Grab Cuts Mechanical Analytics Work From 44% to 30% with AI Agents

Details

Grab has published a detailed account of how it deployed AI agents to automate analytics workflows, reducing the share of mechanical work handled by human analysts from 44% in February to 30% in June 2026. Tasks automated include data preparation, alerting, and recurring reporting.\n\nThe system, called Spartan, operates on a 5-level autonomy framework. At Level 3, agents handle data discovery, query writing, and result validation while humans review outputs. Level 4 lets agents plan and orchestrate workflows with human gates at defined checkpoints. Level 5 — end-to-end autonomy — requires only objective-setting and quality thresholds from humans. Causal interpretation and final business decisions remain exclusively human responsibilities.\n\nSpartan handles natural language requests submitted via Slack, routing them through 50+ skills and 120 analysis frameworks. Root cause questions can trigger multi-dimensional analysis across certified metrics; experiment questions may retrieve existing scorecards rather than re-querying the data lake.\n\nThe reliability of this system depends heavily on data infrastructure: 5,000+ certified tables, 4,000 context documents, and 2,000 golden records maintained by a ContextIQ system that updates context as instrumentation changes. A separate agent (Scarlet) handles pipeline failures through automated root cause analysis.