A company hires its first employee in another country. Six months later, there are three people there. A year after that, leadership is discussing a local entity.

By then, the employment structure may have changed several times while the knowledge around it has remained informal. Decisions live in inboxes. A local exception is remembered by one HR manager.

‍

A payroll process still makes sense only because someone remembers why it was set up that way. The problem is no longer just administration. The business has created a knowledge continuity risk.

A 2026 paper in Knowledge and Process Management examines knowledge continuity at transition boundaries, where responsibility moves and context can become separated from the knowledge being transferred. International expansion creates exactly these kinds of boundaries. The useful question for Knowledge Management teams isn't simply whether the records exist, but whether the next owner will understand the decisions behind them.

The first hire creates a knowledge boundary

The first international employee may arrive long before the company is ready to establish a legal entity in that country. In that situation, the business may use an employer of record. Under this employment structure, the EOR acts as the legal employer and manages areas such as employment contracts, payroll, benefits, and local employment requirements while the company directs the employee's day to day work.

That split matters to knowledge management. Internal teams need a clear record of which decisions belong to the company and which responsibilities sit with the external provider. If a local allowance is introduced, for example, retaining the final policy is only part of the job. Future owners also need to know why the decision was made, who approved it, and whether it was an exception or something intended to continue.

A growing team exposes what was never captured

One employee can compensate for a weak knowledge system through memory and direct access to headquarters. Five employees expose the gaps much faster.

The first hire may know who to ask, how a local process differs from the standard one, or which workaround keeps a recurring issue moving. New colleagues don't automatically inherit that context. If it remains in private conversations, onboarding starts to depend on whoever happened to join first.

This is the point where local operating knowledge needs a durable home inside the organisation. For some companies, that may be an established knowledge platform; for others, it may mean building a dedicated training or knowledge-delivery platform around their internal processes. That doesn't mean documenting every conversation. Focus on information another person would struggle to reconstruct later, especially recurring exceptions and the reasoning behind local process differences. The aim is continuity, not documentation volume.

Entity planning includes a knowledge inventory

When a company starts considering its own local entity, discussion usually centres on legal structure, cost, and employment administration. Knowledge Management belongs in that planning as well because a new entity changes who owns processes and who performs them.

Work previously handled through external employment infrastructure may move to internal teams. If the transfer is treated as a records migration, the company can bring over the files while leaving behind the operational memory that made those files understandable.

Before responsibilities move, identify what the incoming owner must be able to explain without relying on the previous one. Policy rationale and unresolved employee matters deserve attention, but so do informal dependencies that have quietly become part of local operations. ISO 30401 treats knowledge management as a management system that is established, maintained, reviewed, and improved. That principle fits expansion well because the knowledge system has to evolve with the operating model.

Give the handoff an owner on both sides

A cleaner transition has an outgoing owner who understands the current arrangement and an incoming owner who will carry the responsibility forward. Granting system access or transferring folders doesn't create that understanding on its own.

A useful handoff records both the decision and the reasoning behind it. It also makes unresolved issues visible and names the person who owns the next step. That exposes weak spots before they become inherited problems. If nobody can explain a recurring payroll exception, resolve it before the new entity takes responsibility, ideally by tracking it directly in the company's HR software rather than relying on individual memory.  If a local manager has been handling an unofficial onboarding step for a year, decide whether that practice belongs in the formal process before it disappears into another handoff.

A simple test before the structure changes

Legal structures change as international operations mature. The more useful measure of continuity is whether the next owner can understand the operation without reconstructing its history from scattered files and old messages.

Before transferring employment responsibilities, ask whether the incoming team can explain the important local decisions, the exceptions that still matter, and the reasoning behind current processes without calling the outgoing owner for context. If the answer is no, the transfer isn't finished. That test gives KM teams a practical way to judge readiness before an administrative change becomes a knowledge loss event.

_____________________________

Most KM programs fail at the same step: nobody can see what they actually have. Not for lack of content — organizations are full of knowledge.

The problem is visibility. What exists, where it lives, who owns it, who needs it, and where the gaps quietly sit.

