September 29, 2026 How to Overcome Manufacturing’s AI Data Hurdles

To get AI out of pilot and into production, manufacturers must prioritize a data foundation built on standards, context, and governance.

By John Harrington, Co-Founder and Chief Product Officer, HighByte

Artificial intelligence (AI) continues to be a top-line priority across industries, and manufacturing is no exception. According to NTT Data, 38.6% of manufacturing and automotive AI leaders are rebuilding applications with embedded AI capabilities, while 67.6% describe their current AI investment as very significant. These trends signal not just an increasing focus on the integration of AI solutions, but a larger shift in the adoption of strategic AI-driven priorities across the industry.

Even so, strategy does not always translate well to execution. While research from Deloitte found that many teams have moved their first AI use case into live operations, only around one in five of these use cases have been scaled consistently across sites. Manufacturers often struggle to move from AI pilots to real-world implementations, hitting common stumbling blocks that impede effective deployment within facilities with legacy infrastructure.

Contrary to popular belief, this is not strictly a technology problem. Legacy systems—so often maligned as difficult to access or work with—actually create vast amounts of relevant data that can, and should, guide more informed operations.

The true challenge lies in the fact that the operational data generated by these systems is disconnected, inconsistent, and not properly operationalized for AI deployments. It is not uncommon for today’s operators to interact with six to ten different systems in order to get their job done. Combined, these challenges hamstring AI initiatives before they can even get off the ground, reducing their effectiveness and ROI on the factory floor. Rather than focusing on speed and breadth of deployment, the manufacturers that succeed with AI initiatives at scale will be those that first establish a strong, trusted, standardized, and contextualized Industrial DataOps foundation.

Why Manufacturing AI Initiatives Stall

Manufacturers don’t struggle with scaling AI due to a lack of innovation, but because their underlying data ecosystems aren’t ready to support AI solution patterns at scale. When data is spread across a range of ERP systems, MES platforms, SCADA systems, spreadsheets, and legacy equipment, it becomes inaccessible and unusable for the data-hungry AI models that support high-value use cases.

The challenges that inhibit manufacturing AI adoption can be broken down into three categories:

• Data Fragmentation: Data often ends up residing wherever it’s generated in manufacturing technology ecosystems, which are made up of various systems, machines, and unique sensors that were purchased over many years and deployed across multiple networks. With unique integration patterns and protocols governing which systems share what data, there’s no common repository or access pattern for the data generated by these disparate elements of the network. Without discrete, correlated data sets, large language models (LLMs) and other AI solutions don’t know how to access the data, or even what data to use.
• Data Inconsistency and Lack of Utility: Manufacturing systems and machines lack standards on what data is available, how it is defined, and where teams can access it. Even if raw data is accessible, it does not have defined standards on how it is captured, stored, and contextualized. Even when this is determined for one asset, asset class, line, or site, the next will be different and require more work and connections. An LLM will attempt to draw conclusions using whatever data is available—without clear definitions, it will make incorrect assumptions about what data to use, what the data signifies, and what kinds of inter-system relationships should be derived from it. This leads to AI hallucinations, creating a lack of trust in both models and data, and the potential for users to base critical decisions on incorrect information.
• Lack of Governance and Guardrails: Without clear rules governing how data is collected, shared, and used across disparate systems, scaling AI initiatives becomes more difficult. AI agents need specific guardrails and directions to ensure they stay within their assigned roles in manufacturing environments. When systems lack adequate data access controls, security, compliance, and risk management measures, there’s no guarantee that they’ll make correct decisions, leading to more waste, asset breakage, or even physical danger on the factory floor.

These core issues won’t be solved by introducing another AI model or IIoT platform into manufacturing infrastructure. Instead, they require teams to rethink the data architecture that powers AI-driven operations.

Building a Foundation with Standards, Context, and Governance

The effectiveness of any AI system depends directly on the quality and usability of the data it’s ingesting. To ensure quality and reliability, manufacturers must establish a common framework that allows data to move seamlessly across systems, sites, and teams.

This is a multi-step process that includes:

• Standardizing data across the network. All information generated by machines, sensors, and applications should adhere to consistent formats and definitions. Treating data as a product and using common data models helps eliminate discrepancies between assets, lines, systems, and sites that may have historically used, translated, or measured their actions differently. This standardization helps manufacturers compare performance more easily, replicate best practices, and deploy AI more efficiently across locations.
• Adding essential context to data. AI needs to understand the context surrounding an operational or business decision so it can act accordingly. As is, raw telemetry data only offers AI limited value into how a machine is functioning, it does not recognize the relationships within the broader factory environment, or variability across products or raw materials. Contextualizing data products clearly defines the parameters AI needs to understand equipment hierarchy, production processes, asset history, work orders, quality metrics, operational conditions, and more, helping manufacturers derive deeper value from their AI deployments.
• Streamlining data workflows. Aggregating data integrations into a DataOps platform streamlines the deployment of new data pipelines, simplifies integration management, and improves the efficiency and decision accuracy of machines, operators, systems and AI Agents. By maintaining a single source of truth, manufacturers can ensure accuracy across human and AI-backed decision-making.
• Ensuring governed data management. Teams must define when data products are accessible and to which users, as well as establish who is responsible for which data resources. Establishing this comprehensive governance while maintaining data accessibility is critical to scaling operations across the enterprise and to generating reliable, trustworthy systems and results. This can be done by leveraging an integration platform that helps manufacturers define their data products, verify data, and more closely manage data flows.

Moving From Pilots to AI-Powered Operations

The difference between experimentation and AI-driven transformation often comes down to data readiness. A strong, connected Industrial DataOps foundation helps unlock the full value of industrial AI, empowering manufacturers to move beyond isolated use cases into scalable operational applications.

By investing in more robust data infrastructure, manufacturers can unlock high-value AI use cases like predictive and prescriptive maintenance, real-time production monitoring and performance improvements, quality control refinement, supply chain and inventory optimization, and decision intelligence. This model also helps break down long-standing organizational silos, freeing data to flow more seamlessly across IT, OT, engineering, operations, business teams, and AI applications.

A solid DataOps infrastructure enables repeatable success, whether AI-driven or otherwise. With a contextualized and connected ecosystem, manufacturers will be more agile and more prepared to adapt as the business priorities change and when new and impactful technology emerges.

Good Data, Good Results

Manufacturers that continue to struggle with fragmented systems, inconsistent data, unclear ownership, and ineffective governance will find it hard to move beyond limited AI pilot programs.

By prioritizing data products with standardization, contextualization, and streamlined data pipelines—all within a governed and managed system—teams can create the conditions for AI to scale and deliver business value. As factories become more connected and intelligent, the competitive advantage will belong to those who treat DataOps as a core part of their strategy moving forward, into agentic AI and beyond.

About the Author:
John Harrington is the Chief Product Officer of HighByte, focused on defining the company’s business and product strategy. His areas of responsibility include product management, customer success, partner success, and go-to-market strategy. 

John is passionate about delivering technology that improves productivity and safety in manufacturing and industrial environments. He has spent his 25-year career both delivering software to manufacturers and working for manufacturers in operations roles. This experience has given him a unique perspective on how suppliers and end users each play an integral role in implementing new technology solutions. 

John received a Bachelor of Science in Mechanical Engineering from Worcester Polytechnic Institute and a Master of Business Administration from Babson College.

Subscribe to Industry Today