• Biggest wave of costs will only begin with widespread AI adoption
• AI agents make costs harder to predict
• “Financial guardrails” set economic parameters for the use of AI 

Companies are facing a paradigm shift in their use of artificial intelligence. As AI is rolled out more broadly, its benefits increase, but the underlying cost structure is also changing: More and more AI services are being billed based on usage rather than through subscriptions. As a result, finance leaders are increasingly focusing on whether the widespread use of AI actually pays off. Ironically, companies face the biggest wave of costs precisely when moving from pilot projects to production. This is the conclusion of the analysis “AI Cost Management – How Companies Can Scale AI Economically” by management consultancy Horváth.

AI promises benefits through better decision-making, greater efficiency, and new business models. However, the more intensively companies use artificial intelligence (AI), the higher the costs can become. This makes resource consumption an economic challenge. A paradox is emerging: The technology intended to increase efficiency and reduce costs is itself becoming a cost driver. When AI applications are rolled out broadly, costs can temporarily rise even faster than productivity or revenue. The main drivers are a growing number of users and an expanding range of AI applications in production. 

“The real economic test for AI still lies ahead for many companies. Only when AI is used daily by thousands of employees and firmly integrated into business processes will it become clear whether it actually pays off. Companies therefore need to plan and manage AI costs just as professionally as their cloud or IT costs,” says Dr. Stefan Zeibig, Partner in the Competence Center Industrial Goods & High Tech at Horváth.

AI Agents Make Costs Harder to Predict

Agentic AI adds another cost dimension. AI agents do not simply process individual requests. They can independently plan complex tasks, call different models and programs, repeat work steps, or activate additional agents. As a result, a single task can trigger an entire chain of chargeable actions. “What matters is no longer simply how much a single AI request costs. Companies need to know the cost of an entirely AI-supported workflow and the economic value it delivers,” adds Horváth expert Dr. Zeibig.

More AI Does Not Automatically Mean More Value

Yet this kind of economic oversight is still lacking in many organizations. Horváth therefore recommends AI performance management that links costs and resource consumption with quality, risks, and economic value. Companies should manage their AI applications like an investment portfolio: by using standardized business cases, ongoing forecasts, and clear criteria for determining when a pilot project should move into production, be scaled further, further developed, or discontinued.

Companies should also make AI costs transparent in management reporting and establish robust cost-benefit metrics, for example per application, workflow, agent, or customer. Prof. Dr. Ronald Gleich from the Centre for Performance Management & Controlling at the Frankfurt School of Finance & Management also emphasizes: “Companies that want to scale AI economically must avoid becoming dependent on individual providers. Doing so also ties them to those providers’ pricing and business models. Alternative AI models and open-source solutions therefore need to be part of the strategy from the outset. Cost transparency and technological sovereignty must be considered together.”

AI Becomes a CFO Responsibility

As AI is deployed more broadly, responsibilities within companies are also shifting: AI is increasingly becoming a CFO responsibility. While CIOs, CTOs, and data leaders ensure technical performance and business functions are responsible for realizing the benefits, the CFO brings these perspectives together from a business perspective.

In addition to Return-on-Investment, total costs over an application’s life cycle, cash flow effects, risks, and strategic value are critical. “Financial guardrails make the use of AI economically manageable. They help companies continuously balance costs and value and take corrective action in good time when the two begin to diverge,” explains Dr. Zeibig of Horváth.

About the Analysis

In the white paper “AI Cost Management – How Companies Can Scale AI Economically,” Horváth presents eight theses on the future economic scaling of Generative AI and Agentic AI. The analysis focuses on the development of AI costs during the transition from pilot projects to scaled production, new usage-based cost structures, the economic implications of Agentic AI, and the future role of CFOs and controlling. The analysis was authored by Prof. Dr. Ronald Gleich, Dr. Johannes Isensee, Dr. Uwe Michel, Dr. Peter Schentler, Dr. Christian Schnöbel, Prof. Dr. Mischa Seiter, Dr. Stefan Zeibig, and Phil Biehler. Based on their findings, the authors derive AI Cost & Performance Management as a new management discipline. Its aim is to integrate AI costs, usage, and economic value into a common management framework, thereby laying the foundation for scaling AI economically.