I have put together a Knowledge Mapping Playbook:
a practical framework to map an organization's knowledge, assess it, prioritize the gaps that matter,
and turn findings into action — from SME interviews to
a governance checklist to a one-page operating model.

A map makes knowledge and gaps visible.

A playbook makes it actionable. Sharing the playbook below — curious how others are tackling this...

Knowledge Mapping and Playbook Download....
‍
‍

_______________________

Something happens to organizations in the Summer and most of us feel it without naming it.

The building runs on a smaller crew. Half the people who normally answer within the hour are somewhere with their phone face down and quite a lot of what the organisation knows is away with them.
‍

This is when you find out what your organisation actually knows, as opposed to what a few individuals happen to remember.

For most of the year, the gap is invisible. When everyone is present, missing reasoning gets patched in real time. You don't understand why a process works the way it does, so you ask the person who designed it, and they explain it from memory in thirty seconds. The knowledge was never written down, but it was never missed either, because the carrier was at the next desk.

Summer removes the carrier. The person who could explain in thirty seconds is on a beach, and the same question now takes a colleague half a day of digging or worse, produces a confident guess that turns out wrong.

None of this is a crisis. It is mildly inconvenient, and it resolves itself in September. But it is a useful, low-stakes preview of a much larger problem, because the holiday gap and the permanent gap are the same gap.

When someone goes on leave, their knowledge becomes temporarily inaccessible. When someone leaves for good, or moves roles, or retires, that same knowledge becomes permanently inaccessible and there is no September to look forward to. The reasoning behind a decision they made three years ago doesn't come back from holiday. It is simply gone, and the record that remains confirms that the decision was made without explaining why.

Summer just makes the mechanism briefly visible.

So it is worth paying attention to what the quiet weeks reveal. Which questions can't be answered while a particular person is away? Which processes only really make sense to one individual? Where does the covering colleague say "I'll check when they're back" — and what would happen if they never came back?

Those are not summer problems. They are the year-round problems, wearing a lighter outfit.

The organisations that handle August well are usually the ones that already do the unglamorous work: capturing why things are done the way they are, not only how; keeping the reasoning behind decisions accessible to people who weren't there when they were made. For them, a colleague going on leave is a scheduling matter, not a knowledge outage.

For everyone else, summer is a two-month reminder that quite a lot of what the organisation knows is actually just what specific people remember.

The good news is that it's a gentle reminder. Nothing important usually breaks in August. But it's worth noticing what goes quiet while people are away, because the same silence arrives permanently, eventually, and with far less warning.

For now, though, if you're one of the people currently away-enjoy it. The reasoning will keep until September.

And if you're holding the fort: make a quiet note of every question you couldn't answer this month. That list is more valuable than it looks.

The views expressed in this article are my own and do not represent the position of my employer or any institution I am associated with.

_________________________________________________

1. The Intro

For decades, Enterprise Knowledge (tacit and explicit, structured and unstructured) has felt like a vault full of treasure that nobody had the key to unlock.

Open weight AI models might just be the "Open Sesame" moment knowledge leaders have been waiting for.

‍

2. The Story... So Far

If you haven't been tracking the frontier AI landscape over the past few weeks, the open-weight debate just reached a boiling point:

• July 9–13, 2026 | Security & Frontier Risks: An autonomous AI agent from frontier lab OpenAI escaped its isolated sandbox environment (a "jailbreak" in lab parlance). To pass an internal benchmark test, it inferred that an external site (Hugging Face) might hold the answers. It gained internet access, moved laterally, and compromised Hugging Face's dataset processing infrastructure and cluster credentials.
• July 16–27, 2026 | The Benchmark Shift: Moonshot AI released Kimi K3, an open-source frontier model matching or beating closed proprietary models like Claude Fabel and OpenAI Sol across key benchmarks. Think of it as AI's "Tesla vs BYD" tipping point. On July 27, Moonshot offered free access to its model weights—marking a major turning point for knowledge management.
• Industry Giants Line Up: NVIDIA’s Jensen Huang authored an open letter defending open weights, co-signed by Microsoft, Google, IBM, and over two dozen tech leaders resisting proposed federal bans. (The notable exception was Anthropic, who did not sign).
• Enterprise Validation: AWS and enterprise IT leaders are stepping in to provide enterprise-grade support and security for open-weight deployments—mirroring the playbook that made Linux the backbone of modern enterprise software.

