AI Needs Context, Not Just Data

Most organisations already have a solid data foundation. Many have gone further and built an analytic semantic layer on top of it. Governed metrics, clean joins, and consistent definitions for their BI tools.

That is valuable work, but it is not enough for AI. A semantic foundation is needed.  

AI does not just need to know how to calculate the right number. It needs to know what that number means in your business. It needs your language, your relationships between concepts, and the context your people carry in their heads every day. 

Without that, AI will confidently give you the wrong answer. Or worse, the right answer described in the wrong way. 

Why good metrics are not enough for AI 

An analytic semantic layer solves a real problem. It defines how tables join. It governs how metrics are calculated. It makes sure that when someone asks, “what was revenue last quarter?” the answer is accurate and consistent, whether it comes from a dashboard, a spreadsheet, or an AI assistant. 

That layer handles the structured side of your data. It tells AI how to calculate. But it does not tell AI how to interpret what it finds. And it does not tell AI how to talk about it. 

Metrics and meaning are different problems. The meaning is what trips organisations up when they start using conversational AI, or making decisions with it.  

Where AI gets it wrong without context 

Take a pharmacy business. “Dispensed” means “sold", but only if it's not an off-the-shelf product. Everyone in the business knows that. AI does not. Unless someone has made that relationship explicit, AI will treat those as two separate concepts. The query might return the right data, but the AI might describe it in a way that confuses the person reading it. Or it might miss relevant data entirely because it did not know to look under both terms. 

Or consider “legacy customer.” In general language, that sounds like a long-standing loyal customer. Inside your business, it might mean something very specific. Perhaps it refers to customers on an old pricing model, or customers migrated from an acquired company. AI will default to the general meaning unless you tell it otherwise. 

These are not edge cases. Every organisation has many terms like this. “Member” might mean an employee, a subscriber, or a loyalty program participant depending on the division. “Case” could be a patient episode, a legal matter, or a support ticket. “Facility” could be a building, a production line, or a capability. 

The analytic semantic layer will not catch these. The joins are fine. The metrics are fine. The language is the problem. 

AI can get the number right and still lose trust 

This is where it gets subtle. 

Without context, AI can misunderstand the question itself. Someone asks about “active members” (meaning current subscribers), and AI interprets it as anyone who logged in recently. The answer comes back confident, well-formatted, and completely wrong. Not because the data was bad or the query failed, but because AI answered a different question to the one that was asked. 

Even when it does understand the question, the way AI frames the answer depends on context it may not have. 

If your pharmacy tracks dispensing volume but AI describes it as "sales", people notice. If AI calls something a “legacy system” when internally it's called "Orion", the language jars. If it frames a recommendation using the wrong business terms, people do not trust the output. Even though the number underneath was right. 

For decision-makers, wrong language is the same as a wrong answer. It signals that the tool does not understand the business. Once that trust is lost, adoption stalls. 

This is why context is not a nice-to-have. Context is what makes the difference between AI that people use and AI that people abandon. 

What most organisations are missing is what I call a semantic foundation. 

What is a semantic foundation? 

A semantic foundation includes your governed metrics and structures, but it goes further. It captures the business language, the synonyms, the relationships between concepts, and the context that shapes how your organisation thinks and communicates. It is the layer that lets AI understand that “dispensed” and “sold” can mean the same thing. A semantic foundation identifies “legacy customer” as a defined category, not a general description. 

The data world is only just catching up to what knowledge management and ontology practitioners have been doing for decades: 

• defining meaning 

• managing taxonomies 

• making context explicit 

If your organisation has a knowledge or content function, that expertise is directly relevant here. The principle is straightforward. Define meaning once. Govern it. Let both humans and machines use it. 

The value increases as context increases 

You do not need to build a complete semantic foundation before you start. But you do need to recognise that the next layer of value comes from making your business context machine-readable. 

Start with the terms your people already know. The ones where the internal meaning differs from the general meaning. The ones where two teams use different words for the same thing. Make those explicit. That is your starting point. 

The more context you make explicit, the more AI can do safely. The less context it has, the more likely it is to get the language wrong, lose trust, and stall. 

Good data is not enough. Good metrics are not enough. AI needs to understand your business the way your people do. That understanding is the semantic foundation. Building it is the next step forward. 

In the next article, we look at why AI fails when meaning is unclear and what it takes to build trust by design.