AI Tool Sprawl Demands Accountability Before The Bills Arrive

It has never been easier for the global workforce to use AI tools to automate tasks and build applications with little or no IT expertise. For CIOs, the question is no longer simply which tools to approve, but how to enable adoption while maintaining control, accountability and a clear view of both business outcomes and costs.

Business owners remain accountable for outcomes, while FinOps teams and enterprise architects connect adoption and architect decisions with costs, risks and strategic priorities, including whether a tool fulfils a capability gap, duplicates functionality or supports reuse.

José Freitas is Lead Enterprise Architect at the International Air Transport Association (IATA), a global trade association representing 380 member airlines. He previously worked at Deloitte on M&A, manufacturing and enterprise governance, and at IBM on cyber resiliency, also holding chief-architect roles for banking institutions in EMEA.

"Controlling technology costs, it’s not only about what we pay. It’s what we choose, how we use it and what value delivers," said Freitas. AI spending exposes an accountability challenge: organizations need to choose what tools need to be deployed, who is responsible for its costs and risks, and how value is measured. The cloud boom of the 2010s offers a useful lesson: when adoption without clear accountability allowed costs to grow faster than the value delivered.

• Cloud lessons, applied to AI: Establish accountability before spending outpaces value. Freitas noted that cloud deployment often moved faster than cost visibility and governance; AI risks repeating that pattern. Business owners, FinOps practitioners and architects must link model choices and consumption to measurable business outcomes and confuse usage for productivity. In document processing, for example, architects and engineers validate model quality, while FinOps forecasts the cost per successfully processed document, including retries. Use lower-cost models for routine tasks, reserve premium models for tasks that require them, and enforce consumption limits.

• Someone owns the bill, and the outcome: Business owners remain accountable for value, supported by a cross-functional team that tracks spending, duplication and underuse. Freitas recommends starting where visibility is weakest and reviewing usage and subscription costs before renewal. Benefits must go beyond activity: time released through automation should translate into service quality or improved customer experience. An architecture review board can resolve trade-offs within delegated authority, escalating unresolved decisions to the CIO and relevant budget or risk owners.

• Cheap alone, costly together: A low-cost license can hide a larger company-wide bill when teams buy the same tool separately. Freitas said coordinated purchasing and shared enterprise agreements can improve pricing, support and control over accounts and data. Enterprise architects provide cross-organizational visibility to identify duplication and reuse opportunities, working with business owners and procurement to consolidate where it delivers value. Consolidation should not become an end in itself: deliberate redundancy may be justified where business continuity and service criticality require it.

In the approach Freitas advocates, AI governance sits within the broader enterprise architecture framework, connecting experimentation and production adoption to business capabilities, strategic priorities and risk.

Governance should distinguish experimentation from production. Teams may evaluate tools not yet vetted for production under minimal, defined controls, using synthetic data or non-sensitive production data explicitly cleared. Each proof of concept needs a purpose, budget and review point: stop, extend or seek production approval based on value, cost, risk and operational readiness. Business products need equivalent checkpoints to decide whether to scale, adjust or retire them.

Approved tools, standards and reusable architecture patterns should reduce repeated evaluation, with additional review focused on exceptions and material changes. Where FinOps remains retrospective, cost accountability needs to move into selection, design and adoption.

Leadership must provide clear decision rights and an escalation path when cost, value and risk cannot be reconciled. Good architecture cannot guarantee successful delivery, but it should make the trade-offs and required capabilities visible before commitments are made.

"Cost visibility and control need to be orchestrated into how we adopt and use technology, not added after the money is spent," Freitas concluded.

The views expressed in this article are José Freitas’s personal opinions and do not necessarily reflect the views or policies of the International Air Transport Association (IATA).