Manufacturers have visibility but not confidence
Manufacturing has spent the past decade pursuing visibility. Connected assets, industrial IoT, cloud platforms, advanced analytics and digital twins have generated unprecedented amounts of operational information across production, maintenance, logistics and supply chain functions. Yet as AI moves from experimentation into production environments, many organizations are discovering that the challenge is no longer collecting data. It is trusting it.
Weaknesses that once remained hidden inside disconnected systems, inconsistent definitions and fragmented ownership are becoming increasingly difficult to ignore. AI is exposing problems that industrial businesses have often worked around for years through experience, local knowledge and manual intervention. Across the sector, a growing realization is emerging. Success with AI will depend less on deploying sophisticated models and more on creating confidence in the information that feeds them.
Data without direction
Few industries can claim to be short of data. Production lines, warehouses, suppliers, maintenance systems and enterprise applications generate enormous volumes of information every day. The problem is that collecting data and using it effectively are proving to be very different challenges.
For many organizations, digital transformation began with a relatively straightforward objective: capture more information, improve visibility and make better decisions. Years later, the information exists, but many businesses still struggle to translate it into coordinated action across operations.
“There was this huge hype around IoT perhaps eight or ten years ago, and now it is coming back again because companies are trying to build resilience into supply chains and operations,” Maggie Slowik, Global Industry Director for Manufacturing at IFS says. “Manufacturers are revisiting ideas around digital twins and what-if scenario modelling, but many organisations are still sitting on large amounts of operational data that they do not really know what to do with. I still see a lot of manufacturers collecting data without a clear operational strategy behind it. Now AI is forcing companies to confront that problem because they realise they need better quality operational data if they actually want AI to deliver meaningful results.”
The issue is rarely a lack of technology. Most enterprises have invested heavily in systems to support planning, logistics, production, maintenance and supply chain operations. The difficulty comes when those systems need to work together. Many digital estates have been assembled over years, often to solve specific business challenges at particular sites, functions or regions.
“Manufacturers have invested in digital systems for years, but most of those investments were made to solve specific operational challenges inside individual functions, plants or regions,” Greg Hanson, Group Vice President and Head of EMEA North Sales at Informatica says. “Over time, organisations ended up with different versions of supplier, product and operational data across the business.
“The problem is now much more visible because manufacturers are being asked far more detailed questions about sourcing, suppliers, materials and operational exposure. Regulations, supply chain disruption and AI initiatives all depend on connected operational data, yet many organisations are still trying to piece that information together across disconnected systems and teams.”
The result is a familiar pattern. Production, procurement, logistics and planning teams often possess vast amounts of information yet struggle to build a shared understanding of what is happening across the business. Visibility exists, but operational intelligence remains elusive. “Manufacturers do not have a shortage of data,” Hanson continues. “The challenge is connecting it in a way that supports operational decisions. Operational intelligence comes from understanding how data is connected across suppliers, materials, products and operations. If that information is fragmented or inconsistent, manufacturers risk scaling poor decisions rather than improving them.”
That observation sits at the heart of many industrial AI discussions. The challenge is no longer gathering more information. It is creating a consistent operational picture that decision-makers across the business can rely upon.
Why trust breaks down
The vision of a connected enterprise remains attractive, but the reality inside many industrial organizations is considerably more complex. Operational environments have evolved over decades through acquisitions, regional expansion and independent technology decisions. Bringing those environments together often exposes fragmentation that has existed for years.
“The problem is not necessarily that manufacturers lack systems, it is that those systems are fragmented across the organisation,” Slowik says. “AI has the potential to connect the operational thread all the way from design through to manufacturing and supply chain execution, but companies have to decide how much they are willing to consolidate and integrate their operational environments. That is where many organisations still struggle.”
Different functions frequently maintain their own datasets, standards and definitions, creating multiple versions of reality across the same enterprise. Procurement teams, production managers and supply chain specialists may all be working with information about the same products, suppliers or assets, yet interpreting and managing that information differently. This explains why master data management has moved from being a specialist IT concern to a strategic business priority.
“Organisations need a consistent view of suppliers, materials, products and manufacturing environments,” Hanson explains. “Otherwise, teams end up working from different versions of the same operational data. Strong master data improves sourcing, production and supplier decision-making. It gives manufacturers better visibility into where materials are coming from, how they are sourced across regions and the operational or regulatory exposure attached to them.”
