Why Organizational Misalignment Drives Up Enterprise AI Costs
Artificial intelligence has moved rapidly from experimentation to enterprise adoption. Organizations are deploying AI across customer service, software development, finance, human resources, marketing, legal services, and operations with the expectation that it will improve productivity, accelerate decision-making, reduce costs, and strengthen competitive advantage.
The potential remains substantial. However, many executive teams are encountering an unexpected consequence: AI spending is increasing faster than the measurable business value it was intended to create.
This should not be entirely surprising. In September 2025, we cautioned that organizations were repeating the mistakes of earlier ERP and CRM implementations by launching generative AI initiatives without clearly defined measures of financial, operational, and organizational return. In February 2026, we examined how AI tools were spreading across corporate and personal devices faster than organizations could establish effective governance in AI Is the New SaaS: Shadow Growth, Policy Failure, and Why Governance Must Catch Up Fast.
Those conditions have now converged. Organizations scaled AI without consistently defining its expected return, while adoption expanded beyond the visibility of traditional governance. The resulting cost is not simply a technology-pricing problem. It is the financial consequence of adopting AI faster than the organization could align its structure, responsibilities, accountability, and decision-making around it.
Why Enterprise AI Costs Are Increasing
Previous generations of enterprise software followed a predictable financial model. Organizations acquired licenses, funded implementation projects, paid annual maintenance, and increased spending as users or capabilities were added. These investments could become expensive, but executive teams could model the continuing life-cycle cost.
Generative AI increasingly operates on a consumption basis. Model interactions, documents analyzed, automated workflows, API requests, and AI-agent activity consume computational resources that become operating expense. Individually, these transactions may appear insignificant. Across an enterprise, they accumulate rapidly.
Automated agents add another dimension. An agent can make multiple model calls, retrieve information from several sources, use external tools, evaluate its output, and repeat portions of the process without further human action. One business request may therefore produce numerous chargeable interactions that are largely invisible to the employee initiating the activity and the executive approving the investment.
The FinOps Foundation reports that 98 percent of its 2026 survey respondents now manage some form of AI spending, compared with 31 percent two years earlier. Reuters has also reported that some companies exhausted annual AI budgets within months as usage-based consumption grew faster than anticipated. (FinOps Foundation, Reuters)
Lower-cost models, prompt optimization, model routing, and spending limits can reduce consumption costs. They cannot determine whether the organization is using AI for the right work, duplicating capabilities across departments, or producing measurable business value.
Those are organizational questions.
How Broad AI Access Became Uncoordinated Consumption
Much of today’s AI cost growth did not begin with formal pilots. Organizations purchased enterprise access, distributed accounts broadly, activated AI capabilities embedded in existing software, and encouraged employees to experiment. The objective was to accelerate adoption and allow useful applications to emerge through experience.
Broad access was not inherently a mistake. Organizations needed employees to understand the technology and identify where it might improve their work. The problem was that access frequently expanded before leaders established which business problems AI should address, what information could be entered, how results would be validated, or how value would be measured.
Enterprise platforms also represented only part of the emerging environment. Employees continued using personal accounts, business units acquired specialized applications, and AI capabilities appeared through routine updates to productivity and enterprise software. Organizations accumulated sanctioned platforms, departmental tools, embedded capabilities, internally developed solutions, and personal usage without making a coordinated decision to create that operating model.
The rapid growth resembles the earlier expansion of cloud computing and SaaS. Business units acquired applications and provisioned resources independently, while costs accumulated across vendors, accounts, and departments. FinOps eventually emerged because traditional budgeting and technology governance were not designed for continuously variable consumption.
Cloud taught organizations to manage infrastructure consumption. SaaS taught them to manage decentralized software acquisition. AI will require them to manage intelligence consumption, along with the work, decisions, knowledge, and costs it affects.
The AI Bill Is a Symptom of Organizational Misalignment
Unexpected AI spending is often treated as a technology-pricing or governance problem. In many organizations, however, the cost is a visible symptom of a deeper structural issue.
Leadership encouraged adoption, Technology enabled access, business units pursued local opportunities, employees incorporated AI into their work, Finance monitored spending, Legal addressed risk, and Human Resources considered workforce implications. Each function acted within its traditional responsibilities. What was frequently missing was an organizational structure connecting those decisions at the enterprise level.
