Knowledge sources
We select the documents and pages the assistant may use, and agree who checks, updates and removes them when they are no longer valid.
Customers and employees should not have to search several places for the same answer. Zipwolf develops AI assistants for defined tasks, using selected information sources, clear response boundaries and a handover to a person when further judgement is needed.
Discuss your project
This service suits teams handling repeated questions about their offer, procedures or internal documents. We first check whether accurate, current sources exist. An assistant is not a shortcut around unclear rules or information nobody maintains.
We select the documents and pages the assistant may use, and agree who checks, updates and removes them when they are no longer valid.
We define question types, tone and the situations in which the assistant should ask for clarification or stop. It receives no authority beyond what the business has approved.
We test realistic and difficult questions, then define conversation handover, issue logging and reviews of what needs improving in the sources or rules.
We distinguish a public chatbot from an internal assistant, then agree its users, subject area and prohibited actions.
We organise the information and create test questions, assessing accuracy, clarity and behaviour when an answer is not available.
We agree access, monitoring and who takes over complex enquiries. Broader use follows only after a review of performance.
Illustrative example
A visitor asks whether a service covers their situation. The assistant uses an approved service description, asks agreed follow-up questions and prepares a summary for the team. It does not promise unconfirmed prices or deadlines. Exceptions and unclear cases are passed to the responsible person.
Illustrative example
An employee needs the procedure for a particular request. The internal assistant finds an authorised document and includes the source with its summary. Conflicting documents or missing answers go to the process owner. Access is restricted by user and the rules agreed for the system.
AI can produce incorrect answers that sound convincing. Testing, boundaries and human handover are therefore part of the scope. Sensitive decisions and irreversible actions are not delegated without separately agreed controls and permissions.
It can be included, but each agreed language needs its own source, terminology and answer testing. We do not assume equal quality without checking.
No. It may be for employees only. The channel and integration options are agreed after reviewing your systems and access requirements.
We first decide what is needed and who may access it. Confidential information is not uploaded to an external service before data handling, access and service terms have been agreed.
Tell us who asks the question and where the correct answer is currently kept.
Send an enquiry