The foundation that makes your AI yours

The five layers of knowledge management

Knowledge management is not a single problem, so it does not have a single solution. It is a set of interconnected capabilities that span five distinct layers. Unlike Lumio, most organisations have invested unevenly across these layers. A common pattern: heavy investment in a data platform, clean pipelines, and reliable reporting – but the data is uncatalogued, its existence known to only a handful of teams, and inaccessible to the AI assistant the engineering team is piloting. Even though the technology is excellent, the knowledge is accessible only to those who already know where to look.

Understanding these five layers – and the most common issues within each – is the starting point for understanding why AI initiatives stall, and what it actually takes to make them work.

#1 Data foundation

Everything starts here. The data foundation is the raw material the organisation runs on. Records from day-to-day operations, internal documents and knowledge, financial and HR data, information from outside the organisation. Think of it as one big container holding everything the organisation produces, unsorted and unstructured, before anyone has decided what matters.

The problem is not that organisations lack data. It is that they have no clear picture of what they have, where it lives, or whether it can be trusted. Operational records sit in systems that only a handful of people know how to use correctly. Documents are scattered across shared drives, personal folders, and email threads. Finance and HR data lives in systems that were never designed to work together. Teams across the organisation are subscribing to the same external data sources without anyone knowing.

The result is an organisation that stores far more useful knowledge than it knows it has.

#2 Technology

Data in its raw form is not useful. It has to be collected, moved, transformed, and made accessible before anyone, human or AI, can do anything with it. This is where the technology layer comes in: everything that sits between the source systems where data is created and the tools where data gets consumed.

The technology layer is the infrastructure that makes data movable and usable. The storage environments, the processing pipelines that extract and transform data, the integrations that connect systems to each other, and the search and retrieval infrastructure that AI tools need to function.

Here too, the characteristic problem is not a lack of technology. As organisations grow, they adopt more systems to support more processes. Over time, data spreads across them, often ending up disconnected and scattered, with no shared architecture and no one owning the connections in between. Data moves between them through manual exports, scheduled email attachments, and brittle integrations that are vulnerable to system changes. When one of these connections breaks, the failure rarely announces itself. It shows up weeks later as a wrong number in a report, after decisions have already been made based on the bad data.

Modern data platforms like Microsoft Fabric, Databricks, Snowflake, and cloud-native processing environments exist to address this. But the technology alone does not solve the problem. An organisation that migrates its chaos into a modern platform still has chaos. Technology requires governance to function, and governance requires technology to scale.

#3 Semantic layer & metadata

This is the layer that most organisations underinvest in and that most AI implementations stumble over. The semantic layer is what gives data meaning – a catalogue that inventories what exists and who owns it, a taxonomy that classifies content consistently, an ontology that defines how the organisation's key concepts relate to each other, and lineage tracking that records where data came from and what has been done to it.

Without this layer, two teams can use the same word to mean different things – and an AI assistant cannot tell the difference between an active grant and a completed one or understand that ‘committed spend’ means something different in the finance team than it does in the programmes team. It will answer questions confidently using whichever definition it encountered most frequently in the data it was trained on.

The semantic layer does not create new knowledge; it makes knowledge understandable – to people and to machines.

#4 Governance

Governance is the layer that determines whether everything else holds together over time. It is the system of ownership, standards, accountability, and process that keeps knowledge trustworthy as the organisation grows and changes. This typically includes named owners for significant assets, quality standards enforced at the point of creation, a decision forum with real authority to set and enforce data standards, data contracts between teams that produce and consume data, and access policies that are actively managed.

Without governance, investments in the other layers tend to degrade over time. Technology investments can stall as people revert to familiar habits. Catalogues go stale as new assets are created without being registered. Quality standards exist on paper but are not always followed in practice. A knowledge base that looked well-organised at launch can look like a slightly more expensive version of the original chaos two years later. 

Governance is also the layer where the gap between intent and reality is often widest, and most consequential. Many organisations believe their governance is stronger than it actually is, because the policies exist even when they are not always being followed.

#5 User interface & applications

The top layer is where all the investment in the four layers below it becomes useful to the people who actually need it. This is where dashboards, search tools, AI assistants, and knowledge portals live. The tools people actually open, type into, and rely on.

This is where most organisations start. But in our view, this is where the work should land, not where it should begin. A search tool deployed on top of ungoverned, unstructured, semantically inconsistent data only returns noise. An AI assistant deployed on top of a well-governed, catalogued, semantically coherent knowledge base returns answers that people can trust, trace, and act on.

The organisations that find AI tools disappointing have almost always reached for this layer too early. Piloting AI tools here is fine, even useful, as a way to learn what good looks like and surface what the layers below need to support. But scaling them depends on the four layers underneath.

What this means in practice

Each of these five layers contains a set of specific, actionable capabilities – the concrete things an organisation needs to have in place for that layer to function. A data catalogue. Named data owners. Documented pipelines. A taxonomy applied consistently across content. A governance forum that meets and makes decisions. Taken together, they constitute an organisation that is genuinely ready to use AI reliably, at scale, with justified confidence in the outputs it produces.

Most organisations are not there yet. Most are not even sure where they stand. They have a sense that their data is messier than it should be and that their governance is weaker than they would like, but no clear picture of which specific gaps matter most, which sequence of investments will close those gaps fastest, or how far they actually are from being AI-ready in any meaningful sense.

Our approach: the knowledge management for AI readiness assessment

The knowledge management for AI readiness assessment gives organisations a structured, honest picture of where they stand across all five layers, creating the foundation for a practical roadmap.

The assessment works in three stages:

1. Establish the baseline

Working with a cross-functional group that includes both technical and leadership perspectives, we score the organisation's current capability across each of the five layers. We look at what exists on paper and what is actually being used. We look at where the formal picture and the operational reality diverge – because that gap is almost always where the most important governance work is needed.

The output is a current-state capability map: an honest, specific picture of where the organisation is strong, where it has gaps, and where it has blind spots.

2. Identify the priorities

Not every gap matters equally. Some capabilities are foundational – without them, nothing else works reliably. Others deliver value independently, in the short term, while deeper structural work is underway.

We work with the organisation to identify which gaps are creating the most friction today, which are creating the most risk for AI initiatives specifically, and which represent quick wins that will build confidence and momentum for the harder work ahead.

3. Build the roadmap

The assessment concludes with a sequenced, realistic plan – capability by capability, with indicative timelines, named owners, and clear dependencies. This creates a working plan that the organisation can begin executing the following week.

The assessment typically takes four to six weeks and involves structured interviews, documentation review, and a facilitated workshop with the leadership team. The output is a single, clear answer to the question that most organisations cannot currently answer: where do we start?

If your organisation is investing in AI – or planning to – the question is not whether you need this foundation but how far along you already are, and what the most direct path forward looks like from where you stand today.

That is exactly what the assessment is designed to tell you, so your AI can deliver results people trust and act on.