Notion, SharePoint, or an Owned Intelligence System: Where Should Company Knowledge Live?

Company knowledge should live in a store your business owns, kept current and traceable, sitting above whichever tools you already use to hold the documents. The tool is the part of this question that matters least, and almost everybody asking it expects the answer to be a tool name.

It is a fair thing to expect. The question usually arrives in a practical form. Somebody has proposed moving the team onto Notion. Somebody else points out that everything is already in SharePoint and the licences are paid for. A third person mentions that half the actual detail lives in a shared drive nobody has tidied since 2023. The decision feels like a procurement question, so it gets treated like one, and it gets postponed like one.

Then the same problem shows up again eighteen months later in the new tool.

What the question is actually asking

"Where should our knowledge live" is four questions wearing one coat. They have different answers, and only one of them is really about software.

The first is where the files sit. This is the storage question, it is genuinely solved, and any of the options handles it.

The second is whether the reasoning is anywhere. Not the proposal, but why you priced it that way. Not the client list, but why you stopped taking that kind of work.

The third is whether the thing can answer a question. Not return ten documents that mention the word, but tell you what your position is and show you where that came from.

The fourth is what happens when you want to leave.

Businesses compare tools on the first question and then wonder why the other three are still open. That is the mechanism behind the eighteen-month loop, and it is not a failure of research. The IBM CEO study found that half (50%) of respondents acknowledge that the pace of recent investments has left their organization with disconnected, piecemeal technology. Buying carefully and ending up scattered are not opposites. They are the normal outcome of solving the storage question repeatedly.

Four places, judged honestly

Dimension	Shared drive	Wiki (Notion, Confluence)	Microsoft 365 and SharePoint	Owned intelligence system
Holds the files	Yes, and cheaply	Yes	Yes, with proper permissions	Yes, and it is usually fewer files
Holds the reasoning	Only if somebody wrote it down	Only if somebody wrote it down	Only if somebody wrote it down	By design, that is what goes in
Answers a question	No, it returns filenames	Search, then you read	Search, then you read	Yes, with the source attached
Stays current	No	Decays quietly, nobody can tell	Version history, but no view of truth	Review dates and named owners
Cost to leave	Low, files come out	Files come out, structure does not	Files come out, the estate does not	Low by design, or it is not owned
What it is genuinely good at	Cheap durable storage	Writing and organising	Permissions, compliance, scale	Being asked things

Read down the "holds the reasoning" row. Three of the four say the same thing, which is that the tool will hold it if a human writes it, and none of them will ask. That row is the answer to why the tool choice keeps disappointing people. You can move a business from a shared drive to a wiki, run the migration properly, train everybody, and end up with the same knowledge gap in a nicer interface.

The row that changes the decision

For most of the last decade the third row did not matter much, because nothing could answer a question anyway. It matters now, and it is the reason this has stopped being a filing debate.

Pointing an AI assistant at an existing document store is the obvious first move, and nearly every team tries it. What comes back is fluent, confident and impossible to check. The model is not the problem. It is reading pages with no dates, no owners and no sources, and it cannot tell a settled position from an abandoned draft any better than a new starter could. It just sounds more certain about it.

This is the practical difference between a document store and an intelligence system. One holds files. The other holds a smaller set of current, cited pages, so an answer arrives with the document behind it and a date you can check. We have written elsewhere about what actually goes into one and how it is maintained, and the honest summary is that the discipline is the product. The software underneath it is close to interchangeable.

What it costs to leave

The fourth question is the one nobody asks at the point when the answer is still negotiable.

Licence costs are visible and comparable, and they are not the exit cost. The recoverable part is the files, and most tools hand those back. What does not leave is everything the tool held in its own shape, which is the structure, the links between pages, the permissions model and the accumulated context that made the files navigable in the first place. Export the pages and you get the pages. You do not get the thing that made them findable.

We saw this at its sharpest with a UK commercial flooring business running on a closed legacy platform. It cost £565 a month for nine users and held 13,000 contacts, and because it had no API and no structured export, the customer data was effectively trapped inside it. The records existed. They were the company's records. None of that helped, because there was no path out and no path in. Replacing the foundation freed up 50% of staff time, and the point worth taking is that the cost had been running quietly for years before anyone priced it.

The government's own guidance treats this as a buying decision rather than a leaving decision. The AI Playbook tells teams drafting requirements to consider strategies to avoid vendor lock-in and to consider intellectual property rights and who will have these if new software is developed. Both belong in the conversation before a tool holds anything that matters.

There is a second exit question that did not exist a few years ago, and it is about what the tool learns rather than what it stores. UK businesses are markedly unenthusiastic about this. The government's UK Business Data Survey asked the businesses that handle digitised data how they would feel about that material being used to train external AI models, and found that 18% of these businesses said that their business would feel comfortable with its data being used to train external artificial intelligence models, and 48% said that they would feel very uncomfortable. The survey was explicit that the question covers ordinary working material. This included data such as documents, images and customer interactions. Whatever a business concludes about that, it is now part of deciding where knowledge lives, and it is worth deciding on purpose. That is the wider argument about what you are renting when your knowledge sits in somebody else's system.

