What is an AI assistant?

A rule-based chat flow follows a written route. The greeting, the list of options and the answer behind each option are defined in advance. Pick an item from the list and a ready answer appears. Type something outside the list and the flow stalls.

An AI assistant reads a sentence written in free text. It finds the match even when the question is worded differently, and it replies in full sentences. In return, it depends entirely on the sources it has been given.

The two are often combined. Frequent questions with fixed answers are handled by rules, and the rest goes to the AI layer. That keeps both speed and coverage.

In a sound setup the assistant speaks only from the sources it is given. Those sources are the company's own texts: service pages, price lists, opening hours, terms and a list of common questions.

When a question has no answer in those sources, the assistant does not invent one. It says so and passes the conversation to a person. That behaviour is defined during setup, and a gap left undefined fills with a wrong answer.

Three measures keep the risk down. The source list stays short and current. The page behind an answer is logged. Anything that quotes a price or makes a promise goes to a person. Transcripts are read regularly, and a source that caused a wrong answer is corrected.

A handover happens in three cases. The first is a question with no answer in the sources. The second is the visitor asking for a person. The third is a listed topic: complaints, price negotiation and contracts sit on that list by default.

At handover, a summary of the conversation and the customer record open on the colleague's screen. The customer does not have to explain twice. When the record is linked to a CRM, earlier messages appear on the same screen.

Outside office hours there may be nobody to take over. The assistant then records the request, states when someone will reply, and the record goes to the queue for the next morning. Keep that promise realistic, because a missed callback costs trust.

The first place is the company's own website. A visitor types a question into the corner window and reads an answer without filling in a form first. The site content is already the source, so setup there is short.

The second place is WhatsApp. On a number connected to the WhatsApp Business API, an incoming message reaches the system first, and the assistant can write the first reply. For the customer there is one line, and who answers changes behind the scenes.

Typical users are businesses that run on appointments, hotels and service offices. In a hotel, check-in and check-out times, parking, breakfast and cancellation terms come up every day. Those questions have a clear source, so they suit an AI assistant.

The first point is an honest greeting. Tell the visitor they are talking to an AI assistant, and show the route to a person in the first message. Keep the greeting short, and say in one sentence what the assistant can and cannot do.

The second point is the handover list. Decide before setup which topics never stay with the assistant. Complaints, price negotiation and contracts go to a person by default. The list grows later, because the first month of real questions corrects it.

The third point is records and GDPR. Written conversations are stored, and voice recording happens only on request and with consent. If personal data is collected, write down the lawful basis, the retention period and who can read the transcripts. ALTAI Digital hands over a setup with those three points in place.