AI is everywhere in boardroom decks, but not yet in the bottom line. A new global survey of 4,454 chief executives by PwC finds most companies have not turned artificial intelligence into measurable gains, even as leaders insist adopting it is essential to stay competitive.
The headline tension is stark: 56% of CEOs report no revenue growth or cost reduction from AI, while a majority still frame it as table stakes for relevance in their markets.
Survey Signals ROI Stagnation Across AI Investments
Across 95 countries and industries, only 30% of CEOs say AI has increased revenue in recent months. Just 26% cite cost savings. A mere 12% achieved both outcomes at once, underscoring how gains often come with offsetting expenses, from integration and data work to compliance and compute.
The needle is moving, but slowly. An October analysis by Boston Consulting Group found only about 5% of companies could point to clear financial benefits from generative AI pilots, suggesting the leap from experimentation to scaled value remains elusive.
PwC’s numbers also reveal a trade-off pattern: firms that saw revenue bumps frequently incurred higher costs, and those that achieved savings often struggled to lift sales. In other words, AI wins tend to be narrow and local rather than enterprise-wide.
Why AI Feels Unavoidable Despite Limited ROI Today
Despite muted returns, executives say the strategic risk of sitting out the AI wave is higher than the near-term financial risk of investing. Competitors are embedding AI into search, advertising, productivity suites, and CRM—think Microsoft’s Copilot layers and Salesforce’s Einstein features—setting new customer expectations for speed and personalization.
The result is a classic technology adoption trap: AI becomes part of the minimum service level. If everyone else offers smarter support agents or AI-augmented workflows, sticking with manual processes looks archaic, even if the P&L case is not yet airtight.
Where the Money Is and Is Not in Enterprise AI
CEOs say their organizations are experimenting on multiple fronts. According to PwC, about 22% are using AI to stimulate demand, 20% have integrated it into customer service, and 15% deploy it for “direction setting” and planning. The mix reflects a hunt for top-line growth alongside operational efficiency.
Why is ROI hard to capture? Three issues recur in executive interviews and analyst notes: data readiness, change management, and unit economics. Many firms lack clean, labeled, and governed data. Pilots often automate steps without redesigning the process end to end. And inference costs—especially for large models—scale faster than expected without tight prompts, caching, and workload routing.
Quality and risk also matter. Hallucinations, privacy constraints, and sector-specific compliance (from financial services to healthcare) add friction, forcing human-in-the-loop safeguards that dilute headline productivity gains.
Skills Gap Widens as Plans Lag for Generative AI
PwC reports 69% of CEOs expect most of their workforce will need new skills to use generative AI, yet fewer than half have a clear plan to deliver that reskilling. That mismatch slows impact: without retooled roles, upgraded processes, and incentives aligned to adoption, tools sit underutilized.
Early bright spots tend to be narrow and measurable: service-agent assist, marketing content drafting, coding copilots, and document summarization. GitHub’s research, for instance, has shown material time savings on well-bounded programming tasks, but translating such wins across complex enterprises requires robust governance and training.
What CEOs Can Do Next to Turn AI Into Measurable Gains
Anchor AI efforts to P&L owners. Tie deployments to intent-to-buy, conversion, churn, or cycle-time metrics and report net impact after compute and oversight costs. Avoid “pilot purgatory” by selecting use cases with high transaction volumes and clear baselines.
Engineer for cost from day one. Use smaller or domain-specific models where possible, implement guardrails and retrieval to cut hallucinations, and apply FinOps discipline to track token usage, latency, and caching. Treat model choice and prompt design as architecture, not art.
Invest in people. Stand up internal academies, pair power users with process owners, and reward adoption milestones. The firms capturing value are redesigning workflows—moving beyond “bolt-on AI” toward roles and processes built around it.
The message from the C-suite is pragmatic: AI may not be a profit engine yet, but it has already become a relevance engine. The winners will be those who turn that strategic necessity into disciplined, measurable advantage.