Building an internal knowledge base your team will actually use

What problem a knowledge base solves

In most service businesses, a few experienced people act as the human search engine. New staff ask them how to price a particular job, which template to use, what the policy is on client refunds or how a specific client likes things done. Each question takes a few minutes, but together they fragment senior staff time and slow new hires down.

The information usually exists somewhere: in a SharePoint folder, an old email, a Word document on someone's desktop or the notes field of the CRM. The problem is finding it, trusting it is current and knowing it applies. A knowledge base fixes that by putting the right answers in one searchable place with an owner.

Start from questions, not folders

The common mistake is to start by moving every document into a new system. That creates a tidier version of the same mess. Instead, collect the questions people actually ask. For two weeks, have senior staff note every question they get. Check your team chat for repeated questions. Ask new hires what they struggled to find.

Group the questions into themes and write or locate one clear answer for each. A knowledge base of 50 well-maintained answers that cover most daily questions is more useful than 5,000 documents of uncertain age.

Include short client-specific notes where they help, such as access instructions for a commercial property or a client's preferred reporting format, but keep anything confidential in the system of record with proper permissions rather than copying it into a general knowledge base.

• How do I do a specific task? (procedures and SOPs)
• What is our policy on a specific situation? (policies)
• Which template or document should I use? (templates and examples)
• How does a specific client or property like things done? (client notes)
• Who do I ask about a specific topic? (directory and ownership)

Choose where it lives

Use what your team already opens. Firms on Microsoft 365 often use SharePoint and Teams, with Copilot able to search across them for licensed users. Google Workspace firms use Drive and Sites, with Gemini features available on many plans. Notion, Confluence, Guru and Slab are dedicated options. Some practice management and field service platforms include knowledge base features too.

The choice matters less than consistency. Pick one home, set a simple structure and naming convention, and point people to it. Two half-maintained knowledge bases are worse than one.

Write answers in a consistent pattern: a one-line answer first, then the detail, then a link to the full procedure or policy if one exists. Staff in the middle of a task want the short answer. The detail is there for when they need it. This structure also helps AI assistants, which tend to give better answers when the source documents state the main point clearly at the top.

Adding an AI assistant

An AI assistant on top of your documents lets staff ask a question in plain language and get an answer with links to the source. This is often called retrieval-augmented generation: the assistant finds relevant passages and composes an answer from them. Copilot, Gemini, Notion AI, Guru and others offer versions of this, and custom builds are possible for more specific needs.

Two rules keep it trustworthy. First, every answer should cite the document it came from, so staff can check. Second, the assistant should say when it does not know rather than guess. Test it with 30 to 50 real questions before launch and review wrong answers. Most errors trace back to an outdated or contradictory source document, not the AI.

Get permissions right

AI search will find anything a user has access to, including documents that were shared too broadly years ago. Before switching on an assistant, review sharing settings on sensitive folders: HR files, salary information, partner documents, client matters under confidentiality, patient information. Microsoft and Google both provide admin tools to review overshared content.

For law firms, consider ethical walls between matters. For healthcare, make sure protected health information is handled according to HIPAA. For accounting and tax firms, check the knowledge base design against your written information security plan. This review is not optional; it is the main risk in an internal AI assistant.

Pilot the assistant with a small group first, ideally a mix of new and experienced staff. New staff ask the questions the system most needs to handle. Experienced staff spot wrong answers quickly. Their feedback over two to four weeks will tell you which source documents to fix before a wider launch.

Keep it current

Knowledge bases decay. Prices change, policies update, software gets replaced and the old article stays. Give every article an owner and a review date. Show the last-updated date on each page so readers can judge freshness. Set a quarterly reminder for owners to review their articles, and archive anything nobody claims.

Make it easy to report a problem. A 'this is wrong' button or a dedicated chat channel lets staff flag outdated answers, and the owner gets a clear prompt to fix them. When an assistant gives a wrong answer, treat it as a signal that the underlying article needs attention.

Watch for duplicates. Two articles answering the same question in slightly different ways are a common source of wrong answers from AI assistants, because the assistant may quote the older one. When you find duplicates, merge them into one current article and redirect or archive the other.

Drive adoption and measure

Senior staff are the key to adoption. When someone asks them a question the knowledge base answers, they should reply with the link. When the knowledge base does not have the answer, they should add it. New hire training should include how to search it.

Before launch, estimate how many routine questions senior staff field per week and how long new hires take to become independent on common tasks. Measure again after 60 and 90 days. InTheory would typically pilot a knowledge assistant with one team and a defined set of questions before rolling it out firm-wide, because early wrong answers can put people off for good.

• Routine questions to senior staff per week
• Share of test questions answered correctly with a cited source
• Time for new hires to complete common tasks independently
• Articles past their review date