Webinar
Put AI on the Data You Already Have
Summary
OEE, energy and alarm insights from the historian you already run.
Recorded September 15, 2026, with Itanta Analytics.
Most manufacturers are not short on data. They are short on answers. The data sits in historians, SQL databases, SCADA and MES, in systems that were never designed to be read together, and getting one answer out of them usually takes an engineer, a spreadsheet and a week.
In this session, Itanta Analytics and INS3 showed how Itanta’s Prompt-to-Insights (PTI) puts an AI layer on top of that existing infrastructure, without replacing it and without copying the data. The session includes a live demonstration on a Proficy Historian connection hosted by INS3, an OEE application from automotive manufacturing, a benchmark against general-purpose AI models, and what a first trial looks like.
Recording
Key Takeaways
- No rip and replace. Prompt-to-Insights adds an intelligent layer over the historians, databases and systems already running on site.
- Zero data duplication. The agents query data where it lives, and the plant data stays on premise.
- Data-blind by design. The platform works from metadata and schema, so the raw data is never sent to a language model.
- Built for industrial context. On Itanta’s internal benchmark, PTI answered 90.2% of questions correctly, against 15.8% and 17.7% for general-purpose AI models.
- Applies across the plant: OEE, energy, alarms, downtime and root cause, quality and SPC, batch analysis, shift reporting.
- A trial is small and fast: one historian connection and one pressing question, with results on your own data in about two weeks.
Q&A
Access is role-based and modular. For each application, an admin decides who can connect data, edit the knowledge graph, ask questions, and create, edit or delete dashboards. The typical setup is an editor who administers the application and adds viewers. Several people can build the knowledge graph together, and it is versioned. Single sign-on works with Microsoft and Google.
It can use any source PTI has a connector for, but the usual source is a historical data store such as a historian, a SQL database or an ERP. Historians are queried in place, so PTI does not keep its own copy. Live data can also come in over MQTT, including Sparkplug B. The demo showed an Ignition system streaming plant energy data that way.
Both. A client component on your premises queries the data and renders the visualizations, so the data stays in the plant. A cloud component receives only metadata, which it uses to turn a plain-language question into a query that runs on premise.
Only the schema. PTI builds a description of the data’s structure and metadata, and that is what goes to the language models. The raw data is never sent.
Not yet. Today the models run in the cloud. Itanta is building small language models for specific use cases, planned for later this year, which will allow a fully air-gapped deployment.
There’s no need to recreate it. Adding a data source on the existing connection reloads every tag on the broker, and you pick the new ones. You can also run several connections side by side, for example production data from MES and maintenance data from a maintenance system.
Continue the Conversation
Your journey with AI doesn’t have to stop here.
Talk to experts: Book a Discovery Session with INS3 — get a tailored walkthrough of how Itanta can fit into your plant and deliver measurable ROI.
Whether you want to test the software, or explore a pilot project, we’re here to help you take the next step.
Share on :
Keep On Reading
Subscribe to Our Newsletter
Subscribe to our weekly newsletters to get updates regarding our new production, behind the scene process of our art creation and much more.
After submitting this form you will receive an e-mail with a confirmation link that you must click to complete your request. Detailed information on processing and cancellation can be found in our privacy policy.