The thesis. Most organizations now use artificial intelligence (AI), yet boards often review that spending one project at a time. Governing AI as a portfolio of staged bets, with gates, kill criteria agreed in advance and a short list of value metrics, lets a board fund ambition and still stop what is not working. Without that discipline, pilots accumulate and no one ends them.
Use is broad. Value is concentrated.
The pattern in the evidence is now clear. McKinsey’s 2025 global survey found that 88 percent of respondents said their organizations regularly use AI in at least one business function, but only about one-third said their companies had begun to scale their AI programs [1]. Thirty-nine percent attributed any level of earnings before interest and taxes (EBIT) impact to AI, and most of those said less than 5 percent of their organization’s EBIT came from AI use [1].
Value sits with a small group. Boston Consulting Group (BCG) classifies 5 percent of companies as “future-built” for AI, with 35 percent scaling and 60 percent reporting minimal revenue and cost gains [2]. The leaders’ results are material. Compared with laggards, BCG reports 1.7 times the revenue growth, 3.6 times the three-year total shareholder return and 1.6 times the EBIT margin [2].
Chief executives see the same gap. In IBM’s 2025 study of 2,000 chief executives, respondents said only 25 percent of AI initiatives had delivered the expected return on investment over the last few years, and only 16 percent had scaled enterprise-wide. Sixty-four percent said the risk of falling behind drives investment in some technologies before they have a clear understanding of the value [3].
That last figure is the board’s problem in one line. Fear of falling behind is a reason to explore. It is not a reason to keep funding.
Why a portfolio view
A project view asks whether each initiative is on time and on budget. A portfolio view asks three harder questions: whether the collection of bets is balanced, whether capital is flowing to what works, and whether anything is being stopped. Venture investors manage uncertainty this way. Boards can do the same.
Three features distinguish a portfolio from a project list:
- Balance. A deliberate mix of near-term efficiency bets and longer-term growth bets, each sized on purpose.
- Stage gates. Funding released in tranches, each tied to evidence rather than elapsed time.
- Kill criteria. Conditions, agreed before funding, under which an initiative stops.
Balance matters because the largest returns rarely come from efficiency alone. McKinsey defines AI high performers as respondents who attribute 5 percent or more of EBIT, and significant value, to AI use. They are about 6 percent of its sample [1]. Most respondents set efficiency as an objective for AI. High performers were far more likely to also set growth and innovation objectives [1]. A portfolio made only of cost-cutting pilots will rarely produce that kind of result.
A stage-gate framework for AI
The framework below is what a board should ask management to adopt. Each gate releases the next tranche of funding only when the evidence is in.
| Stage | Evidence to pass | Typical kill signal |
|---|---|---|
| Explore | A named owner and a measured baseline | No owner after one quarter |
| Pilot | Results on live data against the baseline | No gain over the baseline |
| Scale | Workflow redesigned, run cost known | Unit cost exceeds unit value |
| Operate | Value tracked, risks monitored | Value decays or risk rises |
Two disciplines make the framework work.
First, kill criteria must be written before money is released, not debated after results arrive. Once a team is hired and a vendor is contracted, every result starts to look like a reason to continue. Gartner predicts that over 40 percent of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls [4]. Cancellation late in a project is expensive. Cancellation at a gate, against criteria agreed at the start, is cheap and frees capital for better bets.
Second, the scale gate should test the operating change, not the model. McKinsey found AI high performers nearly three times as likely as others to have fundamentally redesigned workflows: 55 percent against 20 percent [1]. They were also more likely to have defined processes for deciding how and when model outputs need human validation [1]. Both are things a board can ask to see.
The metrics that matter
Boards do not need dozens of AI metrics. They need a handful that connect spending to value and risk. Measurement is still uncommon. In McKinsey’s survey published in early 2025, fewer than one in five respondents said their organizations tracked key performance indicators (KPIs) for generative AI solutions, even though tracking well-defined KPIs had the strongest link to bottom-line impact of the adoption practices tested [5].
A workable board dashboard has five lines:
- Value realized, in currency, against the business case for each initiative at the scale stage or beyond.
- Unit economics, as the cost to serve one unit of value (a case, a claim, a design), including compute and people.
- Adoption, as the share of the target workflow that actually runs through the new process.
- Portfolio flow, as the number of initiatives entering, passing and stopped at each gate in the period.
- Risk posture, as incidents, open issues and the status of controls against the organization’s chosen framework.
The fourth line is the one most often missing. A portfolio in which nothing is ever stopped is not being governed. For the fifth line, frameworks already exist. The United States National Institute of Standards and Technology (NIST) AI Risk Management Framework organizes the work into four functions: govern, map, measure and manage [6]. Boards do not need to invent their own.
Where the board sits
Oversight structures vary. In McKinsey’s early-2025 survey, 28 percent of respondents whose organizations use AI said their chief executive was responsible for overseeing AI governance, and 17 percent said their board of directors oversaw it [5]. Either can work. Ambiguity does not. Senior ownership also shows up in results: AI high performers were three times as likely as others to strongly agree that senior leaders demonstrate ownership of and commitment to AI initiatives [1].
In board and strategy work, the most useful shift is often the simplest one: asking management to present AI as a single portfolio with gates, rather than as a list of separate projects. It changes the conversation from “how is this pilot going?” to “where should the next dollar go?”
Questions for the board
- Can management show us every AI initiative on one page, by stage, with its owner and its business case?
- What kill criteria were agreed for each initiative before funding, and which initiatives were stopped at a gate in the past year?
- What are the unit cost and unit value of each scaled initiative, and how are they trending?
- What share of the portfolio targets growth and innovation rather than efficiency alone?
- Which risk management framework do we use, and who reports to us on it?
References
- A. Singla, A. Sukharevsky, L. Yee, M. Chui, B. Hall and T. Balakrishnan, “The state of AI in 2025: Agents, innovation, and transformation,” McKinsey & Company, Nov. 2025.
- Boston Consulting Group, “The Widening AI Value Gap: Build for the Future 2025,” Sep. 2025.
- IBM Institute for Business Value, “5 mindshifts to supercharge business growth: 2025 CEO Study,” May 2025.
- Gartner, “Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027,” press release, Jun. 25, 2025.
- A. Singla, A. Sukharevsky, L. Yee, M. Chui and B. Hall, “The state of AI: How organizations are rewiring to capture value,” McKinsey & Company, Mar. 2025.
- National Institute of Standards and Technology, “Artificial Intelligence Risk Management Framework (AI RMF 1.0),” NIST AI 100-1, Jan. 2023.