Artificial Intelligence is changing the way companies think about their internal systems, and in manufacturing, that shift is even more noticeable.
ERPs — those systems that centralize and integrate operations like inventory, procurement, production, and finance into one place — have been around since the late 90s and remain the backbone of most industries worldwide. But their role is going through a major transformation as AI gains ground.
And something interesting is happening right now.
While AI moves forward at breakneck speed, a recent survey from manufacturing data platform Caddi found that 69% of organizations surveyed plan to invest in physical assets, but only 33% have plans to invest in operational software like ERPs and manufacturing execution systems, commonly known as MES.
That represents a 60% drop compared to the previous year.
In practice, manufacturers seem to be prioritizing investments that deliver immediate, measurable results on the factory floor. The problem is that choice could end up being expensive down the road when it comes to AI integration.
Experts point out that AI doesn’t arrive to replace the ERP — it demands that the ERP be much better than it is today.
Fragmented data, legacy systems, and disconnected processes create a scenario where AI, instead of helping, can end up generating wrong answers faster and with more confidence than ever before.
And that’s when the damage gets worse.
In this article, we’ll explore how this relationship between AI and ERPs is playing out in practice, what the real risks of this integration look like, what major manufacturers are already doing, and why data modernization might be the most important step before any automation.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
What happens when AI meets an outdated ERP
When we talk about Artificial Intelligence in industrial environments, the first image that comes to mind is usually robots, automated conveyor belts, and smart sensors scattered across the factory floor. But the most critical part of this transformation isn’t in the hardware — it’s in the data that drives the decisions. And that’s exactly where ERP systems enter the picture, or at least where they should.
The problem is that many of these systems still run on architectures designed for a world that no longer exists — one where information was entered manually, reports took hours to generate, and integration between departments was handled through email or spreadsheets.
As Srinivas Chippagiri, a senior member of the technical staff at Tableau, a Salesforce company, put it, AI raises the bar on the data foundation rather than replacing the ERP entirely. According to him, manufacturers with clean, well-integrated ERP data can layer AI on top and get real value, while those with fragmented data in aging systems just get wrong answers with more confidence and faster.
When you connect an Artificial Intelligence tool to an environment like that, the result can be worse than having no AI at all. Language models and predictive algorithms depend on clean, consistent, and well-structured data to work properly. If the ERP feeds the AI duplicated, outdated, or poorly categorized information, the model will process it all with the same confidence and speed as always — except now it’s delivering wrong answers at a much larger scale. Experts in industrial systems architecture call this phenomenon error amplification, and it’s one of the biggest risks of rushing the integration between AI and legacy systems.
Another point that rarely gets discussed is the impact on operational teams. When an AI system makes an incorrect recommendation with a high degree of apparent certainty, the operator or manager tends to trust it — after all, the machine said so. This shift in accountability can lead to poor decisions in purchasing, inventory management, production planning, and even preventive maintenance.
Modernize instead of ripping and replacing
Cache Merrill, founder and CTO of Zibtek, a company that works with industries on ERP modernization, custom software, and AI initiatives, has a very practical take on this. He says the company’s clients aren’t choosing between ERP and AI. Instead, they’re starting to use AI around the ERP systems they already depend on.
According to Merrill, there’s a lot of investment tied up in existing systems and legacy data, so it can be hard to justify replacing everything just to access newer AI capabilities. In many cases, it makes more sense to modernize the parts of the ERP that are causing problems, improve the way data flows between systems, and only then introduce AI or automation where there’s a clear business case.
For him, the best use case is when AI agents can take over work that normally requires someone to bounce between multiple systems and reports. A simple example is inventory analysis. To handle that, a person would need to look at current stock, recent orders, purchasing information, supplier data, and forecasted demand before deciding what deserves attention.
Merrill explains that an AI agent could pull all of that together, identify patterns, and prepare the next step — instead of someone doing all that work manually. In his view, the manufacturers that benefit the most are the ones working both sides of the problem: modernizing the ERP and data where needed, while using AI and automation to eliminate unnecessary manual work around it.
Why data modernization comes before automation
The logic seems straightforward, but in practice, companies tend to flip the order. They invest first in Artificial Intelligence tools, then discover that the available data isn’t good enough to support the models, and end up scrambling through a correction process that costs more time and money than it would have if the foundation had been prepared in advance.
In the manufacturing sector, this mistake is even more expensive because the systems are connected to physical operations, where a wrong decision can mean wasted materials, line shutdowns, or delivery delays.
Modernizing an ERP doesn’t necessarily mean throwing everything out and starting from scratch. In many cases, the smarter path is a layered strategy where existing data is migrated, normalized, and progressively enriched while new integrations are built on top of a more flexible architecture. Modern ERP platforms already offer open APIs, microservices support, and compatibility with real-time data pipelines, which makes it much easier to connect with AI tools without a complete replacement of the legacy system.
