Frameworks

Your ERP data sits in silos nobody trusts. Make it AI ready one source at a time.

Chris McCarthyNovember 3, 20265 min read
erp data ai readinessai ready datalegacy erp ai integrationdata governance manufacturingcio ai strategy

Your ERP data sits in silos nobody trusts. Make it AI ready one source at a time.

By Chris McCarthy, Founder & Chief AI Officer, OB.1

Your ERP data is AI ready when each source that feeds a use case has a named owner, a known level of quality, and a defined path into the AI system. Most mid-sized manufacturers do not have that yet. The fix starts with one question: is your data AI ready, and who owns each source?

This is part two of our Pilot to Production series. This one is for IT directors and CIOs.

The pressure lands on IT first

Leadership has seen the demos. Now they want AI on top of the system you already run. That system might be an AS/400 that has run the plant for twenty years, an older SAP instance, NetSuite, or Dynamics with a decade of customizations. Half the real data sits beside it in spreadsheets, shared drives and someone's inbox.

You are not alone in this. In RSM's 2026 middle market AI survey, 32% of manufacturing respondents said data quality, availability and lineage issues limit broader AI deployment. Another 27% pointed to integration with legacy systems.

I spent ten years running ERP and global IT for manufacturers before I started OB.1. The pattern does not change. Transactions live in the ERP. The judgment about which numbers to trust lives in people's heads.

Why bad data turns into confident wrong answers

A person who knows the item master is messy works around it. They know the old part numbers, the duplicate vendors and the unit-of-measure mistakes. A model does not. Feed it inconsistent data and it returns fluent, confident answers about inventory, pricing and lead times that are wrong.

That is the risk you carry. When the answer is wrong, the business will ask IT why.

Data practices are the gap. Gartner reported in February 2025 that 63% of organizations either lack or are unsure whether they have the right data management practices for AI.

The question to ask: is your data AI ready, and who owns each source?

Start with one use case. Say it is answering customer order-status questions. List every source that answer depends on: sales orders, the production schedule, shipping records, and the spreadsheet planning keeps on the side.

Then score each source on four things:

CheckWhat to ask
OwnerWho decides what is correct in this data?
QualityHow often is it wrong, and how do we know?
FreshnessHow old is it when someone reads it?
AccessCan a system read it safely, with the right permissions?

A source with no owner is the first red flag. Data with no owner never gets fixed, because fixing it is nobody's job.

What a good answer looks like

You do not need a full ERP replacement to start. You need a clear map for one use case:

  • Every source listed, with an owner from the business and a contact in IT
  • A short note on known quality problems and who fixes them
  • A rule for what the AI may read, and what it may never change
  • A staging layer, where needed, so the AI reads a clean copy and never touches production tables

That last point matters most for IT. Keep AI read-only against production until the data, the controls and the owners are proven.

How OB.1 fixes this

The Operational Blueprint looks at platforms, process and people together. For a data problem, we trace one use case from the question someone asks to every system that holds part of the answer.

You get a source-by-source map with owners and quality notes. You get the access rules AI needs to stay inside, written in plain language your leadership team can sign. And you get a ranked plan for which data to fix first, tied to the use cases worth funding.

From there, build partners work to that spec in the Architect phase. Our Govern service then keeps the access rules and monitoring in place once the system runs. OB.1 designs and governs. We do not sell build hours, so our advice on platforms stays neutral.

Read the rest of the series

  1. Costs climb faster than your prices
  2. Your ERP data sits in silos nobody trusts (this article)
  3. The plant you just bought runs a different system
  4. Staff paste pricing into public AI tools
  5. Your best tech retires, and the fixes leave too
  6. Automation could ship the wrong order
  7. Teams lean on the system and lose the judgment to step in

Book a Blueprint Session

Sources

Frequently Asked Questions

What does AI-ready data mean for a manufacturer?

Data that has a named owner, a known level of quality, and a safe, defined path into the AI system for a specific use case. Judge readiness one use case at a time.

Do we need to replace our ERP before using AI?

Usually not. Many use cases can read a clean staged copy of ERP data while the ERP keeps running. Replacement is a separate decision with its own business case.

Who should own ERP data quality?

The business function that creates the data, such as purchasing for vendor records or engineering for the item master. IT owns the systems and access. Both names belong on the map.

How do we keep AI from changing production data?

Start read-only. Let the AI read a staged copy, require human approval for any change, and log every action. Widen access only after the controls prove out.

Get Started

Book a Blueprint Session.

That is the first conversation.

chris@ob1ai.co | (234) 602-0500

Talk to OB.1

Tell us where AI fits in your business. We start with the Operational Blueprint and Diagnostic AI Consulting, and we reply within one business day.