How AI drove data optimization for oil and gas capital projects

As part of its digital transformation initiative, a client in the oil and gas industry aimed to enhance clarity and efficiency in its engineering processes. It engaged its internal capital projects design team to develop clear engineering requirement statements and establish relationships among them, allowing relevant requirements to be easily identified for design, procurement and construction. However, traditional methods were labor-intensive and prone to errors. To streamline the process, improve predictability and enhance the accuracy of the engineering requirements catalog, the client contacted Ernst & Young LLP to explore potential solutions.

Understanding the client’s problem necessitated a deeper dive into the existing capital project development and execution processes, with a focus on the use of engineering requirements. The primary issue was the volume of content in a library of more than 750 documents. Each document contained 30 pages that had more than 100 requirements, with cross-references to other documents. The client’s vision was to increase throughput dramatically and create a system that could process this large volume of content in less than a month.

The client aimed to reduce the use of an internal set of requirements by leveraging industry standards, enabling end users — such as engineering, procurement and construction (EPC) contractors or subcontractors, equipment manufacturers, subject-matter experts and client engineers — to easily access these requirement sets for design, procurement and construction activities.

“Our goal is to create a more cohesive, user-friendly and digitalized requirements library that enables our capital project teams to deliver projects with greater precision and effectiveness,” the client said.

As part of the transformation, the client decided to first rationalize engineering requirements against industry standards and rewrite them based on technical standards from the International Council on Systems Engineering (INCOSE) and the Easy Approach to Requirements Syntax (EARS). This effort would bring consistency and clarity in requirements. To facilitate easy access to the client’s capital projects and engineering teams, the project team needed to assign metadata tags to every requirement statement. Each metadata tag would need to be selected from a library based on an equipment hierarchical taxonomy with over 1,000 options.

Artificial intelligence (AI) presented a potential solution to address these inefficiencies by transforming unstructured project and engineering standards into structured requirements data. A method to rewrite documents and assign metadata to engineering requirements automatically was needed. The client had limited capacity, as discipline engineers could only dedicate 20% of their time to the project. Despite involving subject-matter experts in the initial manual efforts, the tagging outputs were suboptimal, leading to reduced process throughput.

Maintaining smooth processes in revenue-generating operations and capital project execution work had to be the top priority, leaving limited time for implementing big-picture projects — even initiatives that would ultimately streamline processes.

In the EY team’s view, AI made the most sense as a potential solution to automate the rewriting and tagging functions, allowing more time for validation and other value-adding tasks and reducing the overall restructuring effort. However, the client was unsure of whether AI technology could efficiently and accurately manage the specificity and complexity of the engineering standards.

To address these concerns, the EY team needed to demonstrate the capability of AI to improve accuracy and efficiency, build the client’s confidence in the technology and guide it through the necessary steps for integration into its operations.

The EY team introduced the client to the idea of an AI-based solution through a proof-of-concept (POC) application, designed and developed within the secure EY Fabric environment. This POC utilized publicly available data and mock standards to create rewrite and metadata application agents powered by Generative AI technology housed in the cloud. These AI application modules were integrated through workflow processes on a web application, creating a seamless and efficient operation.

This showcased the model’s ability to accurately tag and rewrite documents, and its initial success in doing so helped build the client’s confidence. The team then proceeded to apply the solution to the client’s actual engineering standards documents, tasking subject-matter experts to provide feedback and verify the model’s accuracy and relevance.

The experience of the EY team proved to be a strong asset in this effort. With backgrounds in AI, digital transformation and engineering, team members brought a wealth of industry-specific knowledge. Their hands-on experience in the oil and gas sector, including time spent on rig sites and at refineries, provided them with firsthand insight into the client’s concerns, enabling productive dialogue. The team comprised AI and data analysts, digital specialists, project managers, electrical engineers, geophysicists and chemical engineers, all contributing to a well-rounded and well-informed approach.

The web application was enhanced during the pilot phase to include value-added features such as identifying similar requirements, building cross-references, and offering import and export capabilities.   The EY architecture encoded standards documents into multi-step prompts, using Generative AI for rewriting requirements. Additionally, cognitive services and vector database storage were utilized for efficient requirement-to-metadata matching, and confirming tag relevance using a combination of Generative AI and Machine Learning models.

Throughout implementation, the EY team maintained a focus on responsible AI, confirming that the solution adhered to governance principles and included human oversight. This approach helped the client improve its processing capabilities. It also resulted in a smooth transition to an AI-driven workflow, setting the stage for future scalability and innovation.