The phrase covers two products that have almost nothing to do with each other, and a search for it returns both. One encodes rules and reaches conclusions; the other stores answers people wrote and helps somebody find one. Telling which you are shopping for takes one question, and getting it wrong wastes a procurement cycle.
The expert system: rules that reach a conclusion
A knowledge base in this older sense is a set of encoded facts and rules that a program reasons over to answer a question, usually in a narrow domain: eligibility, diagnosis, configuration. It is bought where the same judgement has to be applied consistently thousands of times and where the rules are stable enough to be written down.
The knowledge base: answers people wrote
What most companies mean today. Articles, search, ownership and review. It reaches no conclusions and makes no decisions; it helps a person find something another person wrote. Its risks are staleness and coverage rather than a rule that was encoded wrongly.
The question that tells them apart
Does the software have to produce an answer by itself, or help a person find one? If a human reads the result and applies judgement, you want the second and the whole procurement is different: cheaper, faster, and judged on search quality rather than on how the rules are maintained.
Where the two now touch
Retrieval that composes an answer from your articles looks like the first and is the second underneath: it is not reasoning over encoded rules, it is quoting what somebody wrote, well or badly. That distinction matters for who is responsible when the answer is wrong, which is still whoever owns the article.
Questions people ask about knowledge based software
Which one does a support team need?
The second, almost always. Expert systems are bought where the decision is repetitive and the rules are written down somewhere already, which describes underwriting and eligibility far more often than it describes support.
Can one product be both?
Some decision tools sit on top of a base, walking an agent through a branching question set that ends in an answer. That is useful and it is a third thing, and it is worth buying only where the branching is genuinely the hard part.
Does the older meaning still matter?
In regulated and technical domains, yes, and the vocabulary survives in job titles and academic writing. If a vendor uses the phrase without saying which they mean, ask before the second meeting.