Demonstrating ROI is one of the biggest challenges of convincing leaders to invest in AI, especially for conservative sectors like energy and utilities. But there is a model to demonstrate AI ROI, and we are starting to see it work.
AI is transforming how we work and live, but the changes come so fast that quantifying the bottom-line AI ROI can be challenging. While that trend is improving overall, with 59 percent of companies seeing revenue gains and 42 percent enjoying lower costs with AI, the E&U sector continues to lag in AI implementations.
In this blog, we’ll share the story of one E&U company bucking the trend. A valve inspection company that serves the oil and natural gas industry, our client exhibits several best practices of AI adoption:
- Start with a business strategy
- Tackle a small project first as a proof of concept (PoC)
- Use tools you already have
- Evaluate any new tools by how they can grow and change with your business
By using our client’s successful examples of these principles, we will show that any industry can document AI ROI in safe ways that encourage further investment and even greater gains.
Start With Business Strategy Before You Implement AI
Successful AI projects start with a business strategy. What problem are you trying to solve? What are the benefits of solving the problem? How do they fit into your long-term company goals?
The problem we set out to solve for our client was straightforward — but it had a far-reaching, strategic impact. The company installs, maintains, repairs, and tests valves that must meet precise specifications for safety and reliability. A failed valve could lead to disaster.
The challenge? Ensuring that each of the thousands of valves on a pipeline performs as specified is a slow, costly, and labor-intensive process. A single pipeline will have valves of various configurations from multiple manufacturers. Once installed, our client’s quality control technicians must periodically inspect each valve.
Using hard-copy or PDF manuals, the technicians undertook a time-consuming process to retrieve specs for each part of each valve. The process was not only costly and time-consuming but also imprecise. Different manufacturers would use various combinations of text, tabular data, and graphics to present the same types of information.
As a result, the maintenance process was slow, costly and inaccurate because it required a separate quality assurance team to check and correct the quality control technicians’ work.
Here is one way to phrase our client’s situation strategically:
- The Problem: The inspection process is expensive, slow and imprecise.
- The Benefit: Improved efficiency (pipeline performance and employee’s time), saved costs (less costly inspection process and avoid failure or repairs), enhanced safety, increased reliability, met sustainability goals.
- Long-term Goals: Automating the inspection process with AI is a step toward integrating it into the client’s supervisory control and data acquisition (SCADA) system. As the AI learns more about the pipeline, it can take on a self-healing role, making adjustments that prevent problems. This capability can help the company operate its existing pipelines and move into other fields, like biofuels.
In our client’s case, AI efficiency solves a problem that enables long-term objectives to help the client grow and maintain its competitive edge.
Take On a Small Project — Then Make It Smaller
One challenge to securing approval for AI projects without a clear line to ROI is how overwhelming they can seem. Leaders may agree that long-term goals are valuable, but completing an AI project can appear insurmountable.
As our client did, start with a fairly straightforward problem rather than a massive AI transformation. Then, work to limit the focus.
For example, our client chose to start with a pilot program focusing only on one of their many manufacturers. While there would be some variation among that manufacturer’s documentation, it would not be as great as working with documents from multiple manufacturers.
Our team further broke the project down by adopting a parallel, two-track approach. In the first track, our AI experts built an ML model capable of detecting anomalies. In the second track, data architects deployed a large language model (LLM) to extract the data from the documents that would be used to train the model. For this project, the LLM was only a means of feeding the data into the ML model for anomaly detection.
Use Resources You May Not Know You Have (Yet)
Many AI projects become easier when you realize that AI features are already built into many apps your business uses daily. The familiarity of working within well-known platforms will lower stress levels while saving the time and expense of finding and purchasing new tools.
In a more advanced way, we used the element of familiarity with our client by adapting existing, proven frameworks to build our ML and LLM models. That solid foundation allowed our AI experts in the U.S. and our India practice to focus more on solving specific tasks, like working with a large number of diagrams and tables.
We used segmentation technology to extract data from images and diagrams and used the LLM to identify and understand table structures to achieve this task. The LLM then reassembled slices of images and pieces of tables into the JSON format for the ML to read.
Use Agile Tools
The best tools are those that can grow and adapt with you instead of painting you into a corner.
For our client, we used a propriety AI agent called Agent C. Each time our team gathered a critical piece of information, such as “This company’s body seat specs look like this,” Agent C incorporated that chunk of data into an always-expanding database that continuously adapts new information, speeding up data extraction. Because Agent C retains each component of information, it can grow beyond the scope of this project to future projects and other industries.
The Path to Showing AI ROI
Your AI projects can have demonstrable ROI. Our client knew the time and financial cost of gathering accurate data from the field. We built on our existing ML and LLM capabilities to create a solution that greatly reduces those costs. As the tools generate more evidence of their savings, leaders can more easily see the benefits of more ambitious AI products.
Are you ready to explore how artificial intelligence can fit into your industry but aren’t sure where to start? Our AI experts can guide you through the entire process, from planning to implementation. Talk to an expert