Enterprise AI Data Governance Reveals Double the Bad Answers
A new study indicates that 68% of enterprises have identified AI agents confidently providing incorrect answers due to missing or inconsistent business context. This issue is more prevalent in companies with a managed semantic layer.
The context fed to AI agents is frequently failing within enterprises, according to VentureBeat's Pulse Research. Over the past six months, 68% of companies have traced confident but incorrect AI agent answers to missing or inconsistent business context. Most commonly, these failures are not isolated incidents but recurring.
Counterintuitively, enterprises building or operating a governed semantic layer, which provides company-specific definitions and relationships, report recurring failures at more than twice the rate of those without one. This semantic layer does not cause the failures but serves to detect and trace them more effectively.
The research, surveying 101 enterprises, suggests that context failure is no longer an anomaly but a condition. Thirty-seven percent of companies report recurring failures, compared to 32% who experienced them once. The majority of these issues stem from data context deficiencies rather than model errors.
The industry's adopted solution, a governed semantic or context layer, is gaining traction: 32% of enterprises are using it in production, with another 31% piloting or building one. Response correctness is the primary success metric for 38% of businesses, indicating a shift toward prioritizing context governance over mere data movement.