Most companies precisely track spending on labor, raw materials and capital equipment, but far fewer can quantify how much they spend on making decisions — even though decisions shape nearly every business outcome from pricing and production to capital allocation and customer service. Decision costs often remain invisible because they are embedded in everyday activities: managers, engineers, analysts and frontline employees spend large amounts of time gathering information, weighing alternatives, coordinating and obtaining approvals. As organizations grow, meetings, review cycles and governance multiply, making decision‑making one of the largest but least visible operating costs.
For decades firms had little choice but to bear these costs because high‑quality decisions depended mainly on human judgment, analysis and coordination. Increasingly, however, companies are using artificial intelligence (AI) to optimize and speed up “decisioning” — the process of setting strategy and selecting next‑best actions across functions. Early pilots show promising results such as lower operating costs, faster product launches and higher customer satisfaction, but few companies have successfully replicated these benefits at scale, calling AI’s true economic value into question.
AI creates value beyond labor substitution
Debate often centers on productivity and cost cutting, including the controversial idea of replacing human workers with machines. That lens can overstate labor reduction potential while underplaying the value from better and faster decisions — for example, improved use of existing assets or seizing opportunities that would otherwise be missed.
AI should be understood not only as a labor‑saving technology but also as one that enables cheaper and better decisions. As more decisions flow through AI, firms can achieve better resource allocation, faster execution and more consistent outcomes. Decision throughput — the share of decisions that AI informs, accelerates or automates — can therefore serve as a leading indicator of future financial performance.
Returns, costs and a different infrastructure model
Determining whether AI can generate attractive returns on invested capital comes down to whether returns exceed costs. For AI both sides of that equation differ from prior technological revolutions. Unlike earlier general‑purpose technologies, AI does not require each company to build or own the underlying infrastructure. Hyperscalers invest hundreds of billions of dollars in data centers and provide AI capabilities as a shared service. In 2026 alone, capital expenditures by the four largest hyperscalers are expected to exceed $700 billion. Companies can thus access frontier AI through usage‑based pricing rather than large up‑front capital outlays.
For many tasks, AI‑mediated decisions have better unit economics than human coordination. Klarna, for example, compressed resolution time for customer service issues from 11 minutes to under two minutes using AI tools, which the company estimated would save $40 million in 2024. Research indicates AI can reduce decision‑making costs by over 90 percent; tasks that required roughly $40 in human labor in 2023 may now cost fractions of a cent.
If firms could replicate these gains broadly, AI could reach cost parity with human labor in about two years — a much faster adoption curve than prior technologies. Steam power took roughly 55 years to reach similar economic thresholds, the electric dynamo about 30 years, and the internet about seven years.
Why many companies still see little impact on earnings
If AI can cut decision costs so dramatically, why do many organizations report little or no measurable impact on earnings? The answer often lies in implementation. Most firms layer copilots, chatbots and dashboards onto existing processes, generating modest financial gains. By contrast, companies that redesign end‑to‑end workflows around AI often achieve roughly 20 percent EBITDA uplift, about three dollars of profit for every dollar invested, and payback periods of one to two years.
Cost reduction alone does not explain such improvements. SG&A (selling, general and administrative) costs typically run 5–12 percent at most companies; even large cuts in SG&A would not account for a 20 percent EBITDA increase. Instead, much of AI’s economic value comes from improving how organizations make decisions: enabling better decisions, speeding execution, reducing coordination losses and unlocking growth opportunities.
Where AI’s economic value actually comes from
Our analysis finds that AI’s value often arises from a sequence of effects:
- Faster decisions that shorten cycle times and accelerate action.
- Better decisions that improve outcomes and efficiency.
- Improved asset utilization through more frequent, better informed choices.
- Greater organizational flexibility to respond to changes.
- The ability to capture opportunities that slower competitors miss.
These effects compound over time. In some industrial companies, the largest value from AI has come from these less visible sources rather than head‑count reductions.
Metrics and management: what leaders should measure
Many organizations default to measures such as the number of AI tools licensed, models deployed or pilots launched. These metrics signal ambition but reveal little about potential returns — a company can have many tools without using them effectively. A better metric is decision throughput: the percentage of decisions that AI informs, accelerates or automates. Rising throughput shows that AI is embedded in core workflows and is influencing more decisions. For example, a manufacturer that moves from weekly to hourly production schedule adjustments can optimize schedules multiple times per day.
Throughput alone is not sufficient: operating costs, decision quality and business impact matter. CEOs should prioritize increasing both AI decision throughput and decision precision. Six priorities can help (covered in more detail in companion materials): governance and accountability, insourcing versus outsourcing choices, performance management, scaling practices, data and technology investment, and change management.
Example: Reckitt
Reckitt deployed RGMx, an AI‑enabled revenue growth management platform developed with McKinsey, to transform commercial decision‑making. Instead of spreadsheets and siloed debates, pricing, promotion and portfolio choices became data‑driven and collaborative. AI integrated market, consumer and category data, enabling leaders to model multiple what‑if scenarios, predict outcomes and evaluate tradeoffs before implementation.
Reckitt rolled out full versions in major markets and simplified versions elsewhere to build data discipline and consistent, fact‑based decision processes. It also used AI to create store‑level improvement programs by analyzing point‑of‑sale trends, inventory signals, promotional calendars, SKU features, planograms, store performance and execution feedback. The result: sales execution shifted from a downstream activity to an engine of commercial excellence.
Conclusion for leaders
The economics of AI are often framed as a labor story, but that narrative does not explain the magnitude of returns achieved by leading adopters. Those gains arise because AI enables organizations to make better decisions faster — improving asset use, increasing flexibility and allowing firms to capture opportunities that slower competitors miss. The advantage will not come from simply possessing AI, but from building an organization that consistently translates better decisions into better business outcomes.
Lari Hämäläinen is a senior partner in McKinsey’s Seattle office, and Steffen Fuchs is a senior partner in the Dallas office.
