Discipline 05

Operate

Is it working in production, and do the economics hold?

Most AI programs cannot answer the question a CFO asks first: what did this change, in numbers we can defend? Adoption gets reported as success. Time saved is estimated rather than measured. Projected value is blended with realized value. The result is a program that feels productive and cannot prove it, and that is the program whose budget gets cut.

Operating AI well starts before deployment, with a baseline and a measurement charter signed off by the business owner and a finance partner. It continues with the harder work of turning task-level gains into something that moves a financial statement line, which usually means redesigning the workflow so released capacity goes somewhere deliberate. And it ends in the boardroom, where the report has to separate actuals from projections, name the risks, and ask for decisions.

Agents add a new cost problem. Their spend is decided at runtime, not at design time, so the discipline extends to per-task attribution, budget caps enforced in the runtime, and cost measured against outcomes rather than calls.

Read01

Measurement Design

How to build a baseline-first measurement system that produces numbers boards believe and business leaders can act on.

6 min read
02

Financial Linkage

How to turn task-level efficiency gains into P&L impact through workflow redesign, pre-registered attribution and a finance partnership.

7 min read
03

Board Reporting

What boards need to see on AI: portfolio health, risk posture, realized value and strategic alignment, on a two-cadence reporting cycle.

6 min read
04

FinOps for Agents

Financial operations for agents: per-task cost attribution, budget caps enforced in the runtime, and cost measured per outcome rather than per call.

6 min read
ProofProof

Decision Artifacts

Worked board memo, investment scorecard, model inventory, phase-gate review and risk classification worksheet, each with a downloadable template.

10 min read
Next discipline06 OrganizeWho owns AI, and how does adoption spread?