The verdict, which is duller than the question deserves

Keep your tools. Add the layer.

If the business runs on Microsoft 365, the answer is not to migrate to a wiki. If it runs on Notion and people like it, the answer is not to move to SharePoint. Those migrations cost months, disrupt everybody, and leave all four questions exactly where they were. The storage question was never the one that was hurting.

What is missing in almost every case is a small owned store above whatever you already have, holding the decisions and the reasoning behind them, with a named owner and a review date on each page, capable of being asked a question and answering with the source attached. It is deliberately small. Three or four well-chosen sources beat forty dumped in at once, and a first build can run on the tools you already own, which is usually the right call.

That is also why this is not really a build-versus-buy decision, and not the same question as whether an internal team or a partner should build it. Who builds it is a separate conversation with a separate answer. Where it lives is settled first, and it is settled by deciding that the reasoning belongs somewhere you control rather than distributed across suppliers by accident. That principle is the one we have set out in full on owning your intelligence.

What this costs

Doing it properly with a partner sits in the £6,000 to £25,000 range for a first phase, with smaller scoping and consulting engagements from around £3,000. The spread depends on how many systems have to be connected and how much of the reasoning is already written down rather than living in people's heads.

The honest test before spending anything is whether you can name where a decision from last March is recorded, and whether anyone can find the reasoning behind it in under five minutes. If the answer is no, the tool was never the problem. If you want that assessed before committing to a build, a knowledge audit is the smaller first step and it answers the question with evidence rather than opinion.

Practical takeaways

• Stop comparing tools on storage. All four options handle files, and that has not been the constraint for years.
• Ask the four questions separately, because storage, reasoning, answers and exit have different answers. Only the first one is a software decision.
• Before you consolidate, check whether the reasoning exists anywhere at all. Migrating knowledge you never captured just moves the gap.
• Judge any candidate on whether it can be asked a question and answer with a dated source. If it returns filenames, it is storage.
• Price the exit at the point of buying. Ask what leaves in a usable form, in what format, and what stays behind in the vendor's shape.
• Decide deliberately what your tools may learn from. Roughly three quarters of the UK businesses that handle digital data are uncomfortable with it training external models, and most have never made that an explicit choice.
• Keep what you have and build the layer above it. The migration you are considering probably solves the wrong problem.

Frequently asked questions

Where should a growing company store its knowledge?

In the tools you already use for the documents, with one owned layer above them that holds the reasoning and can be asked a question. Almost every business that asks this expects the answer to be a tool name, and the tool is the part that matters least. Shared drives, wikis and Microsoft 365 are all perfectly good places to keep files, and moving between them changes very little. What changes the answer is whether anything above those files records why the business decided what it decided, keeps that current on a schedule, and gives an answer with the dated source attached. Most businesses have the storage and not the layer, which is why the question keeps coming back.

Is Notion good enough?

For writing and organising, yes, and it is better at that than most of what it replaced. The limitation is not the software, it is what a wiki asks of the people using it. Pages get written once, by whoever had the energy, and nobody can tell you six months later which ones are still true. That is fine for a handbook and it is not fine as the thing your business or an AI assistant reasons from. If you already run on Notion, keep it. Add the discipline that a wiki does not enforce on its own, which is a named owner, a review date on every page, and a source behind every claim.

Is SharePoint a good place for company knowledge?

It is a good place for documents and a poor place for answers, and the distinction is the whole point. SharePoint and the wider Microsoft 365 estate handle storage, permissions and version history properly, which is genuinely valuable and hard to replicate. What it does not do is tell you why a decision was made, or which of the four documents matching your search is the one still in force. Search returns files and leaves the judgement to you. If your business already runs on it, the work is not migration, it is building something above it that can answer.

Can AI answer questions from our wiki?

It can, and that is the problem, because it will answer just as confidently from the pages that went stale two years ago. Pointing an assistant at an existing wiki or document store is easy and most teams try it. What comes back is fluent, plausible and impossible to check, because the underlying pages carry no dates, no owners and no sources. The fix is not a better model. It is giving the model something worth reading, which is a smaller set of current, owned, cited pages, so that every answer arrives with the document behind it and anyone can see when it was last true.

What does it cost to move off a knowledge tool?

Rarely the licence, and usually more than the export button suggests. The recoverable part is the files, and most tools will give those back. The expensive part is everything the tool held in its own shape, which is the structure, the links between pages, the permissions and the accumulated context that made the files navigable. One flooring business we worked with had 13,000 contacts in a closed platform with no API, and the data was effectively trapped even though the records technically existed. Ask what leaves in a usable form before you sign, not when you want to go, because that is the point at which the answer is still negotiable.