More data, more problems? According to Gartner, this is a real possibility for B2B manufacturers in the age of AI — the research firm predicts that by 2027, 60% of organizations “will fail to realize the anticipated value of their AI use cases due to incohesive data governance.”
It makes sense; as manufacturers adopt the always-connected architecture of Industry 4.0, the volume of data generated and its use by AI tools can pose challenges for management, security, and efficiency.
The result is Gartner’s rather pessimistic prediction. Unless top manufacturers overhaul current governance efforts, spending on AI may not live up to intelligent expectations.
Manufacturers aren’t just inputting numbers and managing spreadsheets. To keep pace with Industry 4.0 expectations, companies must build predictive supply chains, create digital twins, and deploy autonomous agents, all of which require robust AI infrastructure.
For AI to be effective, however, data is required.
Lots, and lots, and lots of data.
While volume is no problem for manufacturers, recent estimates from IBM suggest that manufacturers generate approximately one terabyte of data per day. Ensuring this data is effectively collected, analyzed, and applied is more challenging.
This is made more challenging by the diversity of data collected. Gone are the days of single or limited-source data. Now, companies must manage distributed data sources that may be internal, external, or a combination of both.
Three categories of manufacturing data are common: people, process, and product. People include both clients and employees.
While anonymized people data is useful for large-scale trend and demographic analysis, failure to properly obfuscate identifying elements can create compliance issues.
While many manufacturers collect first-party data — information provided by clients on request — there’s an increasing push for zero-party data, which is information offered freely by clients or customers to facilitate personalized experiences.
Process data is collected from sources such as production line machinery, programmable logic controllers (PLCs), computerized maintenance management systems (CMMS), and enterprise resource planning (ERP) tools.
Collection and analysis of this data can help pinpoint performance problems before they impact productivity.
Product data rounds out the list. This includes information listed on product detail pages (PDPs), internal documents, and shop-floor data about batch quality and consistency.
More accurate and relevant data means better decisions about price, market positioning, and new product development.
In many cases, the collection and management of this data is a goal, not a byproduct. While the first generation of big data put companies in a reactive position as information volumes skyrocketed, infrastructure couldn’t keep pace.
Top manufacturers now recognize the value of owning and managing this data from end-to-end.
AI-driven governance can help manufacturers make sense of disparate data without putting information at risk.
The underlying idea of AI governance is straightforward. Instead of relying on humans to classify data, identify challenges, and apply analytics, organizations use AI tools capable of connecting the dots.
Using a combination of machine learning (ML) and available training data, these tools “learn” how to correctly classify data risks and integrate operational priorities into decision-making.
Consider the rise of agentic AI. While autonomous agents reduce reliance on human agents and provide clients with in-depth data on demand, companies need strict rules around the type of data these tools can access and what they can do with this data.
This is the role of AI-powered data governance — determining if requests for data access, use, and analysis should be permitted or prevented based on the type of data requested, the nature of the AI function, and the way data will be used or changed.
Key benefits of AI governance include:
AI isn’t a silver bullet. While it offers the potential for improved governance, it also carries concerns around misuse and misinformation.
According to EY survey data, 71% of employees say they are worried about AI adoption. Top concerns include cybersecurity (75%), ethical and moral issues (71%), and the quality of AI outputs (66%).
If governance tools are compromised by malicious actors, the results range from lost data to modified operational parameters, which can lead to incorrect actions or recommendations.
Consider a recently discovered piece of generative AI (GenAI) malware that uses an image to gain access. As noted by Tech Radar, when the infected image is resized to fit model prompt parameters, the resizing causes artifacts that are interpreted by the AI engine as commands, leading to risky and unexpected model behaviors.
AI governance models also require consistent oversight to ensure data is used ethically and morally.
For example, if governance models permit the use of clients’ personal or sensitive company data in decision-making, manufacturers may be in violation of legislation such as the General Data Protection Regulation (GDPR) or the California Consumer Privacy Act (CCPA).
Low-quality AI outputs round out the list.
Consider a manufacturer using AI governance solutions to help ensure new products both follow company naming conventions and don’t run afoul of copyright issues.
If governance tools haven’t been properly trained and tested, they may return results that appear legitimate but are in fact confidently incorrect. If manufacturers act on this data without double-checking for consistency, they could find themselves facing legal challenges.
AI-driven data governance maturity doesn’t happen overnight. Instead, it requires consistency and repetition to produce reliable results.
While there’s no single way to achieve maturity, four steps on the path are common.
Modern data governance frameworks require a single source of truth. AI helps enable this approach by collecting data from multiple sources, including client-provided information, internal product pages, and publicly available assets.
Data quality plays a key role in effective governance and decision-making. This is because AI outputs are only as good as their inputs — even the best models can return wrong answers if they don’t have access to high-quality data.
Common causes of low-quality data include inaccuracy, duplication, and bias. To improve governance recommendations, companies need to implement automated data quality checking solutions that both identify issues and alert staff to act.
Historical data may be stored in locations or formats that make it difficult to access or use. Optimizing this data enables integration with governance systems and provides additional context that can inform governance decisions.
Optimization may require reformatting, removal, and re-entry, or relocation to enable the use of cloud-based and other resource-scalable services.
AI doesn’t work without a human in the loop. While tools are evolving, the artificial nature of machine intelligence means governance tools will always benefit from human oversight.
This includes creating and evaluating ML models, reviewing and improving outputs, and flagging issues with inaccuracy, bias, or security.
The rise of client personalization preferences has moved the B2B model closer to its business-to-consumer (B2C) counterpart.
From the need for rich product content and seamless product information management (PIM) integration to the creation of omnichannel ecosystems that meet customers where they are, top manufacturers can’t afford to ignore the impact of digital transformation.
AI represents the next stage of this B2B change. What began with data analysis and simple chatbots has evolved to include adaptable AI agents and iterative models that become more accurate and intuitive over time.
Without effective governance, however, AI is a non-starter. Granular control over data sources, data uses, and data outputs is critical to reduce the risk of AI bias, minimize security risks, and ensure regulatory compliance.