Delight the business with fully cooked Enterprise AI

Apart from being in an obvious hype cycle right now, there is indeed real value in Ontology-Driven AI Data Standardization. A bit of a mouthful, but maybe you've heard of some important parts of it since its currently a hot AI topic. Perhaps you've heard of Ontology-Driven RAG (Retrieval-Augmented Generation) or Knowledge Graph Enhanced AI as more and more vendors want to sell these tools to you. But the real science of practical AI is not so much about the empty tools themselves - it's more about how the correct data inside these tools can drive amazing algorithmic automation that powers the tagging of each individual enterprise data line item to your intended standards. 

Not all ontologies are created equal; you need a business-specific one that's tried and tested on millions of lines over many years in systems similar to yours. Refresh presents a unique opportunity for IT teams to use pre-defined battle tested industry specific ontologies in SAP, Oracle or other enterprise systems to quickly leverage a proven AI technology project delivering tangible business benefits, getting real experience with AI and LLMs, using your existing tools. The alternative is to try to build your ontologies yourselves, but beware, this is not simple and even your business experts do not know the standards, especially for the mine-field of Product Information Management (PIM) Master Data Standards with very disparate (like tens of thousands disparate) MRO / long tail indirect materials. This is a serious undertaking for Product Management (Sales), Spend Management (Procurement) and Operations (Engineering and Maintenance), where the correct ontologies can make massive new savings in the AI era. 

But what is an ontology? Well, the word is as old as the hills, (like Aristotle-old), but in your Enterprise IT domain for services or materials you can think of them this way: a Dictionary gives definitions, a Taxonomy creates a family tree or filing system, and an Ontology builds a living schema of all of these plus more logical rules and relationships for this industry specific enterprise domain. So now we have three words: dictionary, taxonomy and ontology; how do they relate? 

Dictionary: A complete list of unique words or terms paired with text descriptions or definitions. It answers “What does this word mean, what is this thing?” 
Taxonomy: A hierarchical classification system that groups things into broader and narrower categories (like parent/child or genus/species). It answers “Where does this thing belong?” 
Ontology: A formal network of concepts, properties, and specific logical rules, including Dictionaries and Taxonomies, that allows computers to perform automated reasoning. It answers “How do these concepts about the thing interact and what do they mean in context?”

While distinct, they can build on each other. An ontology can incorporate a taxonomy as its hierarchical backbone, adding rich, specific constraints from the dictionary so that artificial intelligence can draw logical conclusions, rather than just look up categories. Say we have a dictionary defining an item class like MOTOR:AC (in every language), then also providing say the top 20 attributes for MOTOR:AC, like RPM, Volts, number of poles, etc (also in every language). Then every attribute value that is expected (again in every language). Then every unit of measure. And what about every priority for each attribute so that short and long texts can be generated in a priority order. And maybe this maps to some other things like tariff codes, spend codes, UNSPSC codes, industrial certification codes and other standards to boot ... all in one file system... what’s that called? That right there is an ontology.

Because our system with its in-built ontologies combine definitions, multilingual attributes, data types, strict validation rules (units of measure, accepted values), and generation logic (priorities for long/short text), it crosses several thresholds: firstly it can be used as a Classification Standard & Class Library: defining the category (MOTOR:AC), the specific attributes (RPM, Volts, number of poles), the valid units (V, kV, mV), and map them to HS/Tariff, UNSPSC, Spend codes etc. A Domain-Specific Ontology: defining the "schema of reality" for a product or service. Because we include strict logic (e.g., if it's an AC Motor, it must have Volts; a specific value triggers a specific short-text rule), and so now you also have a Semantic Model. This functions as a Deterministic Natural Language Generation (NLG) Blueprint: by adding these attribute priorities to generate for example short/long texts, you have created a template system for automated cataloguing.

But hang on, we've saved the best for last...

Is This the Missing Ingredient to Stop LLM Hallucinations?
Yes, absolutely. You have hit on one of the most critical frontiers in modern Enterprise AI:  Large Language Models (LLMs) are probabilistic—they predict the next most likely word based on statistics, not truth. They do not naturally understand that a motor cannot have a "number of poles" of "3" or that 400V and 400,000MV are drastically different engineering concepts. By feeding your exact structure into an AI pipeline, you fix this through an architecture known as Ontology-Driven RAG (Retrieval-Augmented Generation) or Knowledge Graph Enhanced AI. Also rolls off the tongue. We just call it Refresh Software and it comes in Desktop and Server format.

Here is how Refresh acts as the "missing ingredient" to kill hallucinations. The strict Industry Specific Data Standard Ontology acts as a guardrail / truth source before any unstructured Prompt / Request to any LLM Process, ensuring 100% Accurate, Deterministic Text Output. This represents Grounding Reality (Deterministic Guardrails) so instead of asking an LLM to "write a product description for an AC Motor," you pass the LLM the structured ontology with the item and you basically tell the LLM: "You are only allowed to use these attributes in the Ontology. You must follow the priority order (and maybe some "extra" things we'll save for a demo like Tariff codes or UNSPSC, or duplicates using your own PIM or part number patterns). The ontology acts as a strict cage for the AI model. Cross-Lingual Semantic Mapping is also important because your standard maps RPM to U/min (German) for example at the ID level rather than translating the words textually, so the AI doesn't have to guess the translation. This means It maps the token directly to the pre-defined, human-verified dictionary entry, eliminating translation hallucinations, but critically, also saving tokens, saving time, saving AI cost.

Since you have attribute priorities, you don't even need an LLM to generate the baseline text. You can use Refresh's traditional non-AI code to assemble the "short text" deterministically and then use the LLM only optionally to parse any useful but unstructured long texts or PO texts, then use the ontology to validate that the LLM didn't invent new specifications during the rewrite. In short, the system provides the Symbolic AI (logic, rules, exact data) to anchor the Sub-symbolic AI (neural networks, LLMs). Merging these two is widely considered a holy grail for enterprise-grade generative AI. 

For a true lowest total-cost-of-ownership, any approach you choose should re-use your existing IT assets, not demand new ones. For a Refresh project, no additional hardware is needed, no custom development is needed and your existing system makes calls to Refresh, not the other way around so there are no security concerns. Of course, there are no modifications to standard ERP programs. Since all of these often-hidden support costs are minimized by the software design, your costs and risks are much lower than any other approach. IT’s overall headcount for a typical Refresh project is only a few days so your already stretched IT teams are unaffected, and the business is happy fast.

The Refresh Software is currently being used at many leading companies and governments running for example SAP, S/4HANA upgrades, Oracle, Coupa, Infor and other leading ERP systems. 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 their data.

Book a quick demo of the software today using the contact form below or browse some example customer case studies and drop us a question if you have any.