OpenAI's July 14 guidance for managing AI investments recommends five moves: improve visibility into usage and spend, evaluate efficiency by outcome ROI, govern advanced workflows before scaling, fund workflows that compound, and match capacity to proven demand.
Primary source: OpenAI, “How to manage AI investments in the agentic era”.
The hard part is the denominator. “This agent used $800” says little. “This workflow cost $14 per accepted reconciliation, including review and rework” can support a decision.
Here is a one-page ledger I would require for an agent pilot.
Define one accepted outcome
Do not start with tokens, seats, or tasks launched. Define the business state that counts after review.
workflow: vendor-invoice-reconciliation
accepted_outcome: "invoice matched, exceptions reviewed, result posted"
owner: finance-ops
pilot_window_days: 21
minimum_sample: 100 invoices
quality_gate:
false_postings: 0
exception_recall: ">= 0.98"
reviewer_minutes_p50: "<= 3"
A generated draft is not an outcome if a person must rebuild it. An agent run is not successful if its result never enters the system of record.
Capture the complete cost
AI cost
+ orchestration and observability
+ human review
+ rework
+ incident handling
+ allocated implementation cost
= total workflow cost
Use a table with declared variables:
| Variable | Meaning | Example only |
|---|---|---|
C_model |
model and tool-call spend | $600 |
C_platform |
workflow infrastructure | $200 |
H_review |
reviewer hours | 35 |
R_hour |
loaded reviewer rate | $45 |
C_build |
pilot build cost allocated to window | $2,000 |
N_accept |
accepted outcomes | 850 |
total = C_model + C_platform + H_review * R_hour + C_build
cost_per_accepted_outcome = total / N_accept
With the illustrative numbers, total cost is $4,375, or about $5.15 per accepted outcome. These are not benchmark claims; replace every value with measured data.
Compare against the real baseline
The baseline must use the same unit and quality gate:
| Metric | Manual baseline | Agent pilot |
|---|---|---|
| attempted invoices | 1,000 | 1,000 |
| accepted outcomes | 920 | 850 |
| false postings | 0 | 0 |
| total cost | $6,000 | $4,375 |
| cost per accepted outcome | $6.52 | $5.15 |
| median cycle time | 18 min | 7 min |
A lower cost per attempt can hide lower completion. A faster median can hide an unacceptable tail. Report attempted, accepted, escalated, rejected, and incorrectly completed counts.
Run sensitivity before buying capacity
The largest uncertainty is often human review, not token price.
break_even_review_minutes =
(baseline_cost_per_outcome - non_review_agent_cost_per_outcome)
/ reviewer_cost_per_minute
Calculate three scenarios:
| Scenario | Acceptance | Review minutes | Decision |
|---|---|---|---|
| downside | 70% | 7 | stop |
| expected | 85% | 3 | continue pilot |
| upside | 93% | 1 | prepare controlled scale |
If the decision only works under the upside case, it is not ready for an annual commitment.
Add governance as a costed requirement
Advanced workflows need named controls before scale:
controls:
approval_owner: finance-ops-lead
permission_scope: draft-only
audit_retention_days: 90
rollback: disable posting credential
incident_owner: platform-oncall
review_expiry: "2026-08-31"
A control with no owner or expiry is a sentence, not a control. Draft-only access may reduce automation upside, but it also limits the pilot's blast radius. Price both review labor and risk reduction honestly.
Hard stop conditions
Stop or redesign when any condition is met:
- one unauthorized irreversible action;
- the accepted-outcome denominator cannot be reconstructed;
- more than 10% of runs lack traceable cost;
- reviewers routinely redo the work outside the captured workflow;
- the downside scenario exceeds the baseline after two iterations;
- capacity is being purchased to solve demand that has not been measured.
Scale only after quality passes, unit economics survive the expected and downside cases, and operational ownership exists.
The most useful part of agentic investment guidance is not permission to spend more on agents. It is the shift from infrastructure consumption to workflow outcomes. A model can become cheaper while a workflow remains expensive; a costly model can be economical if it reliably removes a constrained, high-value step.
What denominator does your AI dashboard currently omit: accepted outcomes, reviewer time, rework, or failures?
Top comments (0)