AI Promised a Revolution. Companies Are Still Waiting for Returns

Three years after ChatGPT's explosive debut, the promised AI revolution in business remains largely unrealized. Despite massive investments in artificial intelligence infrastructure, the vast majority of companies are struggling to see meaningful returns, according to recent executive surveys and interviews with business leaders.

The disconnect between AI's potential and its current business impact has emerged as one of 2025's most significant technology stories, with implications that extend far beyond quarterly earnings reports.

The Numbers Tell a Sobering Story

Two major surveys paint a stark picture of AI's business performance. Research and advisory firm Forrester found that only 15% of 1,576 executives surveyed during the second quarter reported profit margin improvements due to AI over the past year.

Consulting firm BCG's findings were even more concerning: just 5% of 1,250 executives surveyed between May and mid-July saw widespread value from their AI investments.

These figures stand in sharp contrast to the enthusiasm that swept through corporate boardrooms following ChatGPT's November 2022 launch. Companies worldwide created dedicated task forces to embrace generative AI, eager to integrate the technology into products and workflows.

Now, Forrester predicts that companies will delay approximately 25% of planned AI spending by a year in 2026 as organizations reassess their strategies.

"The tech companies who have built this technology have spun this tale that this is all going to change quickly," said Brian Hopkins, a Forrester analyst. "But we humans don't change that fast."

When AI Gets Too Polite: Real-World Implementation Challenges

The gap between AI capability demonstrations and practical business applications has surprised even experienced technology executives.

CellarTracker, a wine-collection app, encountered an unexpected problem when building an AI-powered sommelier using OpenAI's technology. The chatbot performed well with general recommendations but struggled with honesty about specific vintages.

"It's just very polite, instead of just saying, 'It's really unlikely you'll like the wine,'" said CellarTracker CEO Eric LeVine. The company spent six weeks in trial and error, designing prompts that gave the model permission to deliver negative assessments.

This issue—what AI researchers call "sycophancy"—encourages users to chat more but impairs the model's ability to provide genuinely useful advice.

Other companies have faced more fundamental obstacles. Cando Rail and Terminals, a North American railroad service provider, tested an AI chatbot to help employees study internal safety reports and training materials. The project hit an unexpected wall: the models couldn't consistently and correctly summarize the Canadian Rail Operating Rules, a roughly 100-page document outlining industry safety standards.

Sometimes the models forgot or misinterpreted the rules. Other times they invented them entirely—a phenomenon AI researchers call "hallucination." The company has spent $300,000 on AI development efforts and suspended the safety documentation project.

"We all thought it'd be the easy button," said Jeremy Nielsen, general manager at Cando. "And that's just not what happened."

The Human Element Returns

Human-staffed call centers and customer service operations were expected to face heavy disruption from AI. Instead, companies are discovering limits to how much human interaction can be delegated to chatbots.

Swedish payments company Klarna made headlines in early 2024 when it rolled out an OpenAI-powered customer service agent that it claimed could do the work of 700 full-time agents. By 2025, however, CEO Sebastian Siemiathowski acknowledged that some customers preferred human interaction.

While Klarna's AI now handles the equivalent of approximately 850 agents' work on simple tasks, complex issues are quickly referred to humans. The company is developing a second-generation chatbot for 2026, but human agents will remain central to operations.

"If you want to stay customer-obsessed, you can't rely [entirely] on AI," Siemiathowski said.

U.S. telecommunications giant Verizon is similarly reemphasizing human customer service agents in 2026 after attempts to delegate more calls to AI systems.

"I think 40% of consumers like the idea of still talking to a human, and they're frustrated that they can't get to a human agent," said Ivan Berg, who leads Verizon's AI-driven service operations for business customers.

The company, with about 2,000 frontline customer service agents, now uses AI to screen calls and gather customer information, then directs people to either self-service systems or human agents. This approach frees up agents to handle complex issues and pursue new activities like outbound calls and sales.

"Empathy is probably the key thing that's holding us from having AI agents talk to customers holistically right now," Berg said.

The Jagged Frontier: Excellence and Failure Coexist

Large language models demonstrate what researchers call a "jagged frontier" of capabilities—rapidly conquering complex tasks in mathematics and coding while failing at comparatively trivial ones.

"It might be a Ferrari in math but a donkey at putting things in your calendar," said Anastasios Angelopoulos, CEO and cofounder of LMArena, a popular AI benchmarking tool.

