The State of AI in 2025

AI Adoption Status

Core finding: AI has reached near-universal adoption, but scaling remains the critical challenge.

The 2025 survey marks a watershed moment in enterprise AI: 88% of organizations now use AI in at least one business function, up from 78% in 2024 and 55% just two years ago. The adoption curve has effectively plateaued—we've moved from "will companies adopt AI?" to "how effectively are they using it?"

However, the more revealing metric is what happens after initial adoption. Only 38% of organizations have successfully scaled AI beyond pilot projects. The majority remain in what McKinsey calls "experimentation mode"—running proofs of concept that never reach production at scale.

The gap between adoption and scaling represents the core challenge facing most organizations. Having AI tools is no longer a differentiator—the competitive advantage lies in deploying them effectively across the enterprise.

AI Agents Emergence

Core finding: AI agents are the next frontier, with 62% experimenting but only 23% scaling.

Perhaps the most significant development in the 2025 landscape is the emergence of AI agents—systems that can autonomously perform multi-step tasks, make decisions, and take actions with minimal human oversight. Unlike traditional AI that responds to single queries, agents can plan, execute, and adapt.

The survey reveals a familiar pattern: 62% of organizations are experimenting with AI agents, but only 23% have scaled them into production. This mirrors the broader AI adoption curve from 2-3 years ago, suggesting we're at the early stages of another major technology wave.

What organizations are using agents for:

• Customer service automation — handling complex inquiries end-to-end
• Software development — code generation, testing, and deployment workflows
• Data analysis pipelines — automated insight generation and reporting
• Supply chain optimization — real-time decision-making and adjustment

Early movers report that agents deliver 2-3× the productivity gains of traditional gen AI tools, but require significantly more investment in workflow redesign and governance frameworks.

Industry & Function Leaders

Core finding: Technology leads in AI agents; marketing leads in gen AI; healthcare shows fastest growth.

Not all industries are advancing at the same pace. The survey identifies clear leaders and laggards across both traditional AI and the newer agent-based approaches.

Industries leading in AI agent adoption:

1. Technology & Telecom — 34% have scaled agents
2. Financial Services — 28% have scaled agents
3. Healthcare & Pharma — 26% have scaled agents

Business functions with highest gen AI adoption:

1. Marketing & Sales — 42% regular use
2. Product Development — 38% regular use
3. Service Operations — 35% regular use
4. IT & Engineering — 33% regular use

Healthcare shows the fastest year-over-year growth in AI adoption (up 23 percentage points), driven by clinical decision support, drug discovery acceleration, and administrative automation. However, regulatory constraints mean scaling takes longer in this sector.

The laggards—construction, agriculture, and public sector—cite data infrastructure gaps and workforce readiness as primary barriers, not lack of interest.

Multi-Function Expansion

Core finding: High performers deploy AI across 5+ functions; average organizations use it in 2-3.

A key differentiator between AI leaders and followers is breadth of deployment. Organizations seeing meaningful business impact don't just use AI in one or two areas—they deploy it systematically across multiple functions.

Distribution of AI usage across functions:

High performers are 2.4× more likely to use AI in 5+ business functions compared to average organizations. This multi-function deployment creates compounding benefits:

• Data network effects — More usage generates more training data
• Workflow integration — AI tools communicate across functions
• Organizational learning — Teams share best practices and governance approaches
• Cost efficiency — Infrastructure investments amortize across more use cases

The implication is clear: AI strategy should focus on breadth of adoption, not just depth in any single function.

Company Size Gap

Core finding: Large enterprises ($5B+ revenue) are 2× more likely to have scaled AI successfully.

One of the most striking findings is the correlation between company size and AI scaling success. Organizations with $5 billion+ in annual revenue are approximately twice as likely to have moved AI beyond pilots compared to smaller companies.

AI scaling success by company size:

The advantages of scale include:

• Larger data assets for training and fine-tuning models
• More specialized talent — data scientists, ML engineers, AI product managers
• Greater capital for infrastructure and experimentation
• Established governance frameworks that adapt to AI oversight

However, the survey notes that smaller companies can compete through focused use cases, cloud-based AI services, and industry-specific solutions. The key is choosing battles wisely rather than trying to match enterprise-scale deployments.

Innovation & Value Impact

Core finding: 64% report innovation improvements, but only 39% see measurable EBIT impact.

The gap between perceived benefits and financial results is one of the report's most important insights. While organizations broadly report that AI improves their innovation capabilities, translating this into bottom-line results remains elusive.

