OT/IT Convergence for AI: Winning With Cross-Functional Ownership

Three Teams, One Meeting Room, Zero Alignment

Picture this: It’s Thursday morning. You’re in a conference room with the OT director, the IT infrastructure lead, and the head of data science. The agenda: approve the next phase of the predictive maintenance program.

• The OT director speaks first. His team just got through a production crisis caused by a software update that disrupted a PLC network. He wants any new system to stay off the production network. Her priority: uptime, safety, zero risk to the running process.

• The IT lead goes next. He’s concerned about data governance. The last pilot pushed unvalidated sensor data straight to a cloud model with no access controls, no schema enforcement, no audit trail. His priority: standardization, security, compliance.

• The data science lead is frustrated. Her team has been waiting eight weeks for access to vibration data from six compressors. The data exists, but it’s locked in a historian that OT controls, behind a firewall IT manages, in a format neither team documented. Her priority: clean, fast, accessible data.

All three are right. And nothing is moving forward. The Accelerating Industrial AI in 2026 survey report shows this dynamic plays out across the industry.

The Organizational Friction Behind Stalled AI

Organizational friction is an ownership problem. And it’s the silent killer of industrial AI programs.

The survey puts hard numbers on what many program managers already feel: 20% cite lack of leadership support as a core challenge for establishing a data backbone. 39% cite budget and ROI uncertainty, which often surfaces when no single team owns the business case end-to-end.

Behind these numbers are cultural divides that predate AI by decades. OT teams optimize for reliability and safety. They’ve been burned by IT-led initiatives that introduced instability into production environments. IT teams optimize for standardization and security. They’ve watched OT teams deploy shadow systems that bypass governance entirely. AI teams optimize for speed and data access. They need to iterate fast, and every gate they have to pass through, such as firewall rules, data access requests, schema approvals, etc., slows them down.

Each team has different toolchains, different KPIs, and different risk tolerances. When nobody owns the data pipeline from edge to model, every handoff becomes a negotiation. Data moves slowly because trust moves slowly.

Why Cross-Functional Ownership Isn’t Optional

Walk through what happens when alignment is missing.

OT collects data from PLCs and sensors but doesn’t contextualize it for AI consumption. Tag names are meaningful to control engineers but opaque to data scientists. There’s no metadata describing what each signal means, what its engineering units are, or where it fits in the asset hierarchy.

IT builds a data lake. It’s well-governed, access-controlled, and compliant. But it doesn’t include real-time operational data because connecting to the OT network requires approvals that take months. The lake is full of ERP transactions and maintenance records, but missing the live machine telemetry that AI models actually need.

The AI team builds models on whatever data they can access, which is incomplete and inconsistent. The result: pilots that work in isolation but can’t be scaled, because nobody built for reuse.

The survey’s leader-versus-laggard comparison makes the pattern clear: the organizations that have moved beyond pilots are the ones where ownership of the data pipeline is shared, not siloed.

Patterns That Are Working in Practice

Builders who have broken through describe a common approach: start with a shared data layer that all three teams can contribute to and consume from.

OT owns the edge and protocol conversion. They’re responsible for connecting PLCs, sensors, and SCADA systems to the data backbone, translating proprietary protocols into a standardized format. They know the equipment, and they control what data leaves the production network.

IT owns governance and cloud integration. They define access controls, enforce data schemas, manage the infrastructure, and ensure compliance. They don’t need to understand every sensor; they need to ensure the data flowing through the backbone meets enterprise standards.

The AI team defines what “AI-ready” means and feeds requirements back into the data pipeline. They specify what context, granularity, and freshness their models need. Instead of building custom pipelines, they subscribe to governed data streams that already carry the structure and semantics their models require.

A unified namespace becomes the shared contract—a common topology everyone can reference. The survey shows 13% have a UNS in production and 16% are piloting or planning adoption. These are the organizations where the ownership conversation has already happened. The architecture is a reflection of the alignment, not the other way around.

The Takeaway: Alignment is An Architectural Decision

Alignment isn’t a soft skill. It’s an architectural decision that manifests in how data flows, who owns each stage of the pipeline, and whether the output is reusable or one-off. The Accelerating AI Use Cases in 2026 report captures how hundreds of practitioners across manufacturing, energy, and transportation are navigating the OT/IT/AI tension, including the specific organizational and architectural patterns that separate leaders from laggards.