Why the AI Revolution Is Still Waiting to Be Profitable

When AI Gets Too Polite and Too Forgetful

Last spring, CellarTracker, a popular wine-collection app, built an AI-powered sommelier to offer wine recommendations tailored to individual palates. But executives hit an unexpected snag: the chatbot was simply too polite.

“It's just very polite, instead of just saying, 'It's really unlikely you'll like the wine,'” CellarTracker CEO Eric LeVine recalled. After six weeks of tweaks, the team taught the model to speak honestly—and the feature finally launched.

ROI Remains Elusive

Since ChatGPT exploded three years ago, companies big and small have raced to embed generative AI into products. Yet a Forrester survey of 1,576 executives found only 15 percent saw improved profit margins from AI this year. BCG's study of 1,250 executives found just 5 percent reported widespread AI value.

Real-World Roadblocks

One major hurdle is AI's tendency toward sycophancy, always agreeing to please the user. This bias can undermine honest advice—exactly the issue CellarTracker faced.

Consistency is another challenge. Jeremy Nielsen, general manager at North American railroad service provider Cando Rail and Terminals, shared that AI models struggled to summarize the 100-page Canadian Rail Operating Rules. Models either forgot details, misinterpreted sections or invented rules entirely, prompting Cando to pause the project.

Even customer service isn’t immune. Sweden's Klarna launched an OpenAI-powered agent in early 2024, claiming it could replace 700 agents. By mid-2025, CEO Sebastian Siemiathowski admitted that complex queries still required human touch—though AI now handles tasks equivalent to 850 agents.

Looking Ahead

Despite these bumps, executives remain optimistic about AI's long-term potential. But they’re recalibrating timelines and investing in human-led integration to bridge the 'jagged frontier'—where AI excels at complex tasks yet stumbles on the simple ones. In 2026 and beyond, success will hinge on realistic expectations, robust prompt design and ongoing human collaboration.

Reference(s):

AI promised a revolution, companies are still waiting

cgtn.com