Ambitious ontologies that are designed from the top down to cover an entire enterprise, have a high risk of never reaching completion in our experience. This is because the ontology’s designers end up being overwhelmed by complexity, and tend to lack a near-term objective that is well-scoped and valuable.
When an ontology is used merely as a reference document for humans, rather than embedded in an AI system to actively constrain outputs, the chances of it being under-utilised are also high, which is why we tend to advise against it.
What are ontologies for?
Ontologies are knowledge frameworks that define what words and concepts mean, and map out how they relate to one another. They help humans and machines to interpret the intended meanings of words and data, based on their context, in a way that is consistent.
From a human perspective, ontologies helps cross-disciplinary and multilingual teams to collaborate with lower risk of misunderstandings. From a vertical AI agent perspective, they enable semantic interoperability. In other words, they enable machines to draw logical conclusions more reliably from a range of different datasets, because meanings are aligned. In our experience, this enables AI to take on more valuable tasks, from decision support to data integration.
Are ontologies worth it?
We’ve noticed growing awareness of the need to ground large language models among companies that are trying – and often struggling with – the realities of implementing AI. This has brought ontologies back into consideration, beyond their established niches in medical research and advanced manufacturing.
Palantir’s CEO Alex Karp is a noted cheerleader of ontologies. Only with structured knowledge, he argues, will the outputs of AI be trusted and accurate enough to enable adoption at scale for mission-critical tasks. At Davos 2026, Karp said: ‘If you just buy LLMs off the shelf and try to do any of these things that are regulated, it’s not precise enough. What you’re going to see, especially in America, is people trying to do something like Ontology by hand.’
There is independent research to back him up. The unexpected benefits of implementing a large-scale enterprise ontology for product catalog and engineering data, was described in a 2025 case study. There is little doubt however that the pitfalls can be costly.
What puts people off ontologies?
Ontologies have been overlooked during the recent AI boom despite the benefits they can bring of greater accuracy, reliability and explainability. This is often because AI startup founders are in a hurry. We have found that they tend to be very comfortable with gnarly technical work, but much less comfortable with messy and time-consuming human interactions involving non-technical subject matter experts (SMEs).
The fact is, building consensus among SMEs is the foundation of an effective ontology. It is the reliable way to reach lasting agreement on the definitions of concepts and things in a domain, that AI can reason over. In our view, it is also the way to deliver AI services that domain experts feel they have ownership over, and can trust.
In other words, consensus-building is the human groundwork that can drive successful AI adoption and scale.
How do I create an enterprise ontology that actually works?
A value-driven approach to building an ontology, or AI trust layer, is to start small. Set an objective that users care about, give it a narrow scope, and make sure it is achievable. Once this single branch of the ontology tree is complete, it can then be tested, validated, and used as a case-study to work on another branch. In this way the ontology can grow organically according to what delivers value, rather than attempting to follow an overarching master-plan of the ontological tree that often never gets finished or integrated into operations.
What tools can help me build an ontology?
Emerging ontology-as-a-service (OaaS) platforms – such as our own – can help by automating elements of the discovery process, providing digital approaches to consensus building such as the Delphi Method, and assisting in the ontology’s maintenance and upkeep as new concepts surface over time across the domain.
One way to kickstart the creative process of developing an enterprise ontology, is to ask your SMEs our three key questions.