Create a free Supply & Demand Chain Executive account to continue reading

Fix Data Foundation Before Next ERP or AI Upgrade

Too often, the supply chain and operations leadership team blames the software. But for most multi-plant manufacturers, what failed was the data, not the choice of technology.

Chatchanan Adobe Stock 923084100
Chatchanan AdobeStock_923084100

A supply chain AI pilot promises actionable insights and never-miss shortage and inventory alerts, then quietly stops getting used a few months in. The demo looked sharp, yet the recommendations it generated weren't trusted by users. There were shortages that didn't exist and missing ones that did. Sound familiar? This is a common scenario currently playing out across many global manufacturing enterprises.

Too often, the supply chain and operations leadership team blames the software: swap vendors or recommend a change (or upgrade) to the underlying ERP system. But for most multi-plant manufacturers, what failed was the data, not the choice of technology. Common data issues include inconsistent part numbers, stagnant and inaccurate lead times, and master data and MRP settings that haven’t been updated in years. Layer on plant-to-plant ERP variability and a lack of consistent enterprise data governance, and you have an ever-increasing uphill battle. Until that foundation gets fixed, no technology investment, however sophisticated, will deliver a trustworthy answer.

Manufacturers running multiple plants are building AI and automation ambitions on top of ERP data that lacks the necessary governance and management required to deliver results. Part nomenclature differs from one facility to the next. Lead times reflect what was typed in years ago by someone no longer in the organization, not what suppliers actually deliver today. Master data on the same parts and suppliers often contradicts itself depending on which ERP system you ask. Order policies and MRP settings are set and forgotten.

Much of the current AI hype in the supply chain ecosystem skips past this harsh reality entirely. The conversation jumps straight to agents, copilots, and autonomous decisioning, with an implicit assumption that the data underneath is already reliable enough to support it. Few organizations stop to seriously ask whether that assumption holds. It usually doesn't, and the gap is becoming more expensive to ignore.

What follows is a look at why this challenge is surfacing now, where the breakdown actually lives, what it costs manufacturers, the concrete groundwork worth laying before the next AI purchase, and practical advice for getting there in a scalable way.

Why this is suddenly urgent

Tariff and geopolitical pressures are pushing manufacturers toward supplier diversification, not consolidation, and that's compounding the complexity problem rather than easing it. Per ISM's December 2025 Supply Chain Planning Forecast, only 36% of manufacturers are actively shifting production domestically to avoid tariffs; most are instead diversifying suppliers, which means more supplier relationships, more part variants, and more sites to keep synchronized. Layer on accelerating customer demand and rising product complexity, and BOM proliferation also follows close behind: more components to manage, more supplier-part combinations to track, and faster planning cycles required to synchronize materials across a global factory footprint.

All of that complexity has to live somewhere, and for most manufacturers, that's still the ERP. Leaders are self-reporting the strain. According to the National Association of Manufacturers, 44% of leaders say the data they collect has doubled in the past two years and expect it to triple by decade's end; 86% believe effective use of data will be essential, yet only one in four have high confidence the right data is even being collected. More than 60% have a data management strategy, yet only 15% actually follow it.

Shortages, excess inventory, and underwhelming AI results share a root cause: materials and inventory data that was never built to be regularly maintained and consistent across sites in the first place. That complexity keeps growing. What's changed is that AI now makes the cost of that inconsistency very visible, while bringing the potential benefits and positive business outcomes so tantalizingly close.

Where the breakdown actually lives

The breakdown shows up in specific, recognizable ways on the factory floor and inside the ERP. The same part has a different part number at plant A vs. plant B, or worse, the same part number represents something different depending on the site. Purchasing parameters and order policies rarely match across ERP instances, especially after an acquisition merges new ERPs into your ecosystem with disconnected teams and processes. Lead times reflect whatever was typed into the item master years ago, not supplier performance or contractual obligations today, a gap that only widens as the supply base keeps expanding. And master data ownership is split across procurement, engineering, and operations, with no single team accountable for keeping any of it current.

The default fallback for reconciling all of this is still the spreadsheet. A 2026 ROI report from Specright and UserEvidence found 60% of manufacturers surveyed were managing specification-level data in spreadsheets, with teams losing an average of 468 hours a year to manual work, time that should be used strategically, ideally conducting recurring “plan for every part” exercises meant to keep order policies, safety stock, and cycle stock targets grounded in reality.

The failure mode may not be obvious. An AI agent or forecasting model built on top of poorly managed part master data doesn't throw an error; it generates a confident recommendation to reorder a part that isn't actually short, or reduce stock on one that is because the lead time or part master data that it trusted was wrong. And the problem compounds: every additional plant, ERP instance, acquisition, or new supplier adds another layer of inconsistency.

What it costs to ignore

Yes, the AI pilot might fail, but more importantly, you’re continuing the cycle of excess inventory tied up while shortages and expedite costs remain stubbornly high. You’re also potentially wasting capital on AI/software licenses that get shelved after a failed pilot. And your materials management, procurement, planning, and operations teams continue to spend hours reconciling spreadsheets instead of strategically managing their desks. More suppliers and sites without better data discipline means the problem gets worse before it gets better.

The practical path forward

Fixing this doesn't require ripping out the ERP or ditching the technology pilot, and it doesn't require waiting for a perfect data set that will never exist. Here are some recommended steps:

  1. Start by choosing a system built to fix the foundation, not paper over it, something that normalizes part numbers, units of measure, and lead times across sites and ERPs into one consistent layer, rather than another point solution that assumes the data underneath is already clean.
  2. Start small. Pick one plant, one product line, or one commodity/supplier segment where you can see the data clearly and prove the approach there before scaling anywhere else.
  3. Use that pilot to build trust, not just prove a concept. Prioritize and validate data elements like lead times and master data against what's actually happening on the floor today, not what was typed into the item master years ago.
  4. Assign real ownership of that data so it stays accurate after the pilot ends, not just during it. Trust compounds the same way inconsistency does: once one team sees a recommendation that makes sense and matches reality, they act on the next one faster.
  5. Show a small, specific win: a shortage avoided, a batch of excess inventory identified and cleared, hours of manual spreadsheet reconciliation eliminated. Executives don't need a moonshot to fund the next phase. They need proof that fixing the foundation pays for itself.
  6. Only then should you automate and expand, plant by plant, supplier segment by supplier segment. Treat this as an evergreen loop, not a single project. Every cycle should surface the next highest-value opportunity, and for each one, ask a second question before acting on the first: is the fix operational, or is the real fix the data that surfaced the opportunity to begin with?

The manufacturers pulling ahead right now aren't the ones with the most advanced AI. They're the ones running that loop continuously, plant after plant, until a clean data foundation stops being a project and becomes how they operate.

The constraint on AI value in manufacturing was never really the legacy ERP system, and it isn't really the AI model either. It's whether the data underneath can truly be trusted. As tariffs, market pressures, increased customer demand, and reshoring push manufacturers toward more products, more suppliers, and more complexity, the winners won't be the ones with the most advanced AI; they'll be the ones who fixed their data foundation first and built a culture of disciplined continuous improvement on top of it.

More in ERP
Page 1 of 52
Next Page