Every conversation about AI in manufacturing eventually runs into the same wall: the answers are only as good as the data feeding them. You can have the smartest natural language interface in the world, but if it can't reach the tag coming off your filler line, the batch record sitting in a CSV, or the historian rollups parked in Snowflake, it's just a clever chatbot with nothing useful to say. That's why Prompt-to-Insights (PTI) doesn't start with a question. It starts with your data. The Integrate stage is the first phase of PTI, and it exists for one reason — to make sure that when a plant manager, an OT engineer, or a CIO finally types a question into PTI, the system already has trustworthy, structured, plant-aware data sitting underneath it. No question can be answered well on top of disconnected sources. So before PTI analyzes anything or visualizes anything, it integrates. Here's what that looks like in practice today. Built for the Reality of Industrial Data Manufacturing data doesn't live in one place, and it never will. A single plant might be streaming live tag values over MQTT from a packaging line, pulling shift summaries out of an Excel sheet a supervisor still updates by hand, holding years of historian data in Snowflake, and exchanging quality records as CSV exports between systems that were never designed to talk to each other. That's not a failure of the plant — that's just how industrial environments evolve over decades. PTI is built to meet that reality head-on. Its Integrate layer is standalone — it doesn't depend on the Itanta Analytics Core Platform or any other backbone to function. You can point PTI at the sources you already have, in the formats they already exist in, and start asking questions. No rip-and-replace, no eighteen-month data lake project, no waiting for IT to build a new pipeline. Today, PTI's Integrate stage supports four core source types, each chosen because it covers a real pattern we see on shop floors and in enterprise data stacks every day. MQTT — Live Data from the Plant Floor MQTT is the language the modern shop floor speaks. It's lightweight, it's built for machine-to-machine communication, and it's increasingly the default for how PLCs, SCADA systems, edge gateways, and Unified Namespace architectures publish real-time data. When PTI connects over MQTT, it's tapping directly into the live pulse of the plant. Tag values, machine states, production counts, alarm conditions — whatever is being published to the broker, PTI can subscribe to live and bring into its analysis layer. That means when someone asks "Which line had the most unplanned stops in the last four hours?", the answer reflects what's actually happening on the floor right now, not what a report said yesterday morning. MQTT is also the one source where PTI lets you define a custom namespace. If your team already organizes tags by site, area, line, and asset — or if you're building toward a Unified Namespace — PTI respects that structure and lets you map it explicitly, so questions can be asked in the language your plant already uses. Critically, MQTT data stays on-premise. PTI subscribes to your broker inside your network. Nothing about your live production data has to leave the four walls of the plant for PTI to make sense of it. Snowflake — For the Enterprise Data Layer Plenty of manufacturers have already invested in cloud data warehouses. Production history, ERP extracts, quality data, energy consumption, maintenance logs — it all gets piped into Snowflake so the business has one place to run analytics. The problem is that "running analytics" usually means waiting for someone who knows SQL and the schema to write a query, validate it, build a dashboard, and hand it back. PTI's Snowflake connector closes that gap. Once PTI is pointed at the relevant tables, anyone authorized to ask can ask — in their own words — and get back a chart or table built directly from the warehouse. And just as importantly, the data never leaves your Snowflake server. PTI queries it in place. Your governance, your access controls, your security perimeter — all of it stays exactly where it already is. This is the land-and-expand entry point for a lot of organizations: the data is already centralized, the governance is already in place, and PTI just unlocks it for a much wider audience. CSV and Excel — Because Reality Is Messy Anyone who has spent time in a real plant knows that not everything lives in a database. Shift handover logs, quality check records, downtime reasons, supplier deliveries, manual batch notes — a huge amount of operational truth still moves around as CSV exports and Excel workbooks. Sometimes the source system only offers an export. Sometimes the workflow has been that way for fifteen years and it works. Sometimes a supervisor genuinely finds Excel the fastest way to capture what happened on their shift. PTI doesn't pretend that data doesn't exist. The CSV and Excel connectors let users bring those files directly into the same analysis surface as the live MQTT streams and the Snowflake tables. Suddenly a question like "Compare last month's downtime reasons from the maintenance log against the production output in Snowflake" stops being a three-day analyst exercise and becomes a single prompt. Like MQTT, CSV and Excel data is handled on-premise. The files stay inside your environment, PTI reads them where they sit, and nothing about that operational record has to be uploaded to an outside service. youtube A Quiet but Important Security Story Put the four connectors together and a pattern