Data governance must become an operational discipline
Manufacturers rarely lack data. The harder problem is turning information created across engineering, production and business systems into a reliable asset that can be reused beyond its original purpose. Too many initiatives remain trapped between an encouraging experiment and a data product that people can trust in daily operations.
Gerald Schmidt, Head of Engineering, Digital & Data at DS Smith, believes even the language surrounding experimentation can become an obstacle. “Sometimes it’s the label ‘pilot’ or ‘proof of concept’ that holds us back,” he says. “Building a full data product, with automation, security and governance, needs to be as quick and painless as possible.”
That approach changes the role of governance. It is not there to preserve every project or add layers of approval, but to make it easier to build credible data products, assess whether they work and discontinue those that do not. When the route from idea to governed product is slow, manufacturers risk accumulating pilots without creating a repeatable operational capability.
Governance without bottlenecks
Turning governance into something that supports daily work requires manufacturers to reduce unnecessary human intervention. “There is no escape from self-service and automation,” Schmidt says. “Human touch points are precious and need to be reserved for important decisions in the development life cycle.”
The foundations are practical rather than abstract. Manufacturers need to define which datasets are available, appoint owners and stewards, identify highly sensitive information and create convenient processes for requesting access. When those elements are fragmented or dependent on individual administrators, governance becomes a bottleneck rather than an enabler.
Responsibility also needs to be separated from ownership. “Governance is absolutely everyone’s responsibility, but ownership isn’t,” Schmidt explains. The most appropriate owner will often be a domain expert close to the part of the business that produces and understands the data.
This means approval workflows may need to move away from line managers and administrators toward the people able to judge how the information should be used. For data platform teams, that can require a significant change in how authority is distributed. The purpose is not to weaken control, but to place important decisions with people who understand the business and operational context.
Data quality across a fragmented estate
For large manufacturers, data quality problems often begin with a lack of standardization rather than an absence of effort. Multiple databases, systems of record and data models make it difficult to establish a dependable source of trust. Acquisitions can intensify the problem by bringing together platforms designed around different structures, definitions and operating assumptions.
“The more databases and systems of record we need to cover, the harder it is to attain the elusive single source of trust,” Schmidt says. Standardizing that environment can be expensive and poorly timed against immediate operational priorities. Yet connecting individually sound systems does not automatically produce trustworthy information.
As Schmidt observes, “Three data platforms with satisfactory data quality strung together do not add up to satisfactory overall data quality.” Information can be accurate within each source and still become unreliable when definitions, identifiers or business rules are not aligned across the wider environment. Data quality therefore needs to be assessed across the complete journey, not simply within individual applications.
Technical measurements are only part of the answer. Automated checks within a pipeline can identify missing, inconsistent or invalid data, but they cannot determine whether a data product accurately reflects operational reality. Confidence grows when automated controls and domain expertise align, with people who understand the process confirming that the findings are valid and useful.
“Automated data quality measurement in the pipeline is the easy part,” Schmidt says. “It’s when domain experts use a data product and confirm the validity of its findings that we start to build confidence.”
From dashboards to decisions
The same discipline should be applied when manufacturers assess whether analytics are creating value. Usage statistics may show that employees open a dashboard, but activity alone does not demonstrate that it improves performance. “Daily active users are an insufficient guarantor of business value,” Schmidt says.
A stronger measure is whether a decision can be traced back to the availability of a particular dashboard or data product. Manufacturers should also estimate the financial or operational impact of that decision wherever possible. This creates a clearer connection between engineering and operational information, management priorities and measurable outcomes.
It also shifts the conversation away from how many dashboards have been deployed or how frequently they are opened. A data product used by a relatively small number of people may be extremely valuable if it supports an important maintenance, quality or production decision. Conversely, a widely viewed dashboard may deliver little value if it does not change what anyone does.
Data literacy is essential to making that connection, but training should not be confined to a particular application or cloud platform. “Technologies change rapidly, but data fundamentals evolve slowly,” Schmidt says. “Companies that restrict training to one application vendor or one public cloud provider risk maneuvering their staff into a corner.”
Platform-agnostic learning gives employees a more durable understanding of how data is structured, governed and applied. It allows people to question the information they receive, understand its limitations and assess whether a result is credible. These capabilities will remain useful even as platforms, interfaces and technology providers change.
Manufacturers also do not need to begin by searching for perfect enterprise-wide data. Much of the information they need is already useful within individual teams, and the employees working with it every day are often the strongest candidates to own it. The immediate task is to understand the value those teams already derive and the obstacles they face.
“Our task is to make that data useful outside its local context,” Schmidt says. “It needs to become reusable across business units, with plausible use cases that demonstrate in-year value.”
Governance for service accounts and agents
The next challenge is ensuring governance can support systems that consume and act on data without continuous human involvement. Schmidt argues that manufacturers should apply the principles of Jeff Bezos’s 2002 API mandate, making new data sources visible and controlling access through APIs. The same governed access available to employees can then be extended to service accounts and, increasingly, AI agents.
This will require much more than connecting an agent to a collection of enterprise systems. Data owners must remain in control, access should follow least-privilege principles and strict guardrails must govern what automated systems can see and do. The identity requesting the data may change, but the need for ownership, traceability and controlled access will not.
“The area of data governance I expect to dominate the coming year is the extension of our existing data platforms first to service accounts and then to near-autonomous agents,” Schmidt says. These agents will need to be “controlled by data owners, with least privilege and strict guardrails.”
Governance will be fit for the next phase of industrial digitalization only when it can secure this environment without preventing legitimate data use. The manufacturers best placed to benefit from AI will not necessarily be those with the greatest volume of data. They will be those that can make trusted information available quickly, securely and in a form that both people and automated systems can use.
Join the conversation at the Manufacturing Data Summit
Gerald Schmidt, Head of Engineering, Digital & Data at DS Smith, will be speaking at the Manufacturing Data Summit UK 2026 as part of the panel, “Data governance, quality and literacy: From frameworks to real business outcomes.” The discussion will examine what effective data governance looks like in practice, how manufacturers can connect operational and business ownership, and how data quality and literacy can support better decisions and stronger AI outcomes.
Taking place in London on October 6, 2026, the Manufacturing Data Summit will bring together manufacturing, operations, data and digital leaders to explore how companies can unlock greater value from their information. Through keynote presentations, panel discussions, case studies and roundtables, the event will address data governance, industrial AI, analytics, digital twins, predictive maintenance, cybersecurity and the foundations required to scale innovation successfully.
Register to attend the Manufacturing Data Summit and hear directly from Gerald Schmidt and other manufacturing leaders turning data strategies into measurable operational outcomes.