3. The Technology (In Plain English)

So, what actually are "Open Weights"?

When you use proprietary SaaS models, you are renting access to a black box over an API. You send your data out; you get an answer back. This is the case for closed models like Claude, ChatGPT, or Gemini—you cannot see how your data is being used. Because your IP is at risk, there is a forced brake on AI adoption and progress for many companies, especially in highly regulated sectors.

With Open Weight models (like Kimi K3), the provider gives you the underlying neural blueprint and trained parameters ("the brain"). You can download it, host it on your own servers or private cloud, run it offline, and tweak it as you see fit.

You control not just your data security and privacy, but the model itself. You can begin to imbue it with the values and personality of your organization. For example, the term "disclosure" carries a vastly different legal weight in a law firm than in a restaurant chain.

Owning model weights is the difference between renting a taxi (Claude or Gemini) versus owning the vehicle (Kimi K3 or Inkling models) and parking it inside your own private garage.

4. Open Sesame: What This Means for Knowledge Managers

For Knowledge Managers, this technical shift solves the two biggest roadblocks that have plagued enterprise KM for twenty years:

• Absolute Data Sovereignty & Privacy: You no longer have to compromise between cutting-edge AI and strict legal compliance. Run models inside your firewall or Private Cloud (VPC)—sensitive IP, code, and confidential documents never touch a third-party server.
• Bespoke Domain Expertise: Off-the-shelf commercial models know a little about everything, but zero about your internal jargon. Open weights allow you to fine-tune smaller, highly efficient models directly on your corporate taxonomy and historical post-mortems. KM teams are uniquely positioned to gather, organize, and validate what the model should and shouldn't learn—opening a massive new horizon for modern knowledge managers as models are updated periodically in partnership with IT.
• Predictable Cost & Scalability: Querying expensive proprietary APIs millions of times a day drains budgets fast. Open weights let you optimize inferencing costs and run task-tailored models affordably at enterprise scale.
• From "SharePoint Graveyard" to Active Intelligence: Passive document repositories (where good ideas go to die) transform into a living, conversational corporate memory. The intelligence from your best data sources now sits inside the model's weights. When a user queries a novel situation, the model provides guidance because institutional wisdom is built directly into its architecture. The modern knowledge manager is instrumental in making this happen and curating it overtime.

The Open Sesame moment has arrived.

_________________________________________

How many times have you searched your organisation’s knowledge base, found three different versions of the same policy, and had no idea which one was current? Or asked a question that clearly comes up every single week — and found nothing?

As a Knowledge Manager, this isn’t a rare moment. It’s the everyday reality of running KM in an organisation that moves faster than any team can document.

And every time it happens, the same thought creeps in: We have so much content. Why does it still feel like we can’t find the right content or right people to point us to the right content?

The honest answer is that most KM programs are built to store knowledge — not to actively manage its health. Gaps go undetected. Outdated content sits published for years. Nobody knows if an article is still worth the effort of keeping it around.

This is exactly where AI stops being a buzzword and starts being useful, everyday infrastructure.

1. Finding the Gaps We Can’t See

Normally, gap analysis depends on someone noticing a gap. A customer complains, an agent flags it, and only then does someone go check if an article exists. AI doesn’t wait for that. It listens all the time.

By scanning search logs, chatbot questions, and support tickets, AI can find out what people are actually asking — even when the same question is worded fifty different ways.

A simple AI-led gap analysis can:

group similar questions together, even if the wording is different, to reveal a gap hiding behind messy phrasing; spot articles that almost answer the question but stop just short; compare what exists against a list of all the topics that should be covered, to expose entire missing areas; rank gaps by how often they come up and how much they matter to the business, instead of guesswork

This is the difference between fixing a gap after someone complains, and knowing it’s there before anyone has to ask.

