Home / Point of View / Your ERP Data in Claude
AI & Platform

Your ERP Data in Claude:
The Architecture That Makes Manufacturing AI Actually Work

Everyone wants to ask AI about their business. Almost no one has the data foundation to make the answer trustworthy. The hard part of manufacturing AI isn't the model. It's getting clean, normalized, multi-ERP data into it.

Paul Ausserer, Marquis Data May 2026 10 min read
Works with every major AI platform
Claude Anthropic ChatGPT OpenAI Gemini Google Copilot Microsoft Perplexity Meta AI + any MCP-compatible AI tool

The question everyone is asking

At some point in the last year, someone in your organization asked a version of the same question: "Can I just ask ChatGPT what our gross margin was last quarter?" Or "Can I use Copilot to look at our ERP data and flag which customers are at risk?" Or simply: "Why do I still have to pull a report when I can have a conversation with Claude, Gemini, or any of these AI tools?"

These are legitimate questions. The technology to do this exists. Whether you are using Claude, ChatGPT, Gemini, Microsoft Copilot, Perplexity, or any other major AI platform, the underlying models are genuinely capable of reasoning over complex business data, identifying patterns, and generating insights that would take an analyst hours to produce manually. The bottleneck is not the model. The bottleneck is the data that reaches it.

Asking any AI tool about your ERP data when that data is fragmented, dirty, and spread across three systems is like asking someone to navigate a city using a map with half the streets missing. The AI will give you a confident answer. It just won't be right.

Marquis IQ is built specifically to solve the data side of this problem. Certified connectors pull from every ERP in your portfolio. Master data normalization resolves customers, suppliers, and items to single, conformed records across systems. The result is a clean, unified data layer - the foundation that makes AI actually trustworthy for manufacturing analytics, regardless of which AI tool your team uses.

Why ERP data and AI don't naturally mix

Enterprise AI integrations fail at the data layer far more often than at the model layer. The reasons are structural and familiar to anyone who has managed multi-ERP manufacturing environments:

  • Multiple disconnected systems - Epicor at one plant, Dynamics at another, Sage at a third. Each has its own schema, its own item master conventions, its own customer and supplier naming.
  • No conformed master data - The same customer appears as three records across three ERPs. The same product has different codes, different units of measure, different cost structures in each system.
  • Context the model can't infer - AI doesn't know that "BOEING-WA" and "Boeing Commercial Airplanes" are the same customer. It doesn't know that your Q4 seasonality is driven by defense contract cycles. It doesn't know which items are active versus discontinued unless that's cleanly encoded.
  • Raw ERP data is not analytics-ready - Transaction tables, backlog tables, and open order tables require aggregation, joining, and normalization before they can answer a business question. Pointing AI at raw ERP tables produces expensive confusion, not insight.
The Same Company. Three ERP Records.
Without a data layer

One customer. Three ERP systems. Three different records. When you ask an AI tool to analyze your top accounts, this is what it sees:

ERP System Customer Record Q3 Revenue
Epicor (Plant A) Boeing Commercial Airplanes $1,240,000
Dynamics 365 (Plant B) BOEING-WA $680,000
Sage 100 (Plant C) Boeing Co. $410,000
AI sees 3 customers. You have 1. $2,330,000 invisible
When you ask any AI tool "who are our top 10 customers?" it returns Boeing three times in three separate rows - or misses two entirely. Your total Boeing relationship, $2.33M in Q3, is invisible. The AI isn't wrong. The data is.

The Model Context Protocol: how AI connects to enterprise data

The Model Context Protocol (MCP) is an open standard, developed by Anthropic and adopted across the AI industry, that defines how AI models connect to external data sources and tools. Claude, ChatGPT, Gemini, Microsoft Copilot, and other leading AI platforms all support MCP as the standard protocol for enterprise data connectivity. Instead of each company building a custom, one-off integration between their AI tool and their databases, MCP provides a standardized way for any AI to discover, query, and reason over external data - securely, consistently, and without requiring the AI to understand the underlying data infrastructure.

Think of MCP as a universal adapter. The major AI tools speak MCP. An MCP server exposes data through a defined interface. Whatever is behind that interface - a database, an API, a data platform - any MCP-compatible AI can access it through the same protocol, with the same security model, using the same interaction pattern. Your team's choice of AI tool becomes a preference, not a technical constraint.

For manufacturing companies, this changes the architecture of enterprise AI from "build a custom connection to every system" to "expose one clean data layer through one MCP server." The complexity moves from hundreds of integration points down to one well-defined interface. The question becomes: what is the quality of the data layer behind that interface?

The Marquis IQ architecture: one clean layer, one connection

Marquis IQ is purpose-built to be the data foundation for enterprise AI in manufacturing. The architecture addresses every layer of the data problem before any model ever receives a query.

