SEC-01 / AI Governance

AI governance roles and RACI for mid-market manufacturers

Mid-market manufacturers need a small AI governance group drawn from roles they already have, a clear RACI across the AI lifecycle, and a firm split between what plant managers own (use, process, safety) and what IT owns (platforms, security, data). In the EU, bring employee representatives in early.

7 min read Updated Plant managersCIOsCISOsOperations leadersHR leaders

Key takeaways

  • You do not need a Chief AI Officer. You need six to eight existing leaders meeting monthly for an hour, with a named chair.
  • Every AI use case has one accountable business owner, usually in operations, not in IT.
  • Plant managers own the process, safety and people consequences. IT owns platforms, identity and data protection. OT engineering owns anything that touches control systems.
  • Use a RACI across the lifecycle: idea, triage, buy or build, deploy via MOC, operate, change, retire.
  • In Germany and several other EU countries, works councils have rights over systems that can monitor employees. Consult before you buy, not after.

In a mid-market manufacturer, nobody's job title says "AI". The CIO has three people and a backlog of ERP tickets. The controls engineer is also the OT security team. The quality manager is fighting a customer complaint. Yet someone has to decide whether the new vision system can auto-reject parts, and whether the scheduling vendor can use your data to train its model.

Good AI governance at this size is about clear ownership with minimum overhead. This page sets out a small governance group, the roles around it, a RACI across the AI lifecycle, the split between plant and IT, and how to handle employee representatives in the EU.

Does this apply to you?

If you have more than a handful of AI use cases, or any that write to control systems, release product or touch worker data, you need defined roles. Frameworks expect it too: the NIST AI RMF's Govern function and ISO/IEC 42001 both require defined roles and responsibilities. In the EU, the AI Act places duties on deployers (for example, using high-risk systems according to instructions, assigning human oversight to competent people, and informing workers' representatives before deploying high-risk AI in the workplace). Someone has to own those.

The AI governance group

Keep it small, senior enough to decide, and built from people who already exist:

Seat Typical role Why they are there
Chair COO or head of operations Most AI value and risk sits in operations; can stop a deployment
IT CIO or IT manager Platforms, identity, data protection, SaaS contracts
OT / engineering Engineering manager or lead controls engineer Anything touching PLCs, SCADA, networks, OEM machines
Quality Quality manager Product release, inspection validation, customer requirements
EHS EHS manager Safety functions, worker safety, risk assessments
HR HR director Workforce impact, monitoring, training, works council relations
Legal / compliance In-house counsel or external Regulation, contracts, data protection (can attend on demand)
Secretary AI governance lead Runs the register, prepares triage, tracks actions

In EU sites with a works council, consider a standing invitation for a council representative to the sessions that cover people-related AI, or a separate regular briefing. It builds trust and saves time later.

Cadence: one hour monthly, plus ad hoc for urgent high-tier approvals. Agenda: new register entries, high-tier decisions, incidents, policy exceptions, regulatory updates. Most low and medium tier items never reach the group; they are approved through MOC and the register. The AI governance committee charter gives you terms of reference to adapt.

Plant-floor reality: In multi-site groups, a central group works only if each site has a named AI contact, usually the site engineering or CI lead. Otherwise the group approves things that sites then deploy differently. The site contact is the person who checks the MOC form actually asks the AI question.

Roles around the group

  • Executive sponsor. Managing director or CEO. Signs the policy, sets risk appetite, reports to the board.
  • AI governance lead. 10 to 20 percent of a role. Owns the register, policy upkeep, triage process, training plan.
  • Use-case owner (accountable). A named business person per use case. Owns purpose, outcomes, acceptance and decommissioning.
  • System owner (technical). IT, OT engineering or vendor. Keeps it running, patched, backed up and secure.
  • Data owner. Owns the data class decisions for recipes, process data, quality data, HR data.
  • Human overseer. The operator, planner or engineer who reviews AI output in daily use. Needs training and authority to override.
  • Purchasing. Gatekeeper for AI in software and equipment buys.
  • OT security owner. Often the same engineer as above. Reviews new conduits, remote access and edge devices.

Plant managers vs IT: who owns what

This is where most arguments start. A clean split:

Plant manager / operations owns IT owns OT / engineering owns
Whether the use case is worth doing Platform and tool selection for enterprise AI Integration with PLCs, SCADA, HMIs, robots
Process, safety and quality consequences Identity, access, licensing Network zones and conduits for AI devices
Operator procedures and fallback Data protection and retention OEM coordination and remote access
Training and competence of overseers Cloud and SaaS security reviews MOC for control changes
Stopping use if outcomes are wrong Monitoring of GenAI usage Validation on the line, FAT/SAT

The principle: the person who lives with the consequences on the line is accountable for the use case. IT and OT are accountable for the technology being safe and secure. For the broader IT/OT question, see IT/OT convergence governance.

RACI across the AI lifecycle

R = responsible, A = accountable, C = consulted, I = informed.

