Where AI Transformation Begins: The Pre-Ontology

This article is adapted from Yeoul (@sckimynwa), co-founder of Tiro, originally posted as "Pre Ontology Layer for AI Transformation" on X.

Anywhere AI Transformation comes up these days, one word follows on its heels: Ontology. It is being used almost like a magic keyword.

The problem is that the word is abstract, and discussion of "what an ontology actually is" or "how you build one" is hard to find. It just floats around as the Magic Word that supposedly solves every AI problem.

This article looks at the step that comes before that magic keyword. Pre-Ontology.

Two truths about building an ontology

Two truths that get overlooked:

First, an ontology is not something someone outside the organization can build for you. It is not a deliverable that a consulting firm can drop in after one or two visits.

Second, the process of adopting it matters more than the artifact. The thing being built is less important than how it gets built.

The reason why becomes clear if you look at the etymology of the word.

Ontology is, really, "an agreement on language"

Ontology is originally a philosophy term that means the study of being. In an organization, it relates to "how a group will define a given entity inside the organization, and how those definitions will compose into the organization's world." It is, fundamentally, an agreement.

The key is that the concepts floating inside an organization are not physical things. They are linguistic agreements that people have created, and that is exactly what makes defining and adopting an ontology hard.

Why the word "customer" means different things to different departments

The word "customer” is abstract. Its meaning differs even within the same company:

• For Sales, a customer is the legal entity that signed the contract.

• For Product, a customer is an active user who comes in every week to use features.

• For Support, a customer is the person who opened the latest ticket.

Same company. Same word. Three different definitions across three departments. And this doesn’t just apply to "customer." Revenue, defect, lead, active user – nearly every term you can name gets defined differently from one team to another.

In that case, what happens when you tell an AI to "analyze our customer data"? It is not even clear whose definition of "customer" the answer should be based on. Without agreement on a given entity, neither a human nor an AI can understand the entity well enough to engage with the outside world.

What Palantir actually did at Airbus

A frequently cited example of successful ontology adoption is Palantir at Airbus. The outcome "A350 delivery speed improved by 33%" is striking.

However, what Palantir actually did inside Airbus was not the 33%. Before the 33% became possible, they got thousands of practitioners to use "defect," "part," and "delay" with the same definition.

Here is the real face of an ontology. It is not a deliverable.

It is simply realized internal agreement.

It is top-down, and that is exactly the trap

So, how is that agreement made?

Ontology is fundamentally top-down. It looks like consensus, but in reality, it’s the result of someone higher up making the decision to say "customer means this."

For that top-down decision to be correct, the decision-maker has to know exactly how the practitioners on the ground are actually using the word.

That is not as easy as it sounds. In an organization with several layers of decision-making, it is hard for senior managers to observe the practitioners' language as it is actually used. Information gets abstracted at every layer, and the more abstracted it becomes, the more the word ends up an empty shell.

This is the biggest bottleneck for AI adoption: top-down decisions without the bottom-up context they need. If you cannot understand the practitioner's language, you cannot define their work, and an AI agent added over undefined work will not produce trustworthy results.

An organization with a firm definition vs. one with a clumsy one

A successful AI Transformation assumes an agreed ontology, so closing this gap is not optional.

Once an ontology takes hold, every piece of data and every AI judgment on top aligns to the same language. It compounds over time.

Conversely, in an organization where definitions are flexible, more data means more noise. How fast you close this gap will, in the end, decide your competitive position.

So, Pre-Ontology

A new layer is worth naming here: Pre-Ontology.

If ontology is the shared internal language an organization agrees upon, then pre-ontology is the raw material — the scattered words and phrases in everyday use. It’s how practitioners actually say “customer,” the concepts that word connects to, and the decision-making moments where it surfaces. Pre-ontology is not fixed; it’s a living accumulation of language as it’s used inside the organization.

Put differently: if ontology is a top-down decision being stamped, Pre-Ontology is the material being gathered bottom-up.

Ultimately, the quality of the decision is proportional to the quality of the material. In any company of meaningful size, the success of an AI transformation will be proportional to the quality of ontology adoption, and to accelerate that adoption, Pre-Ontology has to come first. This is exactly why no external consulting engagement can hand you an ontology in one or two visits.

That material mostly comes out of people's mouths

Where does the raw material for pre-ontology — the practitioner’s living language — actually reside?

The answer is surprisingly ordinary: in everyday conversations. Not just in formal meetings, but in the quick exchanges that happen throughout the day. It’s in spoken language that the organization’s real vocabulary takes shape

Slack threads and emails capture a fraction of that context – roughly 10% of the full business context. Why a decision was made, what conditions were attached, or how a senior manager explained “our customer” to a new hire, that nuance lives in speech and evaporates the moment the conversation ends.

The space Tiro fills

Inside this framework, the space Tiro wants to occupy sharpens.

Capture every important conversation in the organization, and process it in context. Record accurately on any interface (phone, smartwatch, desktop), store securely with proper permissions, and let the organization pull its own language and context back out whenever and however it is needed. Tiro is the kind of tool this era demands.

Tiro is not just a meeting notes app. It is the basis for organizational Pre-Ontology.

Closing

In the present day, when language models are more capable than ever, paradoxically, the competitive edge lies in language.

The organizations that siphon practitioner language from the bottom up as a basis for Pre-Ontology and use it to accurately develop an Ontology will be able to take full advantage of this new wave.

In your next AI Transformation project, before you pick a model, pause and ask one thing. "Within our organization, is everyone using the word 'customer' the same way?"

If you hesitate even briefly, the starting point is not the model. It is Pre-Ontology.

Get started