The Marquis IQ Architecture
Built for the AI era
AI Interface - Ask questions in plain English
Claude ChatGPT Gemini Copilot Perplexity + any MCP-compatible AI tool
Model Context Protocol (MCP)
Marquis IQ - The Data Foundation
One clean layer
Certified ERP Connectors
25+ maintained connectors. When your ERP updates, the connector updates.
Master Data Normalization
Customers, suppliers, and items resolved to one golden record across all ERPs.
Normalized Analytics Layer
Revenue, margin, inventory, and working capital pre-computed and queryable.
Custom Fields & Context
User-defined attribution on any object. Customer tiers, segments, deal types, and more.
Certified, read-only ERP connectors
Your ERP Systems - Nothing changes here
Epicor Microsoft Dynamics Sage Infor NetSuite SAP Oracle JDE + 18 more
No changes to your ERP systems. Marquis IQ reads from your existing systems through certified, read-only connectors. Every normalization, enrichment, and Custom Field lives in the Marquis IQ layer - never in the ERP. AI connects once to Marquis IQ through MCP and has access to your entire multi-ERP portfolio, clean and contextualized.

The Marquis IQ data layer handles four things that make AI trustworthy for manufacturing:

What Marquis IQ provides to AI
Certified ERP Connectors
25+ maintained connectors pull transaction data from every ERP on a defined schedule. When the ERP updates, the connector updates. No custom integration work required.
Conformed Master Data
Customer, supplier, and item records are resolved to single golden records across all systems. "Boeing" is one entity, regardless of how each ERP recorded it.
Normalized Analytics Layer
Raw transactions are aggregated, joined, and structured into analytics-ready datasets. Revenue, margin, inventory, and working capital are all pre-computed and queryable.
Taxonomy and Context
Product hierarchies, customer segmentation, and organizational structure are encoded in the data layer - giving AI the business context it needs to produce meaningful answers.

Custom Fields: giving AI the context to understand your business, not just your transactions

Clean, normalized ERP data tells AI what happened. It recorded that a customer placed an order, that an item was shipped, that an invoice was paid. What it does not tell AI is what any of that means in the context of your specific business - which customers are strategic platform relationships, which products are proprietary versus commodity, which end markets you are intentionally growing versus managing for cash. That context was never encoded in the ERP because the ERP was never designed to carry it.

Without that context, AI produces generic answers. "Show me revenue by customer" returns an alphabetical list with dollar amounts. Useful, but not intelligent. The question a PE operating partner actually wants answered is "show me revenue by strategic tier and end market segment" - a view that reflects their commercial framework, not just the ERP's transactional structure. For that answer, the AI needs to know which tier each customer belongs to. That is not in the ERP. It lives in the heads of the commercial team and, until now, in spreadsheets.

Custom Fields close the gap between what the ERP knows and what the business knows. They are the difference between AI that answers data questions and AI that understands your business.

Custom Fields in Marquis IQ's Enrich module let you add user-defined attribution to any data object - customers, suppliers, items, transactions, and more - without changing a single ERP record. The fields are defined by the operating team, populated through the Marquis IQ interface or bulk import, and immediately available as dimensions across every IQ module and AI query. Field types include text, date, number, currency, defined list, smart list, hierarchies, and more - fully managed within Marquis IQ.

The result is a data layer that carries not just the raw ERP facts but the business meaning layered on top of them:

  • Customers - strategic tier, account type (platform / add-on / distribution), region as defined by your operating model, active pipeline flag, commercial segment
  • Items - product category that maps to your commercial hierarchy (not the ERP item class), margin tier, proprietary versus commodity classification, end market designation
  • Suppliers - preferred vendor status, sourcing category, strategic relationship flag, geographic risk tier
  • Transactions - deal type, contract vehicle, channel classification, project code that spans multiple line items

Marquis IQ also supports third-party enrichment - D&B company data, SIC/NAICS industry codes, firmographic attributes - further extending the context the ERP never captured. When AI queries this enriched data layer, it is not reasoning over transaction tables. It is reasoning over a fully attributed, commercially meaningful view of your business. That is what produces answers worth acting on.

What a PE operating partner can now ask

With a clean, normalized data layer behind the MCP interface, the questions a PE operating partner can direct to an AI model become genuinely powerful. Not "give me a summary of the data" - but specific, cross-entity, cross-period questions that would otherwise require an analyst and a week of report building.

