AI adoption starts with the constraint, not the tool

What the paper develops

Most AI requests start one step too late. A team sees a tool, model, or agent and asks where it could fit. Pilots begin. Usage grows. Only then do leaders ask where value should appear.

Start with the recurring work instead. Name what limits that work today. Then compare the changes that could remove the limit. AI may be the best choice. It may also be data cleanup, clearer authority, training, a workflow change, simple automation, a vendor feature, or a decision to wait.

McKinsey reported broad AI use in 2025, while far fewer respondents reported an effect on earnings across the enterprise. RAND and Gartner describe a related risk: AI efforts can fail after teams choose the wrong problem, use weak data, miss controls, or cannot show clear value. These sources use different methods. Together, they show why a working demo is not enough.

A constraint is whatever currently limits value in recurring work. It might be a slow handoff, missing data, unclear approval, poor exception handling, or scarce review time. A useful request names the work, the baseline, the owner, and the evidence that would show improvement.

Compare workflow, data, people, fixed-rule automation, vendor features, AI, and waiting. Choose the least complex option that can move the constraint with enough reliability, safety, speed, and value. Fund a missing prerequisite before adding technology on top of it.

Move from constraint intake to intervention choice, proof design, a rollout decision, and normal value review. Each gate answers one question and leaves a short record. The pilot earns more funding only when real workflow evidence supports it.

Before the next steering meeting, replace one tool name with the recurring work it is meant to change. Write the constraint, baseline, owner, options, and one result that would justify another funding step. Start with the constraint. Then choose the tool.