Most programs spend their first month choosing software and their second month discovering that nobody has agreed what the base is for or who will own an article. Reversing that order costs nothing and changes the outcome, because the decisions that are hard are all organizational and none of them are blocked by the tool.
Weeks one and two: the question list
Pull the questions people actually asked, from the support queue, the internal chat, the inbox of whoever gets asked everything. Rank by frequency. You are looking for the twenty to fifty questions that come up over and over, and the exercise is finished when somebody senior looks at the list and says yes, that is what we get asked.
Weeks three and four: owners, before any writing
Walk the list and put a name against each item: the person who would be asked whether the answer is still right. Expect pushback and treat it as signal, because a question nobody will own is usually a question about a process nobody owns, which is a finding worth having early.
Week five: pick the tool in an afternoon
With the list and the owners in hand the tool choice is small. Load ten real answers into two candidates, ask three colleagues to find five things, and buy whichever one they found them in. The search result is the product; the editor and the branding are not.
Weeks six to ten: write the top half
Owners write their own articles, short, answer first. A central editor passes over them for titles and consistency rather than rewriting. Aim to cover the top half of the ranked list, because half of the questions asked twice is already a base worth searching, and a full list attempted at once produces sixty drafts and no publications.
Weeks eleven and twelve: put it in the path
The base has to appear where the question is asked: a link in the support tool, a bot in the chat channel that searches it, a line in the onboarding checklist. Launching by announcement alone is the most common way an otherwise good base fails, because habits are not changed by an email.
What is deliberately not in the first ninety days
No taxonomy project, no migration of the old shared drive, no analytics dashboard. Each is a way of being busy without answering a question, and each is far easier to do well later against a base that exists than early against one that does not.
Questions people ask about implementation of knowledge management
Who should lead it?
Somebody from the function that owns the questions, with a named executive who will ask a manager why their article is overdue. A project manager with no standing in the business can run the plan and cannot enforce the ownership, which is the part that decides it.
What if we already have hundreds of documents?
Leave them where they are for now. Answer the ranked questions first, linking to an existing document where one genuinely answers it. Migration decided by the question list is short; migration decided by the folder tree never ends.
How do we know it worked?
The same question stops being asked twice. Count that before you start so the comparison means something, and prefer it to page views, which rise for reasons that have nothing to do with anybody being helped.