Companies Struggle to Unlock AI’s Promise

• Generative AI was expected to transform industries quickly, yet many firms are still waiting for tangible results.
• Surveys show only a small fraction of executives report improved margins or widespread value from AI investments.
• Businesses are now rethinking timelines, focusing on smaller, practical applications rather than sweeping transformations.

Early Hype Meets Reality

Since ChatGPT’s debut three years ago, companies have rushed to embed generative AI into products and workflows. Many hoped for immediate gains, but surveys by Forrester and BCG reveal only 15% and 5% of executives, respectively, saw measurable benefits. Analysts predict that by 2026, about a quarter of planned AI spending will be delayed. The gap between expectations and reality reflects both technical challenges and the slower pace of organizational change.

Firms experimenting with AI often encounter issues of bias, inconsistency, and reliability. CellarTracker’s wine recommendation chatbot struggled to deliver honest feedback, while Canadian rail service provider Cando found models misinterpreted safety rules. These shortcomings highlight AI’s tendency to please users rather than provide critical assessments, and its difficulty handling long, complex documents. Companies have invested heavily—Cando alone spent $300,000—yet many projects remain stalled or abandoned.

Human Interaction Still Essential

Customer service was expected to be reshaped by AI, but firms discovered limits to automation. Klarna initially claimed its chatbot could replace hundreds of agents, only to scale back when customers demanded human contact. Verizon also reintroduced human agents after finding that nearly half of consumers preferred speaking to people. AI now handles routine tasks, freeing staff for complex issues, but empathy remains a barrier to full automation.

Researchers describe AI’s uneven abilities as the “jagged frontier,” where systems excel at advanced math yet falter at simple scheduling. Financial firms report that inconsistent data formats can cause AI tools to misread patterns, while investment groups like Prosus note that models struggle with basic context such as geography or timeframes. To address these gaps, AI companies like OpenAI and Anthropic are embedding engineers directly with clients, emphasizing education and tailored solutions. This shift suggests that specialized, sector-focused AI tools may ultimately deliver more value than broad, general-purpose systems.