?
"What was gross margin by product line across all four facilities last quarter, and how does that compare to the same period last year?"
?
"Which portfolio company has the highest days sales outstanding and how has DSO trended over the past six months?"
?
"Show me revenue by end market across all entities in Q3. Which segments are growing and which are declining?"
?
"Which customers reduced their order frequency in the last 90 days? Flag any that represent more than 5% of a facility's revenue."
?
"Show me gross margin by strategic tier - Platinum, Gold, and Silver accounts - across all entities. Are our highest-tier accounts growing or declining as a share of total revenue?"

The last question above is only possible because of Custom Fields - strategic tier is not a field that exists in any ERP. It was defined by the operating team in Marquis IQ and is now available as a dimension across every query the AI can run. That is the difference between AI that reads your data and AI that understands your commercial framework.

These are questions that today require a data analyst, a set of Excel exports from multiple systems, and several hours of reconciliation. With Marquis IQ as the data foundation, they become conversational.

What finance and operations teams can now ask

The same architecture extends to the teams that manage the business day to day. Every question that currently requires opening a report, running a query, or waiting for IT to build something becomes a direct conversation.

?
"Which inventory items across all facilities haven't moved in 90 days? What is the total book value of that inventory?"
?
"How much open backlog do we have for Q4 by facility? Which facilities are above or below their seasonal average?"
?
"What is our price realization on Aerospace components over the past six months? Is it trending up or down?"
?
"Which supplier accounts for the most spend across the portfolio and what is our average payment performance with them?"

The common thread across every question above: the answer requires data from multiple sources, normalized to a common definition, with context that raw ERP tables don't carry on their own. That is exactly what Marquis IQ provides.

The data layer is the hard part. We already built it.

Every company exploring enterprise AI for manufacturing eventually arrives at the same realization: the model is not the constraint. OpenAI, Anthropic, and Google have made the model layer genuinely capable. The constraint is the quality of the data the model receives. Garbage in, confident garbage out.

Building the data foundation from scratch - certified ERP connectors, master data normalization, conformed taxonomies, analytics-ready aggregations - takes years and significant engineering investment. It requires deep knowledge of each ERP's schema, its quirks, its upgrade cycles, and its data quality characteristics. It is not a problem that can be solved with a data engineer and a few months of work.

Marquis IQ is that foundation - built specifically for PE-owned manufacturing environments, across the ERP systems they actually run, with the master data normalization that makes cross-entity analytics possible. The AI era in manufacturing does not require a new data infrastructure investment. It requires connecting the AI to a data layer that was designed for this from the start.

Related reading: Master Data as a PE Strategic Advantage covers the conformed data layer that makes cross-entity AI meaningful. Why Manufacturing Demand Planning Fails covers the data foundation required before AI-assisted forecasting works reliably.

Common questions

Questions about AI, MCP, and the Marquis IQ data foundation.

What is MCP and why does it matter for manufacturing analytics?
The Model Context Protocol (MCP) is an open standard developed by Anthropic that defines how AI models connect to external data sources. Instead of building a custom integration between an AI tool and each individual database or system, MCP provides a standardized interface. For manufacturing, this means AI can connect to a single, clean data layer - rather than needing separate integrations with Epicor, Dynamics, Sage, and every other ERP in the portfolio. Marquis IQ is designed to serve as that MCP data layer: one connection point that exposes normalized, conformed, analytics-ready manufacturing data to AI tools.
Does this require replacing our ERP systems?
No. Marquis IQ connects to your existing ERP systems through certified, read-only connectors. Your ERP systems continue to operate exactly as they do today - they are the source of record. Marquis IQ reads from them, normalizes the data, and provides a clean analytics layer on top. Nothing in your ERP environment changes.
Why can't we just connect Claude directly to our ERP database?
You can connect Claude to a raw ERP database. The result is a model that confidently answers questions using dirty, unmastered, unconfigured data. "Boeing Commercial Airplanes" and "BOEING-WA" remain two separate customers. Items tracked in different units of measure across facilities corrupt every calculation. The model has no concept of your product hierarchy, your customer segmentation, or your organizational structure. The answer comes back quickly and looks authoritative. It is also frequently wrong. The value of Marquis IQ is not the connection - it is everything that happens to the data before the model receives it.
What kinds of questions can AI answer with Marquis IQ as the data layer?
Any question that can be answered from transaction data across your ERP systems - normalized, conformed, and aggregated. Gross margin by product line across all entities. DSO trends by customer and facility. Inventory aging by SKU and location. Revenue by end market and quarter. Customer concentration and order pattern changes. Price realization by product category. Supplier spend and payment performance. The breadth of questions reflects the breadth of the Marquis IQ data model - which covers finance, inventory, pricing, sales, procurement, and operations across every connected ERP.

The data foundation for manufacturing AI. Already built.

Marquis IQ connects every ERP in your portfolio, normalizes master data, and provides the clean, conformed data layer that makes AI answers trustworthy - without replacing a single system.