AI Adoption Starts With the Constraint, Not the Tool

Start with the constraint, not the tool.

Most AI adoption conversations start one step too late. A tool is selected, licenses are bought, pilots multiply, and only then does the organization ask where the value should appear. By that point the work has already been framed around supply: what the technology can do, what the vendor can demonstrate, what a team can try quickly.

Portfolio leaders need the question framed from demand. Which constraint is limiting value in a recurring workflow? What would change if that constraint moved? Which intervention is the minimum effective response? The answer may be AI. It may also be data cleanup, decision rights, training, workflow redesign, deterministic automation, a vendor feature, model routing, or stopping a weak idea before it consumes capacity.

The current evidence base makes the case for that discipline. McKinsey reported broad AI use in 2025, with 88 percent of respondents using AI in at least one business function, while only 39 percent reported enterprise-level EBIT impact. RAND found that AI projects often fail when stakeholders misunderstand the problem to be solved or focus on new technology instead of real user problems. Gartner updated the warning in 2026, finding that at least half of GenAI projects had been abandoned after proof of concept by the end of the previous year because of poor data, weak controls, escalating costs, or unclear business value.

Those are arguments for sharper governance. The useful standard is simple: name the constraint, choose the least complicated intervention that can move it, assign ownership, and decide in advance what evidence would justify the next increment of funding or reliance.

The operating move

Make every AI request name the recurring workflow, the constraint, the minimum effective intervention, the owner, and the evidence that would justify scaling. That turns AI demand into a portfolio decision instead of a sequence of disconnected pilots.

Notes and sources

1. Alex Singla, Alexander Sukharevsky, Bryce Hall, Lareina Yee, Michael Chui, and Tara Balakrishnan, "The state of AI in 2025: Agents, innovation, and transformation," McKinsey & Company, November 5, 2025. Verified July 7, 2026. mckinsey.com
2. James Ryseff, "The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI," RAND Corporation, 2024. Verified July 7, 2026. rand.org
3. Arun Chandrasekaran, "Why 50% of GenAI Projects Fail - And How to Beat the Odds," Gartner, January 26, 2026. Verified July 7, 2026. gartner.com
4. Alexis Krivkovich and Lucia Rahilly, "AI is everywhere. The agentic organization is not yet," The McKinsey Podcast, April 2, 2026. Verified July 7, 2026. mckinsey.com
5. NIST, "AI Risk Management Framework Core," excerpt from AI RMF 1.0, 2023. Verified July 7, 2026. airc.nist.gov