The resulting misalignment allowed local decisions to accumulate without a shared view of value. Business units could select tools without understanding similar investments elsewhere. Technology could report usage without determining whether operations had improved. Finance could identify rising costs without knowing which consumption supported essential work. Human Resources could evaluate workforce implications without complete visibility into the applications changing that work.
No single failure created the problem. The organization’s existing division of authority created gaps among otherwise reasonable decisions.
AI exposes those gaps because it crosses conventional functional boundaries. It is simultaneously a technology investment, operating capability, workforce intervention, data risk, source of business knowledge, and continuously variable expense. An organization designed to manage those responsibilities separately will struggle to manage AI as an integrated enterprise capability.
The unexpected bill is therefore not the problem itself. It is evidence that strategy, decision rights, accountability, operations, workforce planning, governance, and capital allocation are not sufficiently aligned.
AI Usage Does Not Equal Business Value
Decentralized decision-making is not inherently wrong. Business units and domain experts are best positioned to identify valuable applications for AI. Excessive centralization can delay experimentation, separate decisions from operational requirements, and create governance processes more concerned with approval than results.
However, decentralized authority requires enterprise alignment. Without it, individual functions can optimize local objectives while obscuring enterprise-wide cost, duplication, and return.
A content team can increase production while generating more material than the organization can effectively review or use. A software-development team can produce code faster while increasing defects, security concerns, or future maintenance requirements. A customer-service function can automate responses while increasing escalations or reducing service quality. A professional-services organization can save research time without converting that capacity into additional revenue, improved client outcomes, or lower delivery costs.
Existing business processes compound the problem. If AI is layered onto an inefficient process without eliminating unnecessary steps, redundant approvals, or poorly defined responsibilities, the organization may simply add a new expense to its operating model. Automating unnecessary work produces faster unnecessary work.
Active accounts, prompt volumes, and token consumption demonstrate adoption. They do not establish that work is being completed faster, at lower total cost, or with better results.
The technology may be operating as intended. The organization may not be.
Who Should Own Enterprise AI Spending and Governance?
Although rising AI costs may first become visible to the CFO, the underlying questions belong to the executive team. The CEO must determine whether AI advances strategy and enterprise performance. The COO must understand whether operations are improving. The CIO remains responsible for architecture, integration, security, and technology utilization. The CHRO must evaluate workforce capability and knowledge implications. Legal and risk leaders must address data protection, regulatory requirements, and acceptable use. The board expects assurance that material investments are producing defensible value.
Each executive owns part of the picture. No one necessarily owns the enterprise decision.
Assigning that responsibility does not require centralizing every AI choice under one executive or committee. The appropriate structure varies by organization, and domain experts must retain sufficient authority to act on what they observe close to the work. However, someone must hold the integrated view: who can authorize new capabilities, who is accountable when cost or risk crosses functional boundaries, and who answers for business value at the enterprise level.
Until that responsibility is assigned explicitly, it defaults to no one. Responsibility held by everyone in general and no one in particular is the condition that produced the AI bill in the first place.
Regaining Control Requires Organizational Alignment and Governance
Once AI use has spread across the enterprise, restricting accounts, consolidating tools, or reducing token consumption may slow spending. Those actions will not correct the conditions that produced it.
The organization must first establish clear lines of responsibility. Leaders need to determine who can authorize new AI capabilities, which decisions remain within business units, what requires enterprise review, how domain experts participate, and who can expand, redesign, consolidate, or discontinue an investment.
Once those responsibilities are clear, governance and controls can be applied effectively. Approved platforms, access requirements, data protections, consumption limits, performance measures, review periods, and escalation procedures should reinforce the organization’s decision structure rather than attempt to compensate for its absence.
The sequence matters. Organizational alignment defines who has authority, who is accountable, and how competing priorities are resolved. Governance establishes the rules under which decisions are made. Controls provide the visibility and enforcement needed to ensure those decisions are followed.
Without clear organizational responsibility, governance becomes a collection of policies that functions interpret independently. Without governance and controls, realignment remains an organizational chart with little effect on daily behavior. Regaining control requires both, applied in the correct order.
How Organizations Should Measure Generative AI ROI
Technical measurements remain necessary. Organizations should understand usage by platform, model, workflow, department, and accountable owner. Those measures must then be connected to financial, operational, and organizational ROI.
