Grab reduces mechanical analytics work from 44% to 30% using AI agents with human oversight

Grab’s AI agents automate analytics workflows by handling tasks like data preparation, query execution, and reporting. The system routes requests through specialized workflows, reducing the mechanical workload for analysts. This change shifts human effort from repetitive tasks to higher-value activities like causal interpretation and business decision-making. The approach relies on a structured autonomy model, where agents operate within defined boundaries while humans retain final accountability.

The reduction in mechanical work, from 44% to 30%, reflects the system’s ability to handle routine requests independently. Self-service analytics adoption grew significantly, with 67-90% of metric, data, and SQL requests now resolved without human involvement. However, this efficiency depends on Grab’s investment in certified data and context management. The company maintains thousands of certified tables and context documents to ensure agents produce reliable results, highlighting the need for robust data infrastructure.

Grab’s implementation includes operational AI agents like Scarlet, which performs root cause analysis for pipeline failures and can fix or escalate issues. For recurring analytics, agents automate metric commentary and OKR breakdowns, correlating data with operational changes. While this reduces manual effort, it also introduces new dependencies on agent accuracy and context updates. Teams adopting similar systems must balance automation benefits with the cost of maintaining data quality and oversight frameworks.