Not every useful automation is AI
Opening hours need a visible answer. A fixed selector often helps with a few packages. KiteGuide uses three questions and fixed rules. It is neither generative AI nor live chat.
AI may help when questions vary widely and there is a lot of information to search. It should not make binding commitments on its own.
The knowledge base matters more than the chat window
My fictional Raumwerk concept shows an assistant for service questions. It is not a client deployment. The practical task behind it is real: information must be current and consistent.
Start with ten to twenty common questions. Each needs a checked answer and someone to maintain it. The assistant must be allowed to say when information is missing.
Clear limits and a person to turn to
Individual quotes and complaints often need the team. Appointments require a reliable data source. Automated help must identify itself and offer personal contact.
Before launch, establish which data goes to which provider. Plan access, storage and deletion. Credentials must not be exposed in the browser.
Test time savings in a small pilot first
Test simple questions, ambiguous questions and ones that should remain unanswered. Record errors and useful handovers. Count maintenance and correction time too.
Saving ten minutes of conversation helps little if it creates twenty minutes of corrections. A small pilot shows whether the tool is worthwhile. The NIST framework helps structure the understanding, measurement and management of risk.