Reported benefits from AI adoption:

The 25-point gap between innovation improvement (64%) and EBIT impact (39%) reflects several factors:

• Long lag times between AI deployment and financial results
• Difficulty attributing outcomes to AI vs. other factors
• Productivity gains absorbed by other inefficiencies
• Investment costs offsetting near-term savings

Organizations that do achieve EBIT impact typically have clearer measurement frameworks and longer deployment timelines (2+ years of sustained investment).

High Performers Profile

Core finding: Only 6% of organizations are "high performers" seeing 5%+ EBIT impact—and they operate differently.

McKinsey identifies a small cohort of high performers—the 6% of organizations reporting 5% or more EBIT impact from AI. These companies aren't just doing more AI; they're approaching it fundamentally differently.

What distinguishes high performers:

High performers are 3.6× more likely to pursue transformative change rather than incremental improvement. They view AI not as a tool for modest efficiency gains, but as a catalyst for reimagining how work gets done.

Perhaps most importantly, high performers invest as much in organizational change as in technology. They understand that AI success is as much a people problem as a technical one.

Success Practices

Core finding: Workflow redesign and change management separate winners from the rest.

The survey identifies specific practices that correlate strongly with AI success. These aren't about choosing the right model or technology—they're about how organizations implement and operationalize AI.

Top practices of successful AI organizations:

1. Workflow redesign before deployment — Don't just add AI to existing processes; reimagine how work should flow with AI capabilities. High performers are 2.8× more likely to redesign workflows comprehensively.

2. Comprehensive employee training — Beyond tool training, this includes AI literacy, prompt engineering, and understanding AI limitations. Leaders invest 3× more in training per employee.

3. Clear governance frameworks — Centralized oversight of AI initiatives, with defined policies for data usage, model selection, and risk management. 71% of high performers have this vs. 38% average.

4. Measurement and iteration — Establishing baseline metrics before deployment and continuously measuring impact. High performers review AI performance monthly, not quarterly.

5. Executive sponsorship — C-suite involvement in AI strategy and regular progress reviews. In high-performing organizations, 89% of executives are actively engaged vs. 52% average.

Workforce Impact

Core finding: 32% expect workforce decreases, but the picture is more nuanced than headlines suggest.

The workforce implications of AI generate significant attention, and the survey provides concrete data on how organizations view the impact:

Expected workforce changes in the next 3 years:

The 32% expecting decreases versus 25% expecting increases reflects productivity gains enabling the same output with fewer people. However, the largest group (43%) expects no significant change, suggesting AI will transform roles more than eliminate them.

Functions most likely to see workforce reduction:

1. Customer service & support
2. Data entry & processing
3. Basic financial analysis
4. Document review & compliance

Functions most likely to see workforce growth:

1. AI/ML engineering & development
2. Data science & analytics
3. AI ethics & governance
4. Human-AI collaboration design

Organizations planning workforce reductions report they will primarily occur through attrition and redeployment rather than layoffs.

Risks & Mitigation

Core finding: 51% have experienced negative consequences—inaccuracy is the top concern.

The rush to adopt AI has come with consequences. Over half of organizations (51%) report experiencing negative outcomes from AI implementation, a figure that should temper enthusiasm with appropriate caution.

Negative consequences experienced:

Inaccuracy remains the dominant concern at 44%, reflecting the fundamental challenge of AI systems that generate plausible-sounding but incorrect outputs. This is particularly problematic in high-stakes domains like healthcare, finance, and legal.

Mitigation strategies employed by leading organizations:

1. Human-in-the-loop requirements for critical decisions
2. Output verification protocols before customer-facing deployment
3. Regular model auditing and bias testing
4. Clear escalation paths when AI confidence is low
5. Incident response plans for AI failures

The organizations with fewest negative consequences aren't those avoiding AI—they're those with the most robust governance frameworks. Risk management and AI adoption can, and must, advance together.

Conclusion

McKinsey's 2025 State of AI survey reveals an industry at an inflection point. The question is no longer whether to adopt AI, but how to scale it effectively while managing risks.

The 6% of high performers provide a roadmap: pursue transformative change rather than incremental improvement, invest as much in organizational change as technology, redesign workflows comprehensively, and maintain robust governance frameworks.

For the majority of organizations still in experimentation mode, the path forward requires moving beyond pilots to production deployment—a transition that demands sustained investment, executive commitment, and willingness to reimagine how work gets done.

The emergence of AI agents adds another dimension to this challenge. With 62% experimenting but only 23% scaling, we're witnessing the early stages of the next major technology wave. Organizations that master this transition will likely pull further ahead of those that don't.

This analysis is based on McKinsey & Company's "The State of AI in 2025" report, surveying 1,993 respondents across 105 nations and 17 industries. The full report is available for download above.