生成式_AI_的鸿沟_2025_年商业_AI_的现状（英）_26页_913kb

The GenAl Divide Summary

Core Content

This report, The GenAl Divide, analyzes the state of Generative AI (GenAI) implementation in enterprises as of July 2025. It highlights a significant gap between high adoption and low transformation, revealing that most organizations are not realizing tangible business value from GenAI, despite substantial investment.

Main Findings

1. The GenAI Divide: High Adoption, Low Transformation

• Adoption is widespread, with over 80% of organizations having explored or piloted GenAI tools.
• Transformation is rare, with only 5% of integrated AI pilots delivering measurable P&L impact.
• Most tools are generic, like ChatGPT and Copilot, which enhance individual productivity but not enterprise-wide transformation.
• Custom enterprise solutions face significant challenges in reaching production, with only 5% of custom tools achieving full implementation.
• Seven out of nine sectors show little structural change, reinforcing the divide.

2. Industry-Level Disruption

• Only two industries (Technology and Media & Telecom) show clear signs of structural disruption.
• Other sectors remain on the wrong side of the divide, with minimal impact on workflows or business models.
• Sensitivity analysis confirms that Technology and Media & Telecom consistently rank highest, while Healthcare and Energy remain low.

3. The Pilot-to-Production Chasm

• Enterprise AI pilots rarely progress to full implementation, with only 5% of custom tools reaching production.
• Shadow AI economy exists, where employees use personal AI tools for work tasks, often achieving better ROI than official enterprise initiatives.
• Employee adoption is significantly higher than official adoption, with 90% of workers using LLMs regularly, compared to 40% of companies purchasing official subscriptions.

4. Investment Patterns

• Sales and marketing receive the largest share of GenAI budgets (~50%), despite back-office automation offering higher ROI.
• Functional allocation varies by industry: manufacturers and healthcare focus on operations, while tech and media prioritize marketing and content.
• Measurement challenges in back-office functions make it difficult to justify investment, leading to continued focus on high-visibility use cases.

5. Why Pilots Stall: The Learning Gap

• The learning gap is the primary barrier to scaling GenAI: systems that do not learn, adapt, or retain context fail to integrate into workflows.
• Users prefer tools that are flexible, responsive, and capable of learning, which are often lacking in enterprise systems.
• Agentic AI—systems that embed persistent memory and iterative learning—offers a solution to this gap, enabling systems to evolve and improve over time.

6. The Best Builders and Buyers Succeed

• Successful builders focus on narrow, high-value use cases, deeply integrate into workflows, and prioritize learning over broad feature sets.
• Successful buyers demand process-specific customization, evaluate tools based on business outcomes, and prefer systems that integrate with existing tools.
• Trust and social proof are critical in enterprise adoption, with peer recommendations and established relationships outweighing product features.

Key Takeaways

• The GenAI Divide is defined by the gap between adoption and transformation.
• Learning capability is the core differentiator between successful and stalled AI implementations.
• Shadow AI demonstrates the potential of GenAI when users have access to flexible and responsive tools.
• Investment bias toward visible functions like sales and marketing keeps organizations from focusing on high-ROI back-office automation.
• Agentic AI and workflow integration are essential for crossing the divide.

Conclusion

Bridging the GenAI Divide requires a shift from static, generic tools to adaptive, learning-capable systems that integrate seamlessly with existing workflows. Organizations that recognize the value of shadow AI and invest in tailored, intelligent solutions are beginning to see real impact, while those relying on broad, uncustomized AI implementations remain stuck in the pilot phase. The future of enterprise AI lies in systems that evolve with the user, not just generate content.