The third new member of staff started this month. It is going well, but the two weeks of questions are the same two weeks every time. Where is the pricing approval process? What is the client onboarding checklist? Which version of the proposal template is current? The answers exist somewhere, spread across emails, shared drives, and the heads of the people who have been there longest. The owner has been told an AI-powered knowledge base would fix this. It might. Whether building one is worth the time and money right now is a different question.
What choice are you actually facing here?
The real decision is between two approaches: a purpose-built AI knowledge base that structures, retrieves, and surfaces your firm’s information automatically, and a simpler combination of well-organised shared documents and a general AI assistant. Both can reduce the time your team spends hunting for answers. The question is which fits your actual query volume, your team size, and your capacity to govern a new system.
For owner-managed businesses, the tipping point tends to sit somewhere around 15 to 20 people and a meaningful volume of repeating questions. A McKinsey study on knowledge worker productivity found that employees spend roughly 19% of their working time searching for and gathering information. For a 20-person firm on a standard working week, that is the rough equivalent of almost two full-time members of staff doing nothing but looking for answers.
Scale matters here. Zendesk’s guidance on AI knowledge management notes that query deflection only becomes meaningful once a firm is handling hundreds of similar questions per month. Below that threshold, the overhead of building and maintaining a structured system can exceed what it saves.
When does an AI knowledge base start to pay off?
An AI knowledge base tends to earn its keep when your firm handles high volumes of repeating questions from customers or staff, and when inconsistent answers carry real risk. Staff turnover sharpens the case: every time someone leaves, undocumented knowledge goes with them, and that cost shows up in slower onboarding for whoever replaces them.
The conditions that tend to tip the decision towards building are repeat question volume above a few hundred per month, a team that has grown past about 15 people, and operations that span more than one location. Simara’s UK-focused guide to knowledge management argues that centralised, searchable systems pay off particularly once a firm passes around 15 staff or opens a second site.
Regulated sectors add another reason. If your firm advises on financial products, provides legal services, or works in health and care, the FCA, SRA, and CQC all hold businesses responsible for the accuracy and fairness of information given to clients. An AI knowledge base can standardise language and reduce ad hoc responses, but only if it is tightly maintained.
There is also a compounding effect for firms already using AI tooling. Grow London Local’s research reports that 57% of UK SMEs already use AI tools daily. If your team is among them, a knowledge layer lets those tools draw from a single consistent source rather than generating answers from scattered documents.
When are shared documents and a general assistant enough?
For firms under about ten people with straightforward services and low query volume, the overhead of structuring and maintaining an AI knowledge base will frequently outweigh what it saves. A well-organised shared drive, a small number of current procedures, and access to a general assistant such as Claude or ChatGPT will handle the majority of what such a firm actually needs.
Elansio’s practical guide for UK SMEs recommends exactly this as the starting point before committing to anything more complex. The upgrade becomes worthwhile once you have enough well-structured content for AI to work with and enough recurring questions to justify the setup cost.
A few conditions suggest staying with the simpler approach. Your team is under ten people and your services are straightforward. Questions arise rarely and vary enough each time that a structured knowledge base would be hard to populate meaningfully. And, critically, no one in the business currently has capacity to own questions like “what personal data are we putting into this system?” or “who checks the content for accuracy?”.
That last condition is the one many firms underestimate. The ICO’s guidance on AI and data protection makes clear that UK GDPR applies to AI systems processing personal data, including internal knowledge tools. If that governance question has no owner, the simpler path is genuinely safer.
What does it cost to get this call wrong?
Getting the timing wrong in either direction has a cost. Build an AI knowledge base before the foundations are ready, and you spend months on a system with low adoption, inaccurate answers, and staff who stop trusting it. Hold back too long, and you carry a steady productivity penalty while the same senior people keep answering the same questions.
Premature implementation tends to fail on adoption ahead of budget. AI knowledge bases need current, well-structured content to produce reliable answers. If your team does not yet have that material in usable shape, the first several months will be spent migrating documents, correcting errors, and chasing article owners rather than saving time. Brightmine’s guide on AI for UK SMEs makes the point directly: the main barrier to value for UK small businesses is change management and adoption, ahead of any technology cost.
The cost of waiting too long is easier to miss because it is diffuse. It lives in the accumulated hours your senior people spend answering the same questions, in onboarding that runs a week longer than it should, and in customer communications that vary depending on who picks up the call. The productivity penalty is real and ongoing; the question is when the cost of addressing it exceeds the cost of building the system to fix it.
What should you ask before you commit?
Before committing budget and staff time to a build, five questions will tell you whether the timing is right for your firm. The questions are operational ones about your actual query volume, the state of your existing content, and your capacity to maintain a new system once it is running. A vendor pitch can wait until these are answered.
First: how many repeat questions does your firm actually handle per month, from customers and from staff? If you cannot count them, that is itself a finding. Second: do you already have reasonably current procedures, FAQs, and service descriptions in writing, or would you be starting from scratch? Zendesk’s guidance notes that the benefit horizon is typically several months of continuous improvement rather than days, and that starting with well-organised existing content shortens that considerably.
Third: what personal or sensitive data might end up in the system, and who will ensure it meets your UK GDPR obligations under ICO guidance? Fourth: which role, not which person, owns the knowledge base, reviews it periodically, and corrects wrong answers when they appear? Fifth: how will you measure whether it is working? Useful metrics include the rate at which repeat queries resolve without senior involvement, onboarding time for new staff, and first-contact resolution on customer queries. Without a baseline on at least one of these, there is no way to measure return on the investment.
The DSIT’s AI Management Essentials programme is building guidance to help businesses make exactly these decisions proportionately. It is a signal that expectations around AI governance are rising, and that the question of whether your knowledge base is properly governed will matter more as AI adoption increases.
