5 Levers for Disciplined AI Capital Allocation

The Workflow-First Mandate

Most AI projects fail because organizations treat them as "AI strategy" problems rather than workflow problems. Executives often purchase vendor solutions without understanding the specific shape of their own work, leading to a mismatch between the vendor's demo (which covers routine cases) and the company's reality (which is defined by complex exceptions). The unit of decision-making should not be the department or the role, but the specific operating loop—the inputs, the required judgment, the escalation paths, and the definition of success.

The Five Levers of Investment

Once a workflow is clearly defined, organizations must choose how to allocate capital across five levers:

1. Automate (Delete/Eat): Best for high-frequency, repeatable tasks with clear patterns and easy-to-verify outputs. If the work can be defined, it should be automated.
2. Build: Necessary when the workflow is unique, contains proprietary context, or requires specific risk thresholds that off-the-shelf tools cannot meet. This requires the executive to define exactly what "good" looks like to prevent teams from delivering "amazing" AI tools that fail to solve the actual business problem.
3. Buy: A choice between purchasing "primitives" (components that can be stacked into custom workflows) or "whole-workflow" solutions (like Harvey for legal). The risk here is the "shape mismatch"—if the vendor's workflow doesn't overlap 80-90% with your own, the cost of adjustment will exceed the value of the purchase.
4. Hire: Avoid chasing "purple unicorns" (domain experts who are also AI architects). Instead, identify the specific missing capability—such as evaluation design or workflow engineering—and hire for that gap. If a team member can be upleveled in six months to fill the need, training is superior to hiring in a volatile market.
5. Wait: The most counterintuitive but often necessary lever. Because change management resources are finite, organizations must prioritize workflows that provide the most leverage. Waiting on lower-priority workflows allows the firm to focus on high-impact transformations first.

The Executive's New Role

Executives must stop acting as passive buyers and start acting as the "honest third-party" who can evaluate whether an AI output is actually useful. This requires a deep understanding of the workflow's edges and failure modes. Without this, the organization is vulnerable to "AI-washing" and projects that look good in a demo but fail in production.