The argument for a base is usually made with figures borrowed from a vendor and lost the moment anybody asks where they came from. A better argument uses your own company and three measurements that take an afternoon each, taken before the base exists so the comparison afterwards means something.
Measurable: how many different answers one question gets
Pick three questions your people answer daily. Ask four colleagues each, separately, and write down how many distinct answers come back. Repeat a month after the articles exist. This is the benefit that arrives first and the one nobody argues with when it is their own company's numbers.
Measurable: how often the same question is asked twice
Count repeats in whatever channel people actually use, over a fixed window. It is tedious and it is the number that moves for a reason you can explain, unlike page views, which rise when somebody links to an article from an announcement.
Measurable: how long a new starter takes to stop asking
Count the questions a new starter asks in their first fortnight. Do it for the next two hires as well, so you have a baseline rather than an anecdote, then compare once a reading order exists. The improvement here is usually the largest and the least claimed.
Not measurable, and worth having anyway
Arguments that stop being re-run because the answer is written and owned. Decisions that got better because writing the process down exposed that it made no sense. Both are real, both turn up in every base that lasts a year, and neither belongs in a business case, because a benefit you cannot evidence weakens the ones you can.
Questions people ask about benefits of a knowledge base
Why not use the vendor's figures?
Because the first question in the review will be where they came from, and if the honest answer is a vendor's marketing page, the rest of your case inherits the doubt. Your own three numbers are smaller and unarguable.
What if we already launched without a baseline?
Take the measurements now and treat them as the new baseline for the next stage. Half a comparison is still better than a borrowed statistic, and coverage will keep improving for a year.
How long should the window be?
A month either side, at minimum, because question volume moves with the business cycle. Two weeks will show you noise and invite an argument you cannot win.
Does deflection belong in the list?
Only where articles face customers and you can compare contact volume on the covered questions against the uncovered ones. Total contact volume moves for a dozen reasons and proves nothing either way.