Section III, cross-cutting discipline

Financial governance

Agents spend money every time they call a model or a paid tool, and they can spend it quickly. Financial governance makes agent costs visible, puts each one on a named account, stops spending at a limit before it happens, and checks that each agent returns what it was expected to.

What financial governance covers

Cost attribution

Every cost is assigned to a named account and cost center. When a marketing team's agent spends, the spending reaches the marketing budget.

Detailed tracking

Costs are tagged by agent, by type of task, by project and by business objective, and each piece of work records what it cost.

Model mix

Costs are tracked across the model providers an organization uses, so that each step can go to the model that does it well enough for the least cost.

Ceilings are the easy part of financial governance. The view across everything is harder, especially when work passes along a chain of agents. Cost across a chain is covered under Network, and each agent's cost is set against its value by the value owner named in section I.

Control objectives

There are 8 control objectives here. Each one states what it asks for, the risk it addresses and how it is measured. It then says what each risk tier requires, what evidence proves it, and who may check that evidence. The Operating Model page explains the tiers.

FIN-01

Every cost on a named account

Attribute all agent spending to a named account and cost center, so that someone answers for every cost.

Risk addressed: Spending that nobody answers for
Key metric: Share of agent spend attributed to an account

What each tier requires

Tier 1 Exploratory
Experiments charge to the team's sandbox budget.
Tier 2 Operational
Every agent's spending lands on a named cost center.
Tier 3 Critical
Each piece of work records the account that pays for it.
Tier 4 Strategic
Attribution is reconciled with suppliers' invoices each month.

Evidence: Cost records by account, and the reconciliation with invoices.
Who may check it: Internal audit and an outside auditor.

FIN-02

Spending ceilings

Set a limit per task and per period for each agent, and stop work that would pass a limit before it spends.

Risk addressed: Runaway costs
Key metric: Work refused at a ceiling, and spend over budget

What each tier requires

Tier 1 Exploratory
The sandbox has a fixed budget that stops spending when it runs out.
Tier 2 Operational
Each agent has a period limit set by its owner.
Tier 3 Critical
Each agent has limits per task and per period, enforced outside the agent before money is spent.
Tier 4 Strategic
Limit changes need finance approval, and breaches are reported to the executive sponsor.

Evidence: The configured ceilings, the record of refused work and any changes to the limits.
Who may check it: Internal audit, an outside auditor and a partner.

FIN-03

Cost per job

Record what each piece of work cost, and set a baseline cost per task for each agent, so that inefficiency can be seen.

Risk addressed: Hidden inefficiency
Key metric: Share of agents with a cost-per-task baseline

What each tier requires

Tier 1 Exploratory
The experiment reports its total cost.
Tier 2 Operational
Each agent reports cost per task.
Tier 3 Critical
Each piece of work records its cost, and the baseline is reviewed each quarter.
Tier 4 Strategic
Cost per task is compared with other ways of doing the work.

Evidence: Cost per job records and the baseline.
Who may check it: Internal audit.

FIN-04

Unusual spending detected

Detect unusual patterns of spending that may point to a fault or abuse, and alert the owner.

Risk addressed: Cost problems found too late
Key metric: Mean time to detect unusual spending

What each tier requires

Tier 1 Exploratory
The sandbox budget is checked each week.
Tier 2 Operational
Spending is compared with the previous period each week.
Tier 3 Critical
Unusual spending alerts the owner within hours.
Tier 4 Strategic
Unusual spending can pause the agent automatically until the owner reviews it.

Evidence: Alert settings and the record of alerts raised.
Who may check it: Internal audit.

FIN-05

Return on investment

Track each agent's actual return against what was projected when it was approved.

Risk addressed: Agents that destroy value
Key metric: Share of agents meeting their return target

What each tier requires

Tier 1 Exploratory
The experiment reports what it cost and what it showed.
Tier 2 Operational
Actual cost and benefit are reported at the review date.
Tier 3 Critical
The value owner compares the return with the projection each quarter.
Tier 4 Strategic
The comparison is reported to the investment committee.

Evidence: The projection, the actual figures and the value owner's review.
Who may check it: Internal audit and an outside auditor.

FIN-06

Financial reporting

Report agent costs accurately and on time, so that decisions rest on good figures.

Risk addressed: Poor decisions from bad figures
Key metric: Accuracy of reported agent costs

What each tier requires

Tier 1 Exploratory
No separate report is required for an experiment.
Tier 2 Operational
Costs are reported each month by agent.
Tier 3 Critical
Costs are reported by agent, by business objective and by account.
Tier 4 Strategic
Reports are reconciled with the general ledger and reviewed by finance.

Evidence: The monthly reports and their reconciliation.
Who may check it: Internal audit and an outside auditor.

FIN-07

Acting on cost findings

Act on findings that would cut cost, such as sending a step to a cheaper model that does it well enough.

Risk addressed: Inefficiency that persists
Key metric: Share of cost findings closed

What each tier requires

Tier 1 Exploratory
No requirement applies to an experiment.
Tier 2 Operational
Cost findings are recorded.
Tier 3 Critical
Each finding is acted on or declined with a reason within a set time.
Tier 4 Strategic
Savings are measured and reported.

Evidence: The cost findings log with actions and savings.
Who may check it: Internal audit.

FIN-08

Chargeback and showback

Allocate agent costs to the teams that use them accurately enough that teams accept the charges.

Risk addressed: Disputed or wrong allocations
Key metric: Allocation disputes

What each tier requires

Tier 1 Exploratory
No allocation is needed for an experiment.
Tier 2 Operational
Costs are shown to the teams that use the agent.
Tier 3 Critical
Costs are charged to the teams that use the agent, by their use.
Tier 4 Strategic
Charges between organizations follow the contract, and both sides' records reconcile.

Evidence: Allocation reports and the dispute log.
Who may check it: Internal audit and a partner.

Quick reference

ID	Objective	Key metric
FIN-01	Every cost on a named account	Share of agent spend attributed to an account
FIN-02	Spending ceilings	Work refused at a ceiling, and spend over budget
FIN-03	Cost per job	Share of agents with a cost-per-task baseline
FIN-04	Unusual spending detected	Mean time to detect unusual spending
FIN-05	Return on investment	Share of agents meeting their return target
FIN-06	Financial reporting	Accuracy of reported agent costs
FIN-07	Acting on cost findings	Share of cost findings closed
FIN-08	Chargeback and showback	Allocation disputes