A new report from Google Cloud and MIT Technology Review Insights highlights a major trend in how companies use artificial intelligence. As businesses look to move beyond simple pilot programs and use AI agents across their entire organizations, the findings show that success depends on the quality and accessibility of the company’s underlying data.
The report, “Scaling AI Agents with Trustworthy Data,” includes survey results from 300 data and technology executives. It also features interviews with leaders from major companies like HCA Healthcare, Shopify, and Deutsche Telekom. The central message is that while most organizations are starting to use AI agents, they face significant hurdles in moving to widespread adoption.
The Shift to Widespread AI
Currently, only 10% of surveyed organizations use AI agents widely across their business. However, that is expected to change quickly. Within two years, 69% of organizations plan to deploy AI agents at a larger scale. This represents a move from small experiments to operations that span entire companies.
The research shows that organizations often struggle to prepare their data for this transition. For AI agents to perform useful tasks, they need reliable, real-time access to information. Many companies find that their current systems cannot provide this.
Why Legacy Systems Hinder Progress
More than half of the executives surveyed (55%) said that their older, legacy data systems are preventing them from scaling AI. The report identifies four main problems with these older systems:
- Data Silos: Information is trapped in disconnected systems, so AI agents cannot see a complete picture of the business.
- Unstructured Data: Teams cannot easily use data found in documents like PDFs, emails, videos, and call logs.
- Slow Access: Older systems process data in batches rather than in real time, making it hard for agents to react to events as they happen.
- Lack of Context: Agents often lack the business context needed to understand what the data means and how it relates to other parts of the organization.
The Difference Between Leaders and Laggards
The report distinguishes between two groups: “data leaders” and “data laggards.”
Data leaders are defined as organizations that give their AI systems access to more than 70% of their enterprise data. Data laggards provide access to 30% or less. On average, most organizations only grant access to 45% of their data for AI.
This difference in access directly affects trust. Only 22% of data laggards trust the accuracy of their AI agents’ decisions. In contrast, 100% of data leaders report that their AI is either “mostly” or “consistently” accurate. By opening up access to their data, these leading organizations are finding more success and can scale their AI efforts more easily.
Next Steps for Organizations
To succeed in the coming years, organizations are focusing on three main data initiatives:
- Broaden Data Access: Connecting systems to activate both structured and unstructured data.
- Improve Governance: Adding business context to data models so AI agents can work more accurately.
- Use Real-Time Data: Moving from batch processing to streaming systems, allowing for faster decision-making.
The report concludes that trying to force AI into old, rigid architectures is inefficient and costly. Instead, it suggests that companies should aim to build a “System of Action” rather than just a “System of Intelligence.” This approach moves data from a static storage space to a dynamic platform, ensuring that the AI can act with the accuracy and speed required for today’s business needs.
To read more, read the blog announcement.