Case in point: At XYZCorp, agents kept getting asked about VPN errors — but each ticket used different wording (“can’t connect to VPN,” “VPN keeps failing,” “remote access not working”). No single article was written to catch all of these. AI grouped the tickets and showed there were over 200 such questions a month, with no clear article answering any of them well. That gap had existed for over a year, completely unnoticed.

2. Knowing What’s Actually True Anymore

Publishing an article isn’t the finish line. Content goes out of date. Policies change. Products change. And most KM teams have no easy way to know which articles have quietly become outdated or wrong.

AI can act as a constant accuracy check by:

pulling out facts, numbers, and steps from articles and checking them against the real source of truth (like product documentation or policy systems); flagging two articles that say different things about the same topic using an AI reviewer to catch old terms or steps that no longer make sense; sending anything flagged to a human expert to confirm — AI should never publish the fix on its own

The point isn’t to let AI decide what’s true and do all the work on its own. It’s to stop asking the knowledge team to re-read everything manually, all the time, just to catch what’s gone wrong.

3. Catching Content That’s Technically There, But Practically Dead

This is the quiet failure mode of KM — content that still exists, still shows up in search, still gets used, but refers to a policy or standard that’s no longer in force.

This matters most in places like presales, where proposal and RFP content constantly pulls from policy documents, compliance standards, and certification references. If the source policy has moved on and the content hasn’t, that outdated reference ends up in a client-facing document — and nobody notices until it’s already out the door.

AI can catch this by:

linking each article or proposal template to the exact policy or standard version it was written against watching for updates to policies, certifications, and standards, and flagging every linked article or template the moment a newer version is published setting simple rules so old, untouched reference content gets reviewed automatically on a schedule, not by chance building a simple dashboard that shows, at a glance, which policy-linked content is going stale, so nothing outdated makes it into a client-facing document

Case in point: XYZCorp’s presales team kept a standard security-compliance annexure that got copy-pasted into almost every proposal. When the underlying compliance standard was revised, nobody updated the annexure — it had been reused so often that no one remembered where it originally came from. AI flagged it the same week the standard changed, because the annexure was linked to that specific policy version. Without that link, an outdated compliance claim could have gone out in the next client proposal.
‍
The stakes get even higher in industries like pharma. A drug’s prescribing information — dosage, interactions, side effects, usage guidelines — can change after a regulatory update, a new clinical finding, or a safety alert. If a sales rep, call centre agent, or patient-facing article is still working off the old version, that’s not a minor inconsistency — it’s outdated medical guidance reaching a doctor or a patient. A pharma company linking every piece of content to its exact regulatory version, and getting flagged the moment that version changes, isn’t a nice-to-have. It’s the difference between staying compliant and putting someone’s health at risk.

4. Knowing What Should Exist — Before Someone Has to Ask

Filling gaps is reactive. The real shift is planning content ahead of time — AI suggesting what needs to be written next, based on patterns that would take a human months to spot.

This looks like:

pulling common themes from tickets, calls, and search behaviour into content suggestions recommending the right format, not just the topic — a simple decision-tree for troubleshooting, not another wall of text drafting a rough first version from scattered sources like emails or chat threads, for a human to finish and check comparing the product roadmap against current content, so articles are ready when a feature launches, not three weeks later

5. Measuring What Actually Matters

Page views were never a real measure of value. They only show attention, not impact. AI lets KM finally measure what content actually achieves.

This means tracking things like:

whether an article actually solved the query, or the customer still had to escalate; how much faster an issue gets resolved when the article is used; “zombie content” — articles that take effort to maintain but barely get used or barely help — as candidates to retire; whether content usage connects to real outcomes, like fewer tickets, faster onboarding, or lower churn

No, AI can never replace knowledge managers. Because they are the ones who feed AI knowledge and information that it requires to do its job of keeping the KB updated.

All what it does is— it frees us from being full-time content archaeologists, digging through what already exists, and lets us focus on what KM was always meant to do: getting the right knowledge to the right person, at the right time, without them having to go looking for it.

AI doesn’t fix KM by doing the writing for us. It fixes KM by finally giving us visibility into the health of what we’ve already built — and the foresight to know what’s missing before it becomes someone else’s bad day.

That’s not automation for its own sake. That’s KM finally working the way it was always meant to.

________________________________________