By Keith Farris, E/CTRM Director
The energy industry is undergoing a rapid transformation, driven by factors such as the global shift towards renewable energy sources, increasing energy demand, and the growing complexity of energy markets. To navigate this complex landscape, energy companies are turning to artificial intelligence (AI) to optimize operations, enhance decision-making, and reduce error rates. AI and its natural language processing capabilities can be integrated with Energy Trading and Risk Management (ETRM) systems to provide immense efficiency and innovation. However, to maximize these benefits, a targeted, use-case approach is essential.
Let’s ground our AI expectations by using a simple scenario. We’ll start by giving our AI a name, like Emory. As a user, I want to ask Emory questions in the business terms I use most, and I expect Emory to understand our technology widgets and interact with them to answer my questions. When relevant historical data exists, I expect Emory to use it, but when it doesn’t, I expect Emory to intelligently generate reasonably correct data to answer my questions.
Current AI Landscape for Energy Trading
Most current uses of AI in ETRM are existing use cases which have been repositioned or marginally improved through AI. One example of this is RPA (robotic process automation) which has been ongoing in organizations for over 10 years. AI has made these improvements better and faster to realize, but these are improvements, not new solutions.
Another area where AI is currently gaining traction is in the design or solutioning phase. Businesses are identifying the questions they want AI to answer, but they haven’t fully solutioned or implemented how AI will answer those questions. For example, a trader asks “what was my 3rd most profitable day last quarter? Now, if those market conditions happen with these proposed trades, what would happen to my exposure and PnL?”. When the same trader asks a variation of the question again next week, how the AI derives its response will not be the same.
How to Operationalize AI for ETRM
Businesses which are on the leading edge of this evolution have realized that there are at least three key areas needing work before questions can truly be asked and answered.
- Large Language Models (LLM) must be taught to recognize terms like exposure, PnL, and market conditions. We can safely assume that the models have already been taught concepts like ‘profitable’ and ‘quarter’ because those terms are not domain specific. The models must then learn that terms like exposure and PnL, are NOT concepts that the model itself should calculate, but that they represent more complex scenarios which require widgets in the ETRM ecosystem. The models must also learn where to identify the inputs to these widgets (like market data) and how to extrapolate market data for conditions that don’t have any data. These are foundational to any LLM and AI to be used in the ETRM ecosystem.
- Parts of the ETRM ecosystem need exposure in a way that AI can discover and leverage them. This includes data, e.g. data cataloging, as well as logic. This is the point where potential AI challenges, like hallucinations, arise and must be carefully mitigated.
- Lastly, making the AI solution available to the correct users in a manner that is secure, cost efficient and scalable.
The Road Ahead
We are in the infancy of AI utilization in ETRM. As technology and processing power continues to evolve, AI has the potential to revolutionize operational processes and decision making. Imagine AI effortlessly breaking down complex P&L explanations, saving countless hours of work, or automating complex multi-step processes. The possibilities for AI in ETRM are limitless and poised to revolutionize the industry.
At MRE, our experienced team of ETRM experts are working with industry leaders every day to combine the power of your business with the depth of our experience. Reach out to us and discover how AI can strategically enhance your business.