Financial firms have discovered that data formatting differences across sources can prompt AI tools to "read patterns that don't exist," according to Clark Shafer, director at Alpha Financial Markets Consulting. Many companies are now considering the expensive and complex process of reformatting their data to take advantage of AI capabilities.

Dutch technology investment group Prosus built an in-house AI agent to answer questions about its portfolio—tasks that data analysts already perform. In theory, an employee could ask how often a Prosus-backed food-delivery firm was late delivering sushi orders in Berlin last week.

In practice, the tool doesn't always understand what neighborhoods constitute Berlin or what "last week" means, said Euro Beinat, head of AI for Prosus.

"People thought AI was magic. It's not magic," Beinat said. "There's a lot of knowledge that needs to be encoded in these tools to work well."

Industry Response: More Support, Specialized Solutions

Recognizing these challenges, AI companies are shifting their approach to business customers.

OpenAI is developing new products for enterprises and has created internal teams, including its Forward Deployed Engineering team, to work directly with clients on specific problems. The company is helping businesses identify areas where AI can have "high impact but maybe low lift at first," according to Ashley Kramer, OpenAI's head of revenue.

During a recent lunch with media editors in New York, OpenAI CEO Sam Altman estimated that developing AI systems for companies could be a $100 billion market.

Rival AI lab Anthropic, which draws 80% of its revenue from business customers, is hiring "applied AI" experts who will embed with companies. For AI companies to succeed, they must view themselves as "partners and educators, rather than just deployers of technology," said Mike Krieger, Anthropic's head of product.

An increasing number of startups—many founded by former OpenAI employees—are developing AI tools for specific sectors such as financial services or legal work. These founders argue that companies will benefit more from specialized models than from general-purpose consumer tools like ChatGPT.

Writer, a San Francisco-based AI application startup building agents for finance and marketing teams at firms like Vanguard and Prudential, places engineers directly on client calls to understand workflows and co-build solutions.

"Companies need more handholding in actually making AI tools useful for them," said May Habib, CEO of Writer.

Expert Perspective: The Automation Reality

Hamza Baig, founder of the Automation Institute and Hexona Systems, has observed these challenges firsthand while training over 30,000 students in automation workflows.

"The gap between AI capability and business implementation isn't primarily a technology problem—it's a process design problem," Baig said. "Companies are trying to drop AI into existing workflows without redesigning those workflows around what AI actually does well. That's like trying to use a Ferrari to plow a field. You need to match the tool to the task, and often that means completely rethinking how work gets done."

Baig's organization has focused on teaching businesses to identify specific automation opportunities rather than pursuing broad AI transformation initiatives.

"The companies seeing returns from AI are those that started small, measured carefully, and scaled what worked," he added. "The ones struggling tried to transform everything at once. Automation works best when it's strategic and incremental, not revolutionary and wholesale."

High Stakes: Infrastructure Investment Meets Reality

The business struggles with AI implementation are unfolding against unprecedented technology infrastructure investments—in chips, data centers, and energy sources to power them.

Whether these investments can be justified depends on companies' ability to use AI to boost revenue, increase margins, or accelerate innovation. Without demonstrated business value, the infrastructure build-out could trigger a crash reminiscent of the dot-com bust in the early 2000s, some experts warn.

For now, executives maintain that generative AI will eventually transform their businesses. However, they're reconsidering how quickly that transformation will happen within their organizations.

Shashi Upadhyay, president of product, engineering and AI at customer-service platform Zendesk, notes that AI excels in three areas: writing, coding, and chatting. Zendesk's clients use generative AI to handle between 50% and 80% of customer-support requests.

But the idea that generative AI can do everything is "oversold," he said.

Looking Forward: A More Measured Approach

The recalibration of AI expectations in business represents a maturation of the technology sector's approach. Rather than seeking immediate transformation, companies are beginning to identify specific, high-value use cases where current AI capabilities align with actual business needs.

This more measured strategy may ultimately prove more sustainable than the initial rush to integrate AI everywhere possible. As companies learn to work within AI's current limitations while pushing for incremental improvements, the technology may finally begin delivering on its business promise—just not in the revolutionary timeframe initially anticipated.

The question for 2026 and beyond is whether AI companies and their business customers can close the gap between capability and implementation before investment patience runs out.