The State of AI : Global Survey 2025 | McKinsey

The latest McKinsey & Company survey, “The State of AI – Global Survey 2025,” paints a paradoxical picture. Ninety‑eight percent of organizations claim they are using AI somewhere in their operations, yet only about one third are scaling those initiatives across the enterprise. The gap between experimentation and enterprise‑grade deployment is wider than ever before, and the consequences for senior leaders are clear: without a systematic approach to governance, data quality, and KPI alignment, AI will remain a costly showpiece rather than a profit engine.

As an AI Business Strategist at AI2Work, I have spent years translating raw technology into strategic advantage. This article distills the survey’s findings through that lens, offering a playbook for C‑suite executives, Chief AI Officers, and transformation managers who must decide how to move from pilot mode to sustainable value creation.

• Adoption is high but shallow: 88% use AI in at least one function, yet only ~33% scale enterprise‑wide.

• Agents are the hottest trend but underperform: 62% experiment with agents; only 39% see profit impact.

• Profit uplift lags innovation claims: 64% say AI drives innovation, yet only 39% report measurable EBIT gains.

• Data and governance remain the single biggest blocker: 51% of firms have experienced backfiring due to poor data or unclear processes.

• High performers invest >20% of digital budgets in AI, redesign workflows, and secure executive sponsorship.

Bottom line:

AI is everywhere in 2025, but enterprises that treat it as a strategic enabler—rather than an isolated experiment—will outperform peers by 2–3 years on profitability and innovation.

In practice, the transition from pilot to production is akin to moving from a prototype to a scalable product. The survey’s 88% adoption figure masks a critical reality: most pilots are siloed, lack robust data pipelines, and have no clear ROI metrics. The key insight for leaders is that scaling AI demands the same rigor applied to any enterprise‑grade system—architecture, security, governance, and continuous improvement.

Operationally, this means:

• Establish an Enterprise AI Center of Excellence (CoE): A cross‑functional team that sets standards, approves pilots, and monitors performance against agreed KPIs.

• Create a unified data platform: Data must be discoverable, governed, and quality‑checked before it reaches any model. MLOps practices—automated validation, versioning, and lineage tracking—are non‑negotiable.

• Define success metrics early: EBIT uplift, cost savings, or customer NPS improvement should be quantified pre‑deployment and tracked post‑governance.

When these elements are in place, the transition from pilot to enterprise deployment can occur at a rate of 20–30% per year—far faster than the historical average of 5–10%.

The survey shows that 62% of firms are exploring or using AI agents—autonomous systems powered by GPT‑4o, Claude 3.5 Sonnet, or Gemini 1.5. Yet only 39% report any noticeable profit improvement. The disconnect stems from three intertwined factors:

• Unclear business objectives: Many teams launch agents to “improve efficiency” without tying the effort to a specific KPI.

• Data bottlenecks: Agents rely on real‑time data feeds; without clean, validated streams, their recommendations are often inaccurate or irrelevant.

• Change management gaps: End users need training and clear governance around agent interactions; otherwise, adoption stalls.

High performers mitigate these risks by forming

agent squads

: cross‑functional teams that own a single agent solution, align it with a specific business outcome (e.g., reducing procurement cycle time by 15%), and embed the agent within redesigned workflows. This structure accelerates ROI by ensuring that every interaction is purposeful and measurable.

The survey’s 64% figure for AI‑driven innovation contrasts sharply with the 39% who see profit impact. Innovation, while valuable, does not automatically translate into earnings unless it disrupts cost structures or opens new revenue streams.

Strategic leaders should therefore ask:

• What is the direct link between an AI initiative and financial performance?

• Does the innovation enable a new product line, reduce churn, or unlock a high‑margin market segment?

• Can we quantify the incremental EBIT attributable to the AI project within 12–18 months?

Case in point: A global retailer that deployed an AI‑powered demand forecasting agent reduced inventory carrying costs by 9% and increased sales velocity by 4%, translating into a $12 million EBITDA lift in FY2025.

Seventy‑seven percent of organizations remain stuck in pilot mode, with only 23% having scaled agents. This theatrical approach—high‑profile demos that generate buzz but little value—is a costly distraction for boards and investors.

Leaders can break this cycle by:

• Setting clear “go/no‑go” criteria: A pilot moves to production only if it meets pre‑defined ROI thresholds, data quality metrics, and governance checks.

• Investing in MLOps platforms: Tools that automate deployment, monitoring, and rollback reduce operational risk and accelerate time‑to‑value.

The survey identifies poor data or unclear processes as the cause of 51% of AI backfiring incidents. In a world dominated by GPT‑4o, Claude 3.5, and Gemini 1.5, model performance is no longer the bottleneck; upstream data integrity is.

Practical steps for leaders:

• Implement automated data validation pipelines: Real‑time checks flag anomalies before they reach the model.

• Adopt a data stewardship framework: Assign ownership of data domains to cross‑functional stewards who enforce quality standards.

• Establish a “data health scorecard”: Track metrics such as completeness, consistency, and timeliness across all AI feeds.

