A year ago, most large enterprises were still experimenting with generative AI, deploying it alongside their core operations rather than within them. That phase has largely ended. By early 2026, 91% of businesses reported using AI in at least one core business function, up from 55% in 2023. What began as a series of pilot projects has evolved into a large-scale deployment, with measurable effects on costs and productivity.

The implications are becoming increasingly significant. Generative AI is not creating efficiency gains evenly across industries. Early adopters in sectors such as banking, retail, and healthcare are already reducing operating costs and improving productivity at a pace that competitors may struggle to match. As these advantages compound, the gap between leaders and laggards is becoming more structural. For executives evaluating AI investments today, the key question is no longer whether to adopt the technology, but which business functions have already reached economic viability and which industries are gaining the greatest competitive advantage.

Who wins in the AI spend surge

US enterprises spent $37 billion on generative AI in 2025, up from $11.5 billion the previous year. The global generative AI market is estimated at about $83 billion in 2026 and is projected to reach $988 billion by 2035.

McKinsey has identified 63 generative AI use cases across 16 business functions, with an estimated annual economic value of $2.6 trillion to $4.4 trillion at current model capabilities.

However, the average return on generative AI investment is about $3.70 per dollar spent, rising to 4.2x in financial services. Yet only 6% of organizations are "AI high performers," achieving at least 5% EBIT impact. The vast majority are seeing productivity gains, but not the deeper cost transformation that is beginning to separate leaders from the rest.

Figure 01

Generative AI: key cost and adoption metrics, 2026

Across enterprise functions, productivity gains are now measurable. The capture rate is still uneven.

Sources: Deloitte; Federal Reserve research; McKinsey Global Institute; Fullview.io AI Statistics, 2025-2026

Where AI is actually cutting costs

The cost impact of generative AI is not uniform across the enterprise. It varies by function, and the strongest gains are concentrated in a few areas that are further along in maturity. Three stand out in 2026.

Customer operations

Customer-facing AI is the most advanced use case. Enterprises that have moved beyond basic chatbots into full generative resolution workflows are cutting support costs by up to 20%. In sectors like healthcare and telecom, generative AI is now embedded in patient triage, billing queries, and network support, often overtaking traditional analytics workloads.

Software development

Code generation has become a $6.7 billion market and is projected to reach $25.7 billion by 2030. The productivity gains are significant at the individual level, with research showing about a 33% increase in output per hour of AI-assisted work. For software-driven firms, this directly reduces the cost of delivering new features.

Supply chain and demand planning

In operations, generative AI is quietly reducing cost of goods sold by 1 to 2% in areas like forecasting and inventory planning. For large retailers, this translates into millions in annual savings. In manufacturing, predictive maintenance is also reducing costly downtime across distributed production systems, improving efficiency in one of the most expensive parts of operations.

Figure 02

Annual economic value potential by sector (generative AI, fully implemented)

Retail and banking hold the largest individual sector opportunities in dollar terms

Why most companies are not seeing AI pay off yet

A clear risk is emerging across enterprise AI adoption. Around 80% of firms deploying generative AI report no measurable bottom-line impact, despite a 91% adoption rate. The issue is not the technology itself but how it is being integrated into operations.

Most organizations layer AI onto existing workflows rather than redesigning processes around it. Customer service teams may introduce generative AI assistants while retaining the same staffing structures and oversight layers. In such setups, efficiency gains remain at the individual level and rarely translate into reduced organizational costs. This explains why most firms remain outside the small group of high performers.

A second constraint appears in data foundations. Enterprise models are often trained on structured datasets drawn from Western retail systems, formal supply chains, and digitized consumer behavior. Across sub-Saharan Africa, South and Southeast Asia, and parts of Latin America, consumer activity is shaped by informal retail channels, agent networks, and cash-based transactions that are less visible in conventional datasets.

What leaders need to do next