder and managing director

Published on: February 11, 2026
min read: 7 min read

In short

A business chatbot becomes useful when it is limited to a known scope and backed by a controlled document base. It must answer only from your own content: catalogue, terms of sale, opening hours, procedures. Outside that scope it hands over to a human instead of inventing. The three costliest mistakes are letting it improvise about pricing, failing to log conversations, and hiding its artificial nature. Properly framed, it absorbs repetitive questions, qualifies requests outside office hours and passes a summary to the team. Badly framed, it produces commitments the company will have to honour. Visitors must also be told plainly that they are speaking to an assistant, and be able to reach a person at any moment, with the conversation summary attached to the handover.

Why most chatbots disappoint

An assistant plugged into a language model with no scope answers everything, including what it does not know. In customer service that is a legal risk: an answer about a delivery time or a price commits the company.

A useful assistant does the opposite: it knows its limits and says so.

The four conditions

An up-to-date document base. The assistant must answer only from your documents. That base is a deliverable in its own right, and it must be maintained.

A declared scope. Hours, availability, order tracking, technical prerequisites: yes. Negotiation, disputes, contractual commitments: no — immediate handover.

A human bridge. At any moment the visitor must be able to reach a person, with the conversation summary attached.

Traceability. Every exchange is stored, readable in the back-office, and used to correct the document base.

What must be stated plainly

Visitors must know they are speaking to an assistant. Hiding it produces a feeling of deceit at the first clumsy answer, and trust does not return.

The profitability calculation

Count the repetitive questions received per month and the average handling time. Above a few dozen hours, the assistant pays for itself. Below that, a well-written FAQ page does the same job for less.

The right moment

An assistant plugs into an organisation that already works. Without a document base, a callback procedure or a CRM, it will only accelerate disorder. Start with automating inbound requests, then add conversation.

To frame your case, use the project form.

A project in this area?

Two minutes to tell us what you need, and we come back with a first read of your situation.

Reply within one business day

Ask a question