Executive Briefs · · 6 min read
Why Per-Agent ROI Is the Wrong Question
Boards want proof that AI agents pay back, and they are right to ask. But demanding an ROI number for each agent produces false precision and kills the experiments worth keeping. The right unit of measurement is the portfolio.
Douglas Siqueira · Agent Chaos
The pressure is real
In Dataiku's September 2026 survey of 685 CIOs, 97% reported more board pressure to show AI returns. Yet 72% cannot consistently confirm whether all their agents deliver business outcomes, and 76% say their organization lacks mature ROI measurement across most AI initiatives.
The broader picture is similar. In McKinsey's State of AI 2026, 37% of respondents attribute at least some EBIT impact to AI — and only about 6% qualify as high performers, attributing 5% or more of EBIT to it.
Why per-agent ROI fails
Agents rarely work alone. What a triage agent saves shows up in customer service; what service gains shows up in retention. Measured one by one, each agent looks either useless or miraculous.
Shared costs make it worse. Platforms, integration work and human review time get split across agents with arbitrary rules, and the result is a number nobody trusts.
Timing is the third trap. Strategic agents may need a deliberate investment period. A per-agent payback rule applied too early retires them before they could work, while expensive automation of cheap work survives because its spreadsheet looks tidy.
The cost lands in one place. The value shows up in another.
Judged one by one, the first agent looks like pure cost and the last like pure gain. The portfolio view judges the chain.
What to measure instead
Measure the portfolio, the way a CFO already measures capital expenditure. That means one total, two classes and an explicit review rhythm.
- Total cost of the portfolio: running costs, integration and the human time spent reviewing agent output.
- Contribution by workflow, not by agent: the cycle time, cost per case or revenue the workflow moves.
- Efficiency agents are judged on a short, explicit payback window.
- Strategic agents are judged on milestones and a fixed review date, not on immediate payback.
- Retirement is a capability: every agent has a condition under which it is switched off.
A one-line portfolio view
For every agent in production, leadership should be able to read one line: owner, class (efficiency or strategic), monthly cost, workflow outcome, threshold and next review date. If that line cannot be written, the problem is not measurement. It is governance.
Executive Briefings
One brief per month: a pattern, the data behind it and three questions for your team. You also get The Board Questions as a printable one-pager.
Sources
- Dataiku / The Harris Poll — Global AI Confessions Report: CIO Edition (Sep 2026, 685 CIOs)
- McKinsey — The State of AI in 2026, as reported by FM Magazine (Sep 2026)
- Gartner — Over 40% of agentic AI projects will be canceled by end of 2027 (Jun 2025)
Figures come from public research. Several studies are sponsored by vendors that sell related products; each source is linked so you can judge it yourself.
Keep reading
- Agent Sprawl: When More Agents Means Less Control
AI agent sprawl is not about how many agents you run. It is about running them without a portfolio. The signs, the cost and four decisions that contain it.
- Who Owns an AI Agent?
When an AI agent fails, accountability splits across four teams. Why "IT owns it" is the wrong answer, and how to name a single economic owner for every agent.