Enterprise AI, Grounded in Real Data
Ontology-Driven Enterprise AI
Terms such as Ontology-Driven RAG, Retrieval-Augmented Generation and Knowledge Graph Enhanced AI increasingly describe an important principle: general-purpose AI performs better when it can work from structured, validated knowledge about the domain it is trying to understand.
For enterprise material, product and service data, that knowledge cannot simply be invented by an LLM. It has to define what products are, how they are classified, which attributes belong to them, which values and units are valid, how concepts relate to one another, and how the resulting information should be represented in enterprise systems.
Refresh has been building, validating and applying this kind of structured Material Data Intelligence for two decades. The important question for many IT teams is therefore not whether to use an ontology. It is whether they really want to build one themselves.
Why enterprise ontologies are difficult to build
Creating a useful enterprise ontology is much more than defining a hierarchy of product classes. For tens of thousands of disparate materials, spare parts, services and finished products, it can require thousands of classes and relationships together with attributes, permissible values, units, synonyms, multilingual terminology, naming conventions, validation rules and mappings to external standards.
The difficulty becomes particularly clear in long-tail indirect and MRO data. A single enterprise may contain everything from bearings, valves and motors to laboratory equipment, electrical components, chemicals, PPE and specialist engineering parts — each requiring different technical characteristics and domain knowledge. This is not simple. Even your own business experts usually do not know all of the relevant data standards, classifications, attribute models and terminology needed to build a complete enterprise ontology.
The challenge is not a lack of expertise inside the business. It is the breadth and formalization required. No single team normally owns the cross-industry standards, attribute models, terminology, relationships and mappings needed to turn thousands of disparate materials, products and services into one coherent ontology.
This opportunity extends well beyond MRO or indirects. Different ontology requirements arise across Product Information Management (PIM), manufactured and configured products, procurement and spend management, engineering, maintenance, supply chain and sales.
A class library is only the beginning
Many organizations already have product classes, internal naming conventions or business experts with deep domain knowledge. That is useful — but it is not the same as having a complete enterprise ontology.
A usable ontology also needs the relationships between classes, the correct technical attributes for each class, permissible values and units, synonyms, multilingual terminology, naming rules, validation logic and mappings to external standards.
This is where the scale of the problem is often underestimated. Even very knowledgeable business experts usually do not know all of the relevant standards, attribute models and cross-industry terminology needed to formalize thousands of disparate materials, products and services consistently.
Refresh brings much of that structure already developed, validated and refined through two decades of real enterprise data-standardization work.
Its preconfigured Material Data Intelligence includes accumulated classes, attributes, values, dictionaries, taxonomies, ontologies, mappings, validation logic and naming standards that can be used alongside SAP, Oracle, IBM Maximo and other enterprise platforms.
This allows IT and business teams to apply modern AI and LLM capabilities without first having to recreate the underlying domain intelligence themselves — and to deliver measurable outcomes in areas such as inventory reduction, spend visibility, product information, data migration, maintenance and master-data governance.
Read about two decades of Refresh AI and Material Data Intelligence »
From ontology to knowledge graph
For organizations developing their own knowledge graphs, RAG architectures or enterprise AI platforms, Refresh can also provide the underlying material-data structures and relationships.
These can include the ontological concepts and triples needed to connect materials, product classes, technical attributes, values, units, manufacturers, standards and related enterprise information.
In this role, Refresh is not another system of record. It provides a specialist domain-intelligence layer that can make the enterprise systems and AI platforms around it more useful.
Dictionary, taxonomy & ontology — how do they relate?
A Refresh ontology can incorporate all three. For an AC motor, for example, it can define the material class, multilingual terminology, relevant attributes such as voltage, speed and number of poles, valid units and values, attribute priorities and mappings to other standards. Those mappings can include UNSPSC, tariff or HS codes, spend categories, industrial classifications and other customer-specific standards.
The result is not simply a list of words. It is a structured, machine-readable model of the material domain.
