AES: Scaling energy safety audits with Anthropic’s Claude on Google Cloud
Reduces costs of audits by 99%
Accelerates audit results from 14 days to one hour
Doubles annual audit capacity
Increases audit accuracy by up to 20%
AES has significantly increased the accuracy and speed of health and safety audits by deploying gen AI agents built using Vertex AI and Anthropic’s Claude models.
AES has significantly increased the accuracy and speed of health and safety audits by deploying gen AI agents built using Vertex AI and Anthropic’s Claude models.
A massive audit process overhaul
Vertex AI Model Garden makes it easier for us to build AI-powered applications by simplifying the use of pre-trained learning models including Claude. Running Claude on Vertex AI is driving immense improvements, such as a 20% increase in audit accuracy and a 99% reduction in the cost of each audit.
Marwan Sherri
Data Scientist, AES
With more than four decades of experience in the energy sector, AES is helping the world transition to clean, renewable energy. The US-based energy company owns and operates power plants, which it uses to generate and sell electricity to end users and intermediaries including utilities companies and industrial facilities.
AES puts safety first by dedicating more than 1,500 safety management program audits annually. These include both external audits to ensure adherence to standards at individual sites and internal audits conducted by individual sites to self-assess compliance and performance. Each audit takes approximately 100 employee hours to complete, including document reviews and checks for thoroughness and accuracy.
Dr. Sean Otto, Senior Director of Data Science & Analytics, AES, says, “We've trained various team members internally to conduct these audits, even though it's not their primary responsibility. So, any opportunity to reduce the internal effort required while maintaining accuracy is important.”
Otto, along with Dr. Marwan Sherri, Data Scientist, AES, saw the potential of AI-agents to automate decision-making and tasks by processing data, learning from patterns, and optimizing the auditing process. “We needed this process to be at least 100 times faster than a human,” says Sherri. “The faster the response, the better it could generate accurate results and follow instructions reliably.” Sherri also wanted agents to perform without the need for constant monitoring or re-evaluation.
The new way to conduct safety audits with gen AI
Accuracy and speed determined the selection of technology partners for the project. AES chose Anthropic’s Claude models on Vertex AI based on the ability of its foundational model family, Claude, to handle large volumes of API calls and synchronize agents for complex safety audit tasks. “This means that we can synchronize all our agents seamlessly, significantly accelerating the report generation process,” says Sherri.
AES also chose Claude thanks to integrations with its long-term data and analytics provider, Google Cloud. This includes Google Cloud tools such as Vertex AI, which enables businesses to experiment with, deploy, and manage models on scalable infrastructure. AES also benefits from Vertex AI’s built-in security, privacy, and compliance measures to protect AES’s enterprise data.
I love the accuracy of Anthropic’s Claude models and the security and advanced AI tools that Google Cloud provides to utilize these models for our auditing process.
Sean Otto
Senior Director of Data Science & Analytics, AES
“Connecting Claude to Vertex AI wasn’t hard at all,” says Sherri. “We found the solution by enabling Claude to be used through API calls within the Vertex AI environment. This approach resolved relevant security issues such as managing permission levels and configuring service accounts.”
Sherri was impressed by Vertex AI Model Garden, a platform that provides access to state-of-the-art first party, third-party, and open-source models from Google Cloud and its partners, including Anthropic. “The integration of Vertex AI Model Garden and Anthropic's Claude provides a powerful and convenient environment for building AI agents. It offers us easy access to a highly capable language model, simplifies deployment, and accelerates the development process,” says Sherri. In the end, it took just two months to make the AI agents operational, including building knowledge bases and fine-tuning AI agents.
With its agents now operational, AES has seen significant improvements in the audit process, especially at the document reviews stage, which can include up to 400 pages at a time. And because AES is a multilingual corporation, agents help accelerate document analysis when audits involve multiple languages. Agents also reduce the human effort needed to maintain attention and focus on details across so many documents.
After conducting more than 50 agent audits, AES saw audit costs reduced by 99% while results were delivered 99.7% faster. “Our auditors previously spent 14 days completing each audit process. Now, with our Claude-powered agents on Vertex AI, the same work is completed in just one hour,” says Otto. “We can double the number of audits conducted, as our gen AI agents handle half of the workload.” The other half, which involves a ‘human in the loop’ examining the process, remains in place.
AES has also seen a 10-20% increase in audit accuracy. This precision significantly improves audit quality and ensures safety compliance and best practices across all sites.
Otto is full of praise for the Google Cloud team that supported the AI-agent project. “Partnering with Google is one of the best vendor relationships I have ever had.” The project also opens the door to further innovation based on Google Cloud, Vertex AI, and foundation models. “You can see the potential for optimizing energy generation from renewable sources, predicting demand more accurately, and automating maintenance through predictive analytics. It’s a huge opportunity for AES and the energy generation industry as a whole,” he says.
AES is an American utility and power generation company with significant operations across North and South America. It designs, owns, and operates power plants, which it uses to generate and sell electricity to end users and intermediaries such as utilities companies and industrial facilities.
Industry: Utilities
Location: United States
Products: Vertex AI, Anthropic’s Claude models on Vertex AI
About Google Cloud Partner- Anthropic
Anthropic is an AI safety and research company which builds reliable, interpretable, and steerable AI systems.
