Why are companies struggling to profit from AI despite its promise? Generative AI is proving powerful but inconsistent, requiring human guidance to deliver real business value.
 |
| The gap between AI hype and business results highlights the importance of human oversight, specialized applications, and realistic expectations for enterprise adoption. Image: CH |
New York, United States — December 17, 2025:
The AI revolution, long heralded as imminent and transformative, has yet to deliver widespread returns for businesses, revealing the challenges of integrating generative AI into practical operations. While companies have rushed to adopt tools like ChatGPT, many are finding that these models require careful guidance, human oversight, and tailored applications to be truly effective.
CellarTracker’s attempt to build an AI sommelier illustrates these difficulties. The model initially provided overly polite wine recommendations, necessitating weeks of experimentation to coax honest appraisals from the system. Similarly, Cando Rail discovered that AI struggled to consistently summarize complex safety regulations, sometimes inventing or misinterpreting rules. These cases underscore that even advanced AI can be inconsistent when handling nuanced, domain-specific tasks.
Survey data confirms the trend. Forrester Research found that only 15% of executives reported improved profit margins from AI over the past year, while BCG noted just 5% of firms saw widespread value. While executives remain optimistic about AI’s long-term potential, many are adjusting expectations for speed and impact. Forrester predicts that 25% of planned AI spending will be delayed by 2026 as companies reassess priorities.
Customer service applications further highlight AI’s limits. Swedish fintech Klarna initially deployed AI to replace hundreds of agents but quickly found that humans remained essential for complex interactions. Verizon and other firms have similarly adopted hybrid models, using AI for routine queries while relying on human agents for nuanced customer needs. The role of empathy and judgment remains a barrier to full automation.
Researchers describe the current state of AI as a “jagged frontier”: models excel at some tasks, such as math or coding, yet struggle with simple real-world tasks like interpreting timeframes or locations. Variations in data formatting and context can cause AI tools to misread information, requiring companies to invest heavily in data standardization and workflow integration.
AI providers are adapting to these challenges. OpenAI, Anthropic, and startups like Writer embed engineers directly with clients to ensure AI applications deliver tangible results. Specialized, sector-specific models are emerging as a more effective strategy than general-purpose solutions, emphasizing the need for guidance, integration, and continuous human oversight.
Ultimately, the AI revolution remains a work in progress. While the technology is powerful, businesses are discovering that realizing value is not instantaneous. Success depends on blending human expertise with AI, carefully targeting applications, and managing expectations. Companies that embrace this hybrid approach are more likely to capture meaningful benefits as AI continues to evolve.