AI Projects Often Fail Due to Poor Data Quality

At Fortune’s Brainstorm Tech conference in Park City, Utah, executives highlighted a critical roadblock in AI adoption: bad data. The discussion revealed that many AI initiatives fail because companies overlook the importance of clean, consistent data.

Salesforce’s AI Agent Exposed Data Flaws

Shibani Ahuja, Senior Vice President of Enterprise IT Strategy at Salesforce, shared how an AI agent on their website began hallucinating and providing inconsistent results. The issue wasn’t the AI itself but contradictory "knowledge articles" published on the site.

• "The agent helped us identify a problem that always existed," Ahuja said.
• After cleaning up the data, the AI was repurposed as an "auditor agent" to detect anomalies in content.

MIT Study Confirms Widespread AI Pilot Failures

Ashok Srivastava, Chief AI Officer at Intuit, cited a recent MIT study revealing that 95% of AI pilots at large corporations fail, often due to outdated data systems.

• "People don’t invest in data—the foundation of AI," Srivastava said.
• He criticized companies for relying on archaic databases from the 1990s while expecting cutting-edge AI results.

Scaling AI Projects Remains a Challenge

Sean Bruich, SVP of AI and Data at Amgen, noted that even successful pilots struggle to scale enterprise-wide:

• "Pilots might deliver learnings, but scaling is where ROI happens," Bruich said.
• Large corporations often face difficulties transitioning from small proofs-of-concept to full deployment.

Key Takeaways

1. Data quality is non-negotiable for AI success.
2. Legacy systems hinder AI adoption.
3. Scaling requires more than just pilot projects.

For companies aiming to leverage AI, the message is clear: Start with the basics—clean, organized data—before diving into complex solutions.

The Fortune Global Forum will further explore these challenges in October 2025.