Financial measures may include P&L impact, cost avoidance, revenue improvement, or reduced external spending. Operational measures may include cycle-time compression, lower error rates, increased throughput, and improved service quality. Organizational measures may include clearer decision rights, effective adoption by domain experts, stronger workforce capability, and reduced dependence on informal workarounds.
The appropriate unit of analysis is not the token, user, or application. It is the business process and the outcome the investment was intended to produce.
This is the foundation of Brookey & Company’s Expert-Driven Loop (EDL). Rather than applying oversight after AI has produced an outcome, EDL starts with expertise. Domain experts define the business requirement, performance baseline, risks, and expected benefits at the outset, then remain engaged through implementation to validate results before an initiative is expanded.
The same discipline should extend to capital allocation. A zero-based review should not assume that every platform, integration, or workflow must continue because it has already been funded. Each investment should compete for resources based on strategic relevance, demonstrated performance, and credible future value.
The AI bill may have been unexpected, but the conditions that produced it were already visible. Undefined ROI made value difficult to prove. Uncoordinated adoption made consumption difficult to control. Organizational misalignment allowed both conditions to persist.
The technology did not create those conditions. It exposed them.
Frequently Asked Questions
How can an organization regain control of generative AI token spending?
Begin by identifying where AI authority, accountability, and spending responsibility currently reside. Enterprise platforms, departmental subscriptions, embedded AI features, APIs, automated agents, and known personal-account usage should be connected to an accountable business owner and a defined business purpose. Once responsibility is clear, the organization can apply budgets, usage monitoring, model-routing rules, access controls, and performance reviews where they will be effective.
How can companies control AI token costs without stopping productive use?
Avoid imposing uniform restrictions across every user and application. Determine which AI use cases support important business processes, which demonstrate measurable value, and which represent duplication or unmanaged experimentation. Preserve high-value uses, place uncertain applications under structured review, consolidate redundant tools, and restrict consumption that has no accountable owner or defensible business outcome.
What organizational structure is needed to manage enterprise AI spending?
The organization needs clear enterprise accountability combined with defined authority for business units and domain experts. One executive or governing body should own the integrated view of AI cost, risk, workforce impact, and business value, while operational leaders retain authority over approved uses within their areas. The appropriate structure varies by organization, but responsibility cannot remain fragmented among Technology, Finance, Operations, Human Resources, Legal, and individual business units.
Should AI governance or organizational realignment come first?
Organizational realignment should come first. Leaders must establish who has authority, who is accountable, which decisions can be made locally, and which require enterprise review. Governance can then define the rules, and controls can monitor and enforce them. When governance is imposed before responsibility is clarified, policies are often interpreted differently across functions and fail to change behavior.
How can companies rein in rogue generative AI and shadow AI use?
Policies and prohibitions alone are rarely sufficient. Organizations should understand why employees are using unapproved tools, provide approved alternatives that meet legitimate workflow needs, establish enforceable data and access requirements, and create reasonable transition periods. If approved systems do not support the work employees must perform, shadow AI is likely to continue through personal accounts and unmanaged devices.
Who should own generative AI token spending?
Finance can monitor expenditure and Technology can measure consumption, but neither function can independently determine whether AI is improving business performance. Enterprise accountability should sit with a leader or governing structure empowered to integrate strategy, operations, technology, workforce, risk, and capital allocation. Individual AI applications and workflows should also have named business owners responsible for both cost and results.
How should an organization determine whether AI token spending is creating value?
Evaluate consumption against the business process and outcome it supports, not simply the number of users, prompts, or tokens. Financial measures may include cost avoidance, reduced external spending, revenue improvement, or P&L impact. Operational measures may include cycle time, throughput, error rates, and service quality. Organizational measures may include clearer decisions, stronger workforce capability, and reduced reliance on manual workarounds.
How can organizations prevent AI sprawl from recurring?
Require every material AI capability to have a defined business purpose, accountable owner, approved data practices, expected cost, performance baseline, and scheduled review. Decisions to expand, redesign, consolidate, or discontinue an application should be based on evidence. This creates a continuing connection among organizational responsibility, governance, consumption controls, and measurable business value.
Brookey & Company helps boards and executive teams evaluate the organizational conditions affecting strategy execution, technology investment, workforce capability, and enterprise performance. Our OrgCore™ Assessment, AI Workforce Assessment, Expert-Driven Loop, and Zero-Based Budgeting services help organizations align responsibility, establish governance, measure outcomes, and direct resources toward demonstrable business value.