Ken Stillwell explains that generative AI requires reprocessing tasks from scratch, unlike CPU or rules-based systems that handle billions of repetitive transactions cheaply. Using AI where a simple rule or structured workflow suffices, such as replying to an email with a known recipient, drives up computational costs and risk without delivering better outcomes. AI should be reserved for unstructured creativity or synthesis tasks.
Stillwell observes that many AI pilots spin out of control because they are run by engineering mindsets focused on proving technical possibility rather than business value. To prevent pilot sprawl and wasted spend, leaders must approach internal AI initiatives like product managers—focusing tightly on specific business problems, user outcomes, and scalability rather than chasing novel or technically impressive prototypes.
When cloud computing emerged, individual employees expensed AWS instances on corporate credit cards, resulting in unmanageable decentralization that companies later had to rein in. Stillwell explains that Pega preempted this dynamic in AI by centralizing all AI vendor requests through a cross-functional steering committee involving engineering and IT leadership, ensuring unified governance, architecture, and cost controls right from the start.
Calculating exact financial ROI for AI projects early on is often elusive and counterproductive. Stillwell argues leaders should instead evaluate initiatives using clear intermediate operational proxies—such as shaving two days off the financial quarter-end close or accelerating collections by a week. Focusing on directional business outcomes and clear problem definition prevents teams from stalling over complex math or pursuing AI without a concrete purpose.
In the rush to look innovative, organizations penalize teams when AI experiments do not pan out, discouraging employees from admitting failure. Stillwell emphasizes that experimentation inherently involves being wrong. Boards and senior executives must remove emotional attachment and actively reward leaders who shut down unviable AI projects and reallocate resources toward proven business value rather than maintaining zombie initiatives out of competitive anxiety.
Organizations currently obsess over discrete AI ROI calculations because enterprise security, governance, and risk profiles remain immature. Stillwell notes that just like email or word processors, AI will eventually become native enterprise infrastructure where adoption is taken for granted. However, that shift requires developing years of rigorous security and governance standards so enterprises can trust automated systems to manage sensitive operations reliably.