Organizations that invest in these practices see a 30–40% reduction in model drift incidents and a corresponding increase in confidence among end users.

The survey’s high‑performer profile offers a clear template for ambitious enterprises:

• Process redesign: AI is not an add‑on but a catalyst that forces rethinking of core workflows—e.g., automating contract review or predictive maintenance schedules.

• Executive sponsorship: A C‑suite champion who drives cross‑functional alignment and secures budgetary support.

These firms achieve EBIT uplift 2–3 years ahead of peers, as the combination of financial investment, strategic focus, and governance creates a virtuous cycle of learning and scaling.

The survey’s global scope reveals stark regional differences. North America and Western Europe lead in agent experimentation, while emerging markets lag due to data infrastructure gaps. Sectors vary too: financial services and manufacturing show higher adoption of AI for risk management and predictive maintenance, respectively.

Leaders should therefore:

• Create region‑specific roadmaps: Align data strategy with local regulatory requirements and digital maturity levels.

• Partner with local ecosystem players: Leverage regional cloud providers, data vendors, and consulting firms to accelerate deployment.

• Benchmark against peer performance: Use sectoral KPIs—such as cost per transaction in banking or yield improvement in agriculture—to set realistic targets.

The survey indicates a shift in vendor evaluation criteria. Executives now prioritize:

• End‑to‑end technical capability: From data ingestion to model monitoring.

• Integration skills: Seamless connectivity with existing ERP, CRM, and IoT platforms.

• MLOps maturity: Proven pipelines for continuous training, deployment, and rollback.

Choosing vendors that meet these criteria reduces integration friction and accelerates time‑to‑value. In 2025, the leading AI platforms—OpenAI’s GPT‑4o API, Anthropic’s Claude 3.5 Sonnet, Google Gemini 1.5, and Microsoft Azure OpenAI Service—offer varying strengths across these dimensions; selecting the right mix is critical for enterprise success.

High performers are increasingly forming

agent squads

, cross‑functional units that own a specific agent solution. This structure offers several advantages:

• Clear ownership and accountability: Each squad is responsible for the agent’s performance against defined KPIs.

• Rapid iteration cycles: Squads can experiment, deploy, and refine at a pace akin to software development teams.

• Alignment with business outcomes: Agents are embedded within redesigned workflows, ensuring that every interaction contributes directly to revenue or cost savings.

Looking ahead, the rise of autonomous agents may reshape traditional roles—prompting leaders to rethink talent strategies and governance models. The question is not whether AI will change the workplace, but how quickly organizations can adapt their structures to harness agent-driven value.

Only 39% of firms report profit impact from AI despite widespread experimentation. This gap signals a need for deeper understanding of:

• Incentive alignment: How do organizational pay structures and performance metrics influence AI adoption?

• Data pipeline robustness: What are the causal links between data quality, model accuracy, and business outcomes?

• Change management efficacy: Which approaches to training, communication, and governance yield higher adoption rates?

Longitudinal studies tracking pilot-to-scale transitions will be invaluable. For now, leaders can mitigate uncertainty by embedding rigorous evaluation frameworks—combining qualitative stakeholder feedback with quantitative performance metrics—to capture incremental value early.

• Create a Data‑First AI Architecture: Invest in unified data platforms and MLOps pipelines that enforce quality, lineage, and governance from the outset.

• Define KPI‑Driven Pilots: Every experiment must have a pre‑defined EBIT or cost‑reduction target, with clear thresholds for scaling.

• Secure Executive Sponsorship: Allocate >20% of digital budgets to AI and appoint a C‑suite champion who can drive cross‑functional alignment.

• Establish Agent Squads: Form dedicated teams that own specific agents, embed them in redesigned workflows, and track ROI at the squad level.

• Prioritize Vendor MLOps Capabilities: Choose partners with proven deployment pipelines, security compliance, and transparent IP ownership.

• Implement Continuous Monitoring: Use automated dashboards that link AI performance to board‑level KPIs, ensuring accountability and rapid course correction.

• Tailor Global Roadmaps: Align regional data strategies with local regulatory landscapes and partner ecosystems to overcome infrastructure gaps.

By following these steps, enterprises can move beyond the “AI theater” into a future where generative AI—powered by GPT‑4o, Claude 3.5 Sonnet, Gemini 1.5, and other cutting‑edge models—is not just a technology showcase but a strategic lever that drives profitability, operational efficiency, and competitive differentiation.

• Widespread AI use does not equal enterprise value; scaling requires governance, data quality, and KPI alignment.

• Agents are the hottest trend, yet most firms fail to capture profit impact due to unclear objectives and data bottlenecks.

• High performers invest heavily in digital budgets, redesign workflows, and secure executive sponsorship—enabling a 2–3 year advantage in EBIT uplift.

• Data quality is the single biggest blocker; automated validation and stewardship frameworks are essential.

• The rise of agent squads signals a shift toward organizational structures that embed AI directly into business outcomes.

For leaders who act on these insights now, 2025 will be the year when AI moves from experimentation to execution—and where enterprises unlock measurable, sustainable value.