AES cuts safety audit costs 99% using Claude AI agents on Google Cloud's Gemini Enterprise Agent Platform

Overview

AES built AI agents using Anthropic's Claude models on Google Cloud's Gemini Enterprise Agent Platform to automate its health and safety audit process, which previously required about 100 employee hours per audit across more than 1,500 audits a year. After a two-month build, AES ran more than 50 agent audits and saw audit costs fall 99%, audit turnaround drop from 14 days to one hour, audit accuracy rise 10-20%, and annual audit capacity double, with a human still reviewing each agent-assisted audit.

The challenge

AES dedicates more than 1,500 safety management program audits annually, each taking approximately 100 employee hours including document reviews of up to 400 pages, conducted by team members for whom auditing is not their primary responsibility, creating a need to reduce internal effort while maintaining accuracy.

The solution

AES chose Anthropic's Claude models on Google Cloud's Gemini Enterprise Agent Platform, accessed through Model Garden, connecting Claude via API calls within the Agent Platform environment to manage permission levels and service accounts, and built AI agents by creating knowledge bases and fine-tuning them over a two-month period to synchronize agents for complex safety audit tasks including multilingual document analysis.

Reported business value

After more than 50 agent audits, AES cut audit costs by 99%, reduced audit turnaround from 14 days to one hour, increased audit accuracy by 10-20%, and can double the number of audits conducted since AI agents now handle half of the workload, with a human still reviewing each audit.