A Deterministic Natural Language Generation blueprint
The same structure can also control how standardized descriptions are produced.
Classes define the relevant attributes. Validation rules define acceptable data. Attribute priorities determine the order in which information should appear. Multilingual dictionaries define the approved terminology.
Together, these elements provide a Deterministic Natural Language Generation (NLG) blueprint capable of producing consistent short descriptions, long descriptions and purchasing descriptions without asking a generative model to invent the underlying specification.
This is particularly important for enterprise master data, where repeatability and correctness are often more valuable than creative language generation.
Deterministic Guardrails Reduce LLM hallucinations
Large Language Models (LLMs) are probabilistic: they predict likely language patterns rather than verify engineering truth. They do not inherently know that an AC motor should not have a “number of poles” of 3, or that 400 V and 400,000 MV represent radically different engineering realities.
That is why enterprise AI needs structured domain knowledge, permissible values, units, relationships and validation rules around the model — not just a powerful model on its own.
Refresh provides a structured truth source against which AI output can be constrained and validated. Instead of asking an LLM to freely generate a product description, the system can provide the relevant class, permitted attributes, accepted values, units, terminology and generation rules first.
The LLM can then be used selectively for tasks where its flexibility is valuable — such as interpreting unstructured legacy descriptions or purchase-order text — while Refresh validates the resulting structured information against the ontology.
This combines the flexibility of generative AI with deterministic guardrails designed to prevent invented specifications from entering governed enterprise data.
Cross-Lingual Semantic Mapping
Multilingual enterprise data introduces another problem: translating words is not always the same as translating meaning.
Refresh can map concepts at the semantic ID level rather than repeatedly asking an AI model to translate the text itself.
An attribute such as RPM, for example, can map directly to its approved German, French, Spanish or other terminology while continuing to represent the same underlying concept.
This Cross-Lingual Semantic Mapping improves consistency across languages and reduces the need for repeated generative processing — saving tokens, processing time and AI cost.
Symbolic AI + Sub-symbolic AI
Refresh deliberately combines different forms of intelligence rather than relying on a single technology for every problem.
Symbolic AI provides logic, rules, exact data, ontologies, classifications and deterministic validation.
Sub-symbolic AI — including machine learning, neural networks and LLMs — provides powerful pattern recognition, language interpretation and inference.
Refresh uses each where it is strongest. A language model may interpret a messy source description, while the ontology determines which attributes actually belong to that class, which values are permissible and how the final governed record should be constructed.
The result is modern AI anchored by structured enterprise knowledge rather than asked to rediscover that knowledge every time it runs.
Designed to fit existing enterprise architecture
A Refresh implementation is designed to reuse the systems and infrastructure customers already have rather than require another system of record.
Refresh can operate alongside environments including SAP and S/4HANA, Oracle, IBM Maximo, Coupa, Infor and other enterprise and legacy systems.
Standard ERP programs do not need to be modified simply to use Refresh intelligence, and deployments are designed to minimize custom development, infrastructure requirements and ongoing demands on internal IT teams.
This makes Refresh suitable both for major transformation projects such as ERP and S/4HANA migrations and for focused business-led initiatives where procurement, maintenance, supply chain or product teams want measurable improvements in their data without creating another large IT program.
Enterprise AI, Proven on Global ERP Systems
Refresh provides the accumulated domain intelligence needed to move from a general-purpose AI capability to a controlled enterprise data solution, but don't take our word for it.
The Refresh Software is currently being used at many leading reference customers today running for example SAP, S/4HANA upgrades, Oracle, Coupa, Infor and other leading ERP system improvement projects. Take a look at how the global leader in fine paper, the world's largest mining and commodities company, and a leading chemical giant are now using Refresh™ to analyze and successfully drive out new savings from multiple systems. Also, medium-sized companies ranging from a top United States agri-business running SAP and ARIBA through to Oxford University running Oracle have all unlocked savings from Refreshing their data.
See it working on your data
The easiest way to understand the difference is to see Refresh in person. Contact us below for a demo and discussion.