Internal Knowledge Base Build (Notion + Claude)

Until your knowledge is structured and accessible, AI augmentation is theoretical. The knowledge base build is the foundation everything else rests on. A VA, Claude or a Custom GPT can only be as good as what it has to read from, and in most AU small businesses that “something” is scattered across a drive, an inbox, a Slack history and the founder’s head. We fix that first.

What this actually does

A knowledge base build turns the loose knowledge running your business into one structured, searchable, AI-readable source of truth. In practice that means a Notion workspace (for most clients) organised the way modern AI tools can actually use, not just a folder of documents nobody can find.

There are two audiences for it, and we design for both at once:

• Your team reads it like a wiki. SOPs, product docs, pricing, supplier contacts, a decisions log. A new VA or staff member can self-serve instead of interrupting you fifteen times a day.
• AI reads it like a context source. This is the part that has genuinely changed since 2024. With your knowledge structured properly, Claude (via Claude Projects, or programmatically via the Notion API) and Custom GPTs can answer questions grounded in your real documents instead of guessing. “What’s our refund policy for wholesale orders?” gets answered from the actual policy, not a plausible-sounding hallucination.

That second audience is why structure matters more than it used to. A pile of PDFs is fine for a human who can skim. An AI retrieving against your knowledge needs clean, atomic, well-titled documents with consistent fields, because retrieval quality is a direct function of how well the source is organised. Garbage in, confident-garbage out. A good schema is what makes the difference between an AI that answers your team’s questions reliably and one that quietly makes things up.

What 2026 AI can genuinely do with a good knowledge base

We describe what is real, not the brochure version:

• Answer staff and customer questions from your own docs. Loaded into a Claude Project or wired through the Notion API, the AI retrieves the relevant SOP or policy and answers from it, with the source attached so a human can check it.
• Draft from your standards, not generic ones. “Write the onboarding email for a new wholesale account” pulls your actual onboarding steps and tone, because they exist as a document it can read.
• Onboard a new hire or VA without you. The knowledge base plus an AI assistant means a new starter asks the assistant first and you second. The handover that used to live in your head is now a thing they can query.
• Surface what is missing. Building the base is itself diagnostic. The gaps we find (the SOP that was never written, the process only one person knows) are usually the riskiest knowledge in the business.

What it does not do: it does not replace judgement, it does not stay current on its own, and it will confidently answer from an out-of-date document if you let one rot in there. Structure and governance are what keep it honest. More on that below.

What it looks like in a real AU business

• The agency preparing for a VA. A founder doing everything wants to hand client onboarding, reporting and inbox triage to a VA, but it all lives in their head. We audit what exists, write the 15-25 priority SOPs that are missing, and structure it so the VA (and Claude) can run the playbook without the founder narrating it.
• The product business with scattered docs. Specs in Google Docs, supplier terms in email, returns policy in three slightly different versions. We migrate it, de-duplicate it, and give it one schema so “what’s the lead time on SKU X” has one answer, not three.
• The business preparing for sale or handover. Due diligence punishes “it’s all in the founder’s head”. A structured, current knowledge base with a decisions log is an asset a buyer can value and a successor can run. We build it so the business survives the founder stepping back.

How the engagement runs

This is a fixed-scope project, typically 3-4 weeks, from $2,200 AUD. It runs in clear stages:

1. Discovery and content audit. We map where your knowledge currently lives and what state it’s in. You point us at the drives, the docs, the tools, and tell us the goal (onboard a VA, get Claude answering product questions, prepare for due diligence).
2. Schema design. We design the Notion workspace: databases, templates, permissions, and the structure for SOPs, product docs, customer FAQs, a decisions log and supplier contacts. This is the part that makes or breaks AI retrieval, so we don’t rush it.
3. Migration. We bring your existing docs across from Google Docs, Word and PDFs. A 100-document folder is typically a few days: we extract the folder structure, convert what’s relevant and attach the binaries.
4. Authoring the gaps. We write the 15-25 priority SOPs we identify as missing. This is usually the highest-value part, because it captures knowledge that previously existed only in one person’s head.
5. AI integration. Where it fits, we connect Claude Projects (manual ingest) or the Notion API (programmatic) so your AI tools read from the base directly.
6. Training and handover. A team training session, plus maintenance and governance documentation so it stays current. Then 30 days of post-launch tuning.

What you provide: access to your existing docs and tools, a few hours of subject-matter time so we can capture what’s only in your head, and a clear goal. What you get: a structured, populated, AI-ready knowledge base, the missing SOPs written, and the documentation to keep it alive.

Notion or Confluence, and where Claude fits

For most AU small businesses, Notion: more flexible, lower friction, and better AI integration. Confluence if you’re already committed to the Atlassian stack. Either way, the value is in the schema, not the logo.

On the AI side, Claude can read your knowledge base two ways. Claude Projects is the simpler path: you ingest the relevant docs into a Project and Claude answers grounded in them, good for a small, fairly stable knowledge set. The Notion API is the programmatic path: Claude (or a Custom GPT, or an automation) queries Notion live, which suits a larger or frequently changing base. We recommend one in scoping based on how big your knowledge is and how often it changes.

The honest limitations

A knowledge base is not magic, and we won’t pretend the AI layer is either.

• It goes stale if nobody owns it. An AI will answer just as confidently from a year-old policy as a current one. That’s why we ship governance documentation and a maintenance routine, not just a pretty workspace. Someone in your business has to own keeping it current.
• Retrieval isn’t perfect. Even with clean structure, an AI can occasionally pull the wrong document or miss a nuance. We design so sources are always visible, so a human can sanity-check the answer rather than trusting it blind.
• It can’t capture what you won’t tell us. The knowledge only an exhausted founder half-remembers is the hardest to extract and the most valuable. The build is only as complete as the time you give us to mine it.
• It is overkill for very small teams. If you’re a single founder with two staff, a tidy doc folder beats a full Notion build. We’ll tell you that rather than sell you a project you don’t need.

The trade is straightforward: AI is excellent at retrieving and drafting from a well-structured source, and useless without one. This service builds the source.

What “done” looks like

Every DotVA AI services engagement is fixed-scope, fixed-price, and fixed-timeline. No retainers without a defined deliverable. No “discovery phases” that turn into 6-month projects. The scoping call is free; the written quote that follows is what you sign off on.

Why we say no

We turn down roughly 30% of applications. Usually because the team isn’t ready, the goal isn’t clear, or there’s a simpler path that doesn’t need us. We’d rather tell you that on the discovery call than take a brief that won’t deliver. The shortest path to outcome isn’t always us.

Apply below

Two minutes. We read every application within 1 business day. If we’re a good fit, we book the scoping call. If we’re not, we say so – and usually suggest who is.