Lifecycle step Use-case owner AI gov. lead IT OT / Eng. Quality EHS HR Legal Gov. group
Propose use case A/R C C C C C I I I
Register entry and triage A R C C C C C I I
High-tier assessment A R R R C C C C Approves
Vendor due diligence A C R R C I I C I
Contract A I C C I I I R I
Deploy via MOC A I R R C C I I I
Validation and acceptance A I C R R C I I I
Operate and monitor A I R R C I I I I
Model or vendor change A C R R C C I I High tier only
Incident A C R R C R if safety C C I
Retire A R R R I I I I I
Worker-facing AI (any step) A R C C I C R R Approves

Adjust the columns to your organization, but keep one A per row. Two accountable parties means none.

Works councils and unions in the EU

AI that touches people is where governance most often stalls. Plan for it:

  • Germany. Works councils have co-determination rights over the introduction and use of technical systems designed to monitor employee behavior or performance (Section 87(1) No. 6 of the Works Constitution Act). German courts have read "designed to" broadly, so a system objectively capable of monitoring can be enough. Many AI tools, including Microsoft 365 Copilot and MES with operator logins, can fall in scope. Many companies negotiate a framework agreement on AI and IT systems rather than one agreement per tool.
  • Other EU countries. France, the Netherlands, Austria and others have information and consultation rights with different triggers. Check locally.
  • EU AI Act. Employers deploying a high-risk AI system in the workplace must inform workers' representatives and affected workers before putting it into service (Article 26(7)). As of November 2026, Annex III high-risk obligations, including employment uses, apply from 2 December 2027 following the Digital Omnibus.
  • GDPR. Worker monitoring usually needs a data protection impact assessment.

Watch out: Introducing a productivity or quality analytics tool as a "pilot" without consultation is a common way to sour relations for years. In some jurisdictions it can also make the deployment unlawful until an agreement is reached. Involve HR and the council at the idea stage.

In the US and UK there is no direct equivalent, but union agreements, state privacy laws and employee trust still argue for early disclosure.

NIS2 and accountability at the top

If your company is an essential or important entity under NIS2 (many medium-sized and larger manufacturers in the sectors NIS2 lists are), management bodies must approve cybersecurity risk-management measures, oversee their implementation and follow training (Article 20). AI systems in OT are part of that attack surface. Make sure the AI governance group reports into the same management oversight as cybersecurity. See board questions on AI and OT risk.

Effort

  • Setting up the group and charter: 2 to 3 days of the governance lead's time.
  • Monthly meeting: 1 hour times 6 to 8 people, plus 2 to 4 hours of preparation.
  • Use-case owners: an hour or two per use case per year, more for high tier.
  • Works council framework agreement: typically weeks to months; start early.

Common mistakes

  • Making IT accountable for every use case. IT cannot own process outcomes.
  • No OT seat. The group then approves things that cannot be integrated safely.
  • A large committee. Twelve people meeting monthly becomes a status meeting.
  • No authority to stop. A group that can only advise is decoration.
  • Late works council involvement. Costs months.

Monday-morning checklist

  • Name an executive sponsor and a chair.
  • Pick six to eight members from the table above and book a standing monthly hour.
  • Adapt the committee charter and agree tier approval thresholds, using the AI risk assessment tiers.
  • Assign a named use-case owner to every row of your AI register.
  • In EU sites, brief the works council chair on your AI policy draft before it is final. The AI policy guide shows what to include.

This page is general guidance, not legal advice. Confirm regulatory and employment-law obligations with your counsel.

Sources and further reading

  1. Regulation (EU) 2024/1689 (Artificial Intelligence Act), EUR-Lex
  2. Directive (EU) 2022/2555 (NIS2), EUR-Lex
  3. NIST AI Risk Management Framework
  4. ISO/IEC 42001:2023 Artificial intelligence management system (ISO)

This guide is general information for manufacturers, not legal advice. Confirm regulatory obligations with qualified counsel.

FAQ

Frequently asked questions

Who should chair an AI governance committee in a manufacturing company?

Usually the COO or head of operations, because most AI value and risk sits in operations. The CIO is a common alternative where IT already runs digital programs. Avoid making legal or compliance the chair: the group should enable good use cases, not only block risky ones. Whoever chairs needs authority to stop a deployment.

Does a mid-market manufacturer need a Chief AI Officer?

Rarely. With 100 to 2,500 employees, AI governance is a part-time duty spread across existing roles. A named AI governance lead, often in IT or operations excellence, spending 10 to 20 percent of their time coordinating the register, triage and committee is usually enough. Reconsider if AI becomes core to your product.

When does a works council need to be consulted about AI?

It depends on country law. In Germany, the works council has co-determination rights over technical systems capable of monitoring employee behavior or performance, which can include many AI tools. Under the EU AI Act, employers deploying high-risk AI at work must inform worker representatives. Involve them early, ideally with a framework agreement on AI, and check local law with counsel.

What is the difference between the AI owner and the system owner?

The AI or use-case owner is the business person accountable for why the AI is used and its outcomes, for example the quality manager for vision inspection. The system owner is the technical owner who keeps it running, patched and secure, usually IT, OT engineering or the vendor. Both are named in the AI register.

Related guides

Know where your plants stand

Score your AI governance, OT security and data readiness in about 10 minutes each. Results stay in your browser.