Uncovering the hidden cost of unbridled AI use

As agentic AI adoption grows, CFOs, CIOs and board members face hidden costs that are difficult to measure and control, warns Hewlett Packard Enterprise South Africa's President Ntuli. Each prompt or automated workflow can trigger a cascade of interactions, API calls, data retrievals and security checks, potentially leading to a new form of shadow IT.

While traditional AI is billed by the hour or seat, agentic AI operates on a consumption-based model using tokens, making costs vary based on usage. Small phrase changes can significantly impact token consumption and costs. When AI agents operate continuously at production scale, costs can skyrocket, creating an invisible meter that grows with increased use.

Agentic AI spreads across business functions, leading to unmonitored spending, governance challenges and operational risks. To manage these costs, businesses must gain visibility into agent usage, model selection, resource consumption, data access and delivered value. Unexpected expenses arise from data-readiness work, integration, maintenance and ongoing governance, which are often overlooked in initial cost estimations.