The Role · Companion to the Specification
The executive specification for the discipline's owner.
Every wave of technology produced an operations discipline once its flow reached scale, and every discipline that reached the executive level was eventually held by a named role. This page specifies that role for AI.
01 · Position Summary
The Head of AI Operations is accountable for turning AI from isolated tools and pilots into a scalable, governed, economically legible operating capability. The role owns the AI portfolio lifecycle, the governance and evaluation regime, the economics, adoption, and the measurement system.
It is judged on business outcomes, not technology outputs: cost, cycle time, throughput, quality, revenue, and audited risk posture. Engineering builds AI. This role ensures AI improves the operations and economics of the enterprise.
That is the boundary the title draws. Everything below specifies what the owner of it does, how the work is measured, and where the role sits.
02 · The Mandate
The mandate is not a project list. It is the set of recurring functions the role owns for as long as the enterprise runs AI. The Role Standard carries the full detail; this is the shape of it.
Run one inventory and one lifecycle for every AI initiative. Own the gates, the intake, and the governance constitution, and hold exactly one accountable owner per workflow.
Find the value in the workflow, not the model. Map processes as they actually run, redesign around the constraint, and measure gains end-to-end against frozen baselines.
Run the evaluation regime: golden sets, faithfulness measurement, and per-stage error accounting. Match the control architecture to the cost of an undetected error.
Make AI economically legible. Meter fully loaded cost per workflow execution, attribute every unit of spend, and run the routing reviews that turn model choice into procurement.
Convert shadow AI into governed capability on a paved road. Treat adoption as design, not decree, and measure real use, not logins.
Deliver the measurement system, from the operational panel to the board page. Never publish a number the function cannot defend under audit.
03 · How the Role Is Measured
A discipline is its metrics. The measurement system groups into four families, each answering one plain executive question. A sample of the catalog is below; every metric in the standard names the way it can be gamed and the counter.
Is it worth it?
AI ROI against frozen baselines · Fully loaded cost per workflow execution · Time-to-ROI per workflow.
Is it working?
Production deployment rate · End-to-end cycle-time reduction · SLO attainment inside the error budget.
Is it used?
Adoption rate on completed executions · Depth of use by median and quartile · Shadow-AI reduction and conversion.
Is it safe?
Measured error rate on golden sets · Human review rate against class policy · Incident rate and time to contain.
04 · The First 365 Days
The single organizing principle: show a measured win early, and never publish a number the function cannot defend under audit. One year, from ad hoc to a measured system.
Build the portfolio inventory and the first fully loaded spend picture. The readout is usually the first time anyone has seen the whole board, and it alone justifies the role.
Take two or three workflows through the full lifecycle gates with committed owners and frozen baselines. Debug the gates, not just the workflows, and open a public intake.
Pass the first workflows through the production gate and publish measured deltas. Run the first eval-driven route review and capture the routing savings.
Expand the portfolio through the proven lifecycle, implement showback with Finance, and deliver the year-one board readout as an investment case, not a budget ask.
05 · Reporting-Line Doctrine
The reporting line is the most load-bearing structural decision in the role, and it is the one most often gotten wrong.
Reporting-line doctrine
The reasoning is structural, not political. AI Operations is accountable in operational and financial outcomes across functions, and the COO already owns cross-functional outcome accountability. Reporting into Engineering converts the function back into MLOps, where accountability for the artifact displaces accountability for the outcome. Reporting into IT frames AI as a support service to be ticketed rather than an operating capability to be run. The reporting line is a statement about what kind of thing the enterprise believes AI is.
06 · Ideal Background
The strongest candidates are operations leaders who have built AI systems, not AI engineers who have observed operations. The hard problems of the role are accountability design, cross-functional coordination, unit economics, adoption, and governance that survives audit. They are twenty-year operations problems wearing a new technical surface.
The technical fluency is trainable and verifiable. The operator judgment is the scarce input. Candidates who have personally built governed, evaluated, multi-model AI pipelines, even at small scale, have run the entire discipline in miniature and know where it bites.
The profile in one line
The role is, structurally, COO training for the AI-era enterprise. It rehearses the same portfolio: cross-functional outcomes, economics, risk, and change.
The Standards
This page is the summary. The Role Standard gives the complete competency profile, the normative requirements, and the reporting, team, and career path. The AI Operations Specification v1.0 is the 57-requirement enterprise standard the role exists to implement.