The genre has a problem: the case studies that are easiest to find were commissioned by whoever sold the software, and the ones written by the company itself are usually published while the program is still in its first optimistic year. Four questions applied in about two minutes tell you whether what you are reading will transfer to your company.
Who wrote it, and what were they selling
A vendor case study is not worthless, it is evidence about what the product can do when the customer is unusually motivated and supported. Read it for the mechanics and discount the outcome. A conference talk by the practitioner is better, and a write-up published two years after launch is better still, because by then the decay has had time to show.
How long after launch was it written
Almost every failure in this area is a second-year failure: coverage plateaus, owners drift, articles go stale invisibly. A study written at six months cannot have seen any of that, whatever its numbers say, and its numbers will be the best they will ever be.
What was the baseline
A percentage improvement with no stated starting point is decoration. If the study does not say what was being measured before, and over what window, then the only thing it evidences is that somebody produced a chart. This eliminates most of the genre in one question.
Is the mechanism named
The useful part of any case study is the specific thing they did that you could copy: owners assigned before writing, a review date with a consequence, the base surfaced inside the ticket. If the study describes a culture of sharing and names no mechanism, there is nothing in it to take.
Questions people ask about knowledge management case studies
Are there case studies worth reading?
Public-sector lessons-learned programs publish honestly because they are required to, and the failures are documented alongside the successes. They transfer less directly to a small company and they lie less.
What should we do instead of looking for one?
Write your own baseline measurements before you start. Three numbers about your own company beat any number of studies about somebody else's, and you will need them anyway when somebody asks whether it worked.
Do failures get published?
Rarely, and that absence is the most important fact about the genre. Assume that for every published success there are several quiet replacements, and read the successes for mechanism rather than for probability.