emerges that matters enormously to any CIO or OT security lead evaluating an AI tool: your data is never exposed outside your enterprise systems. MQTT, CSV, and Excel data lives on-premise and is processed on-premise. Snowflake data stays inside your Snowflake server. PTI doesn't shuttle your production records, batch histories, or shift logs off to some external model to "learn" from. The intelligence comes to the data, not the other way around. For regulated industries — pharma, food and beverage, oil and gas — and for any manufacturer who has been (rightly) cautious about handing operational data to AI tools, this is a meaningful architectural choice. Two More Pieces Worth Knowing About The Integrate stage also includes two capabilities that deserve a quick mention, even though they each merit a deeper post of their own: SME Review lets subject matter experts validate and refine how PTI understands your data — the meaning of tags, the relationships between assets, the context that turns raw values into operational truth. It's how plant knowledge gets baked into the system instead of being lost. SOP Upload allows standard operating procedures to be brought into PTI alongside the data itself, so answers can be grounded not just in what happened, but in what should happen according to your own documented practice. Both are part of why PTI feels different from a generic AI tool the moment you start using it. More on each of these soon. Why This Combination Matters If you look at PTI's Integrate stage as a whole, you'll notice the connectors aren't arbitrary. They cover the three layers where manufacturing data actually lives: - The real-time layer on the shop floor, where MQTT carries live machine and process data. - The enterprise layer in the cloud, where Snowflake holds the warehoused history that finance, operations, and leadership rely on. - And the operational edge, where CSV files and Excel sheets continue to capture the human, ground-level reality that no automation has fully replaced. A natural language analytics tool that only handles one of these layers will always disappoint someone. PTI handles all three out of the box, and the Integrate roadmap will keep expanding from here as customer pilots reveal new patterns. What This Means for the People Asking Questions The whole point of getting Integrate right is that the user never has to think about it. A plant manager asking "What was my OEE on Line 3 yesterday?" doesn't care whether the OEE numbers came from a live MQTT feed, a Snowflake table, or a manually maintained Excel sheet — and they shouldn't have to. They just want the answer, fast, traceable, correct, and built from data that never had to leave their own environment. That's the promise PTI is built around. Integrate is the unglamorous half of that promise — the work that happens before the prompt is ever typed. But it's the half that decides whether the answer is worth trusting. You already have the data. PTI's Integrate layer is what finally makes it usable. Ready to see it on your own data? Itanta Prompt-to-Insights is in early access now. Connect your sources, ask your first question, and find out how fast "we should look into that" can become "here's the answer." Visit itanta.ai to get started. Frequently Asked Questions What data sources does Itanta Prompt-to-Insights currently support? PTI's Integrate layer supports MQTT for live shop floor data, Snowflake for enterprise cloud warehouses, and CSV and Excel files for operational and manually maintained data. The connectivity layer is standalone and does not require the Itanta Analytics Core Platform. Does PTI work without the Itanta Analytics Core Platform? Yes. Prompt-to-Insights is a completely standalone product with its own integration layer. Customers can deploy PTI independently and connect it directly to their existing data sources. Does my data leave my environment when I use PTI? No. MQTT, CSV, and Excel data is handled on-premise inside your own network, and Snowflake data stays inside your Snowflake server. PTI brings the intelligence to the data rather than moving the data out, so nothing about your production records has to leave your enterprise systems. Can I define my own data hierarchy in PTI? Yes — for MQTT sources, PTI supports custom namespaces, so you can map your tags to the site, area, line, and asset structure your plant already uses, including Unified Namespace architectures. Can PTI combine data from multiple sources in a single question? Yes. Because all connected sources feed into the same analysis layer, a user can ask questions that span live MQTT data, Snowflake tables, and uploaded CSV or Excel files together — without writing any code or SQL. Does PTI understand the context of my specific plant? Yes. Through SME Review, subject matter experts can validate and refine how PTI interprets your data, and through SOP Upload, your standard operating procedures can be brought into PTI so answers are grounded in your own documented practice. Who is PTI designed for? PTI is built for three personas across manufacturing organizations: plant managers and shift supervisors who need answers immediately, digitalization heads and OT/IT engineers who need faster root-cause analysis, and CIOs, directors, and VPs of operations who need dashboards and summaries without using analytics tools directly. Which industries is PTI suited for? PTI is built for manufacturers across Food & Beverage, Automotive, Pharmaceuticals, Oil & Gas, Water & Utilities, and both discrete and process manufacturing environments.