What experts consistently recommend is that companies start by mapping their data flows before making any technology decisions. Knowing exactly where data is generated, how it moves between systems, where the bottlenecks are, and where information gets lost or duplicated is the first step toward a modernization effort that actually works. Without that diagnosis, a company might end up automating bad processes — and then AI will just do the wrong things much more efficiently. 😅
The weight of legacy systems in industry
Remi Beges, CTO and co-founder of Bonx, an AI-native manufacturing ERP company, is straightforward about where the bottleneck lies. In his view, legacy ERP is the single biggest thing holding manufacturers back from achieving meaningful efficiency gains with AI. Most manufacturers today still use a legacy ERP that wasn’t built, at its core, with AI in mind.
Beges explains that these systems were designed with one purpose: to reliably record transactions. In other words, they were built as systems of record, not as systems capable of taking action.
But in theory, today’s AI could not only identify which systems are affected by a delivery delay, but also draft and send emails to customers letting them know their order will be late, while notifying workers about updated processes and schedules. The catch is that this is virtually impossible in a system designed only to record transactions.
AI-powered ERP programs and the risks that come with them
Some companies like Microsoft are launching specialized programs such as Dynamics 365, an agentic ERP system geared toward manufacturing that connects demand, supply, production, delivery, and cost data so AI agents can execute decisions — rather than just answering questions and generating recommendations.
In light of this trend, Beges raises an important warning against handing tasks entirely over to AI, since it could take unintended actions without a human in the loop. In his view, a system acting autonomously isn’t inherently dangerous, and there are situations where it’s perfectly appropriate.
For example, if the ERP detects that a material will run short for a future production run based on planned usage, the company might want the system to figure out which supplier would be best to fill the order and draft a purchase order for the buyer to review. But it probably doesn’t want that order sent out without any human verification.
The key, according to Beges, is to define those boundaries in advance — establishing how the system should and shouldn’t act — while making sure it’s smart enough to proactively flag exceptions and unusual situations.
What major manufacturers are already putting into practice
Despite the industry’s cautious stance on the risks and challenges of implementation, some manufacturers have already taken significant steps in this space. Clorox announced in 2025 a transition to a new ERP system, moving away from manual spreadsheets to a far more automated process. The move is part of a $500 million digitization plan that kicked off in 2021.
Giants like Nestlé have also announced the adoption of AI-focused ERP systems. The company migrated its entire ERP process from data centers to the cloud in 2022, a critical step in preparing its data foundation for new intelligent capabilities.
Schreiber Foods recently announced a partnership with Ascendion to modernize its ERP systems by deploying AI agents. The changes will be rolled out across more than 40 facilities spread over five continents.
In the automotive sector, companies are using predictive models fed directly by the ERP to anticipate component failures on the production line before they happen. This reduces unplanned downtime and improves overall plant efficiency. The differentiator in these successful implementations is always the same: well-organized data with a consistent history and real-time updates.
In the consumer goods sector, major manufacturers are integrating AI into the ERP’s procurement module to optimize orders based on variables like demand forecasting, supplier lead times, and market price fluctuations. The practical result is a significant reduction in idle inventory and improved operating margins — two metrics every CFO tracks closely. But what makes this possible isn’t just the AI algorithm — it’s the quality of the historical data that the ERP has stored over the years.
The investment gap that could prove costly
The Caddi survey revealing a 60% drop in plans to invest in operational software raises a serious red flag. There’s a clear tendency to prioritize what’s visible — new machines, modern equipment, physical automation — over what supports everything underneath, which is the data and management systems. This imbalance creates a paradoxical situation where a company buys cutting-edge hardware but continues operating with a digital infrastructure from the 2000s to make decisions about it.
This gap also has a human dimension. Industrial IT teams often carry years of knowledge about legacy systems, and when modernization arrives without proper planning, that knowledge can be lost in the process. Well-executed modernization involves not just the technology but also the people who will operate and maintain these systems, ensuring that institutional knowledge is preserved and adapted to the new environment.
Experts emphasize that these changes need to be made carefully. Chippagiri highlights that the biggest trap is trusting a probabilistic output inside a system that people treat as an absolute authority. According to him, wrong answers built on outdated or inconsistent data are worse than no answers at all, because people act on them.
This also creates accountability gaps. When an AI-driven action turns out to be wrong, who takes responsibility? That’s a question every manufacturer needs to answer before giving their systems too much autonomy.
The manufacturing sector has a clear window of opportunity right now. Artificial Intelligence tools are more accessible than ever, ERP vendors are competing to offer native AI integrations, and companies that can align a solid data foundation with a well-executed modernization strategy will have a real competitive advantage in the years ahead. Those that wait may discover that the cost of catching up later — with even more fragmented systems and a larger technical debt — will be much higher than investing now. 🚀
