Jesse Stockall is Chief Architect, Flexera.
Enterprise AI is entering a new phase. Organizations are moving beyond experimentation and embedding AI into products and business processes. As adoption accelerates, leaders are finding AI costs easier to understand than its value. New AI applications are appearing faster than traditional governance and budgeting approaches can adapt. The challenge is not simply tracking AI spend. It is understanding what that spend produces.
What has stood out to me as organizations navigate this shift is how quickly the conversation is moving from what AI costs to what that investment actually produces. Thirty-six percent of leaders report overspending on AI applications and 14% cite wasted AI spend, according to Flexera research. Yet spend visibility is only the foundation. To demonstrate ROI, leaders must connect AI consumption to the work performed, the quality of the result and the business outcome achieved.
That requires a progression from discovery to outcome attribution, to continuous optimization and control.
Step 1: Discover What You Have
During the AI experimentation phase, organizations adopted a flurry of new tools, ballooning their technology stacks. As a result, leaders no longer have one clear place to see it all.
Traditional management processes assumed more predictable software, infrastructure and ownership. AI disrupts that model as it appears across compute, applications, agents, workflows and data infrastructure. An agent may reason, invoke tools, call multiple models, access data and trigger downstream actions to complete a task. One business outcome can create costs across several systems.
Organizations should begin by inventorying every AI application, API, model, embedded AI feature and agent in use. When the data remains siloed, each team sees only one part of the investment and no one can assess the economics of the complete system.
With 2027 planning cycles upon us, this AI inventory provides a baseline for deciding which tools to consolidate, which experiments to continue and which use cases warrant additional investment.
Step 2: Connect Costs To Outcomes
As vendors move from predictable per-user pricing toward credits, tokens and consumption-based models, organizations need to understand where AI is being used. But attribution to a user or workload is not the end goal. The question is whether that consumption created value.
The most meaningful unit of measurement is often not a token, prompt or model call. It is a successful outcome. Depending on the use case, that may be the cost per customer issue resolved, deployment completed, document processed or agent task successfully performed. Measuring the complete outcome also exposes the hidden economics of failures, retries, latency, human intervention and rework. A cheaper model or workflow isn’t more efficient if it produces a lower-quality result or shifts additional work to people later in the process.
One lesson I’ve carried into AI cost management is to begin with the decision, not the dashboard. Before launching an AI initiative, leaders should define the decision the data will support, such as scaling a pilot, changing the underlying model or redirecting the investment. Then work backward to determine the cost, usage and business outcome data required.
Organizations should also establish success metrics that reflect the use case. For example, AI-assisted software development illustrates why activity metrics alone are insufficient, since generated code or accepted suggestions show activity, not necessarily value. Leaders need the full picture and should examine specification quality, cycle time, rework, software quality, change failure rate, security findings and customer impact.
High usage does not automatically signal value, and limited usage does not always indicate a poor investment. A smaller, mission-critical use case may deliver more benefit than a broadly available tool used occasionally.
Step 3: Measure Outcomes And Optimize Investment
KPMG data shows organizations with full visibility into AI operating costs are five times more likely to report established ROI, at 15% compared with 3%. However, assessing AI visibility isn’t a one-time activity.
AI has changed both the unit of cost and how quickly that cost accumulates. When consumption is measured in tokens and inference calls, periodic reviews are no longer enough. Organizations need a view of which users, agents and workloads are creating spend and whether that activity supports a defined business result.
Establish guardrails before costs or risks escalate. Financial controls may include budgets, consumption thresholds, approved models or defined spend levels. Operational controls should define what an agent may access, which actions it may perform, when human approval is required and what evidence must be retained.
Controls should match the value, autonomy and risk of the use case. For example, an internal experiment may need a tighter budget and a shorter review cycle. A customer-facing agent may justify higher consumption when it supports revenue or service quality, but it also requires stronger approval, testing, traceability and recovery controls. In software delivery, those controls may include architectural review, security checks, automated testing and an auditable record of agent actions.
As AI gets more complex, organizations need to align tools, models, agents and spend with the value and criticality of each use case. Mission-critical applications and agents may become material AI cost centers as their usage scales. Their controls should reflect not only spend, but the importance of the outcome and the consequences of failure. This allows organizations to support high-value adoption without treating every experiment, workflow and production system the same way.
Preparing For A Successful 2027 Starts Now
What I’ve learned from watching organizations navigate this transition is that AI cost management cannot be separated from AI value management. As organizations look to 2027, they need more than another dashboard.
Every AI application and agent should have a defined purpose, measurable outcomes, appropriate guardrails and clear ownership. Leaders need to understand what AI costs, what work it performs, the quality of that work and whether the result justifies continued investment. For me, that is the most important shift in how enterprises should think about AI ROI. The future of AI ROI is not simply knowing where the money went. It is understanding what the organization gained in return.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?