Not everyone believes large-scale centralization is the answer. Jay Limburn, Chief Product Officer at Ataccama, argues that many organizations focus on moving data rather than agreeing what that data means. “It is super hard, and most teams aim at the wrong target,” he says. “They try to drag every system into one place. In a business grown by acquisition and decades of plant-level decisions, that project never ends. Aim instead for one agreed view: shared definitions, shared reference values, clear lineage and known owners across systems that stay where they sit. Align the meaning and reliability follows. Chase centralisation and you inherit the same trust problems on the far side of an expensive migration.”
That perspective highlights an important distinction. Integration and standardization are not necessarily the same thing. One focuses on technology architecture. The other focuses on creating shared meaning across the organization.
The challenge becomes even more apparent when operational technology data is combined with enterprise systems. Information may reside in historians, MES platforms, ERP systems and edge devices, but its value depends entirely on context. “A historian might record a temperature reading at a certain time of day, but without the context that identifies which line or product run it is and what the acceptable temperature range is, the recorded number is essentially meaningless,” Sunitha Rao, Senior Vice President and General Manager of Hybrid Cloud at Hitachi Vantara says. “Organizations need to connect OT data to ERP records and MES logs to transform raw numbers into real decisions.”
Context, ownership and accountability repeatedly emerge as the foundations of trust. Technology can identify anomalies, automate workflows and improve visibility, but it cannot determine who owns a process, who is accountable for data quality or how operational standards should be applied across a business.
“Mostly organisational, and that is the uncomfortable part,” Limburn adds. “Software profiles data, flags anomalies and suggests fixes all day. It cannot decide who owns the product hierarchy, whose plant standard wins, or how fast a problem gets fixed. People decide those things. The failures I see rarely come down to a missing tool. They come down to vague accountability and governance that lives in a slide deck instead of the actual workflow.”
These weaknesses often remain hidden until organizations attempt to scale successful initiatives across multiple sites. What works well in one facility may struggle elsewhere because the supporting governance, leadership engagement and data maturity are not consistent.
“One of the biggest issues is that different manufacturing sites are often operating at very different levels of data maturity,” Slowik continues. “You might have one highly advanced site with relatively clean operational data and strong leadership support but replicating that success across a wider manufacturing footprint becomes much harder. There also needs to be a clear rollout strategy. Without that broader strategy and governance structure, scaling AI and data initiatives becomes very difficult.”
AI changes the stakes
Data quality challenges are not new. What has changed is the cost of ignoring them. Historically, reporting systems allowed organizations to compensate for inconsistencies through human oversight, operational experience and local expertise. AI operates much closer to the decision itself.
“Reporting forgave a lot,” Limburn adds. “Aggregation, weekly snapshots and a human sanity-checking the dashboard buried plenty of sins. AI removes all three. It runs closer to the decision, on broader inputs, faster than anyone can eyeball, so a small inconsistency that used to average out now drives a recommendation someone acts on. AI does not dirty your data. It just stops letting you get away with it.”
The same pattern is becoming increasingly visible as companies attempt to operationalize AI across production, maintenance and supply chain environments. “AI is exposing data quality problems that manufacturers were often able to work around in traditional reporting and analytics environments,” Hanson continues. “AI systems operate at speed and scale, so small inconsistencies in operational data can quickly become significant operational issues. Our research shows 57% of leaders see data reliability as a major barrier to moving AI projects into production.”
For Rao, these challenges reflect the way industrial environments evolved long before AI became a strategic priority. Most facilities were designed around uptime, safety and throughput rather than data portability, governance or AI readiness.
“Manufacturing sits at the intersection of legacy operational technology and modern digital infrastructure,” she says. “Most digital transformation investments have focused on layering new systems on top of existing foundations without addressing the underlying problem: whether the data itself is consistent, contextualised and trusted.”
As AI becomes more deeply embedded in operational workflows, confidence in outputs becomes just as important as the quality of the models themselves. Frontline teams need to understand where recommendations come from, how decisions are reached and when intervention is appropriate.
“The people doing scheduling, planning and operational management still want to understand where the data is coming from and how decisions are being made,” Slowik concludes. “A lot of operational experts still rely heavily on spreadsheets because they trust what they can see and control themselves. They want to be able to intervene and challenge AI outputs if something does not look right.”
Limburn sees explainability as essential to building that confidence. “You do not rebuild trust by telling an operator to believe the model,” he explains. “You rebuild it by showing the work: where the number came from, what changed since yesterday, who owns the fix.”
For all the attention being paid to AI, the issues raised here are surprisingly familiar. Ownership, governance, context and accountability have been challenges inside manufacturing for years. What has changed is that AI is making them much harder to ignore. Companies that address those foundations will be in a far stronger position than those that continue searching for technological shortcuts.
