Grab Reduces Routine Analysis Work Using AI Agents

Grab, a leading ride-sharing and delivery service in Southeast Asia, is advancing the automation of data analysis tasks using AI agents. As a result, the proportion of routine mechanical work handled by analysts has decreased from 44% in February 2024 to 30% in June. This 14-point reduction over approximately six months demonstrates that efforts to integrate AI into business workflows are delivering concrete results.

First, let us understand what "routine mechanical work" means. In data analysis environments, repetitive tasks such as checking specific metrics, extracting data, and creating queries using SQL (a database query language) occur on a daily basis. These are tasks that can be systematized to some degree with specialized knowledge, and they can be viewed as "obstacles" preventing analysts from focusing on creative analysis. The underlying philosophy of this initiative is that by having AI take on these tasks, human experts can dedicate more time to higher-level judgment and interpretation.

Grab's approach centers on a multifaceted strategy built around AI agents. An AI agent refers to an AI system capable of autonomously executing multiple steps based on given instructions. Beyond the agent's autonomy, Grab combines four elements: the use of quality-assured "certified data," a mechanism for the agent to properly maintain context about tasks, and a framework for human oversight and verification of results. Additionally, Grab has established a "self-service analytics" infrastructure, enabling requests such as metric checks, data queries, and SQL creation to be handled without analyst intervention.

It is important to understand the background behind Grab's initiative. Grab generates vast amounts of transaction data daily across Southeast Asian countries, and the demand for analyzing this data continues to grow with business expansion. Meanwhile, hiring and developing advanced data analysts is costly and time-consuming. Automation through AI is positioned as a practical means to bridge this gap. Recently, many tech companies have explored automating data analysis, but few have published results backed by actual figures.

The significance of this initiative for the industry extends beyond efficiency gains. Grab's case demonstrates a model in which AI agents are not left entirely to their autonomy but are operated in combination with certified data and human oversight, thereby achieving both reliability and practical effectiveness. The insight here is that simply deploying AI does not automatically guarantee precision or trustworthiness; rather, how data quality management and human involvement are designed proves critical.

Going forward, two points warrant attention: how much further the proportion of routine tasks will decline, and how analysts will redirect the freed-up work toward high-value-added activities. The true value of business automation through AI lies not in the work time saved itself but in how the freed resources are utilized. Grab's case represents a practical attempt to address this question and may serve as a reference for other companies facing similar challenges.