AI in Energy Trading: Speed, Transparency, and Smarter Decisions

Keith Farris:

I think it's a little bit like high school algebra. You've got to be able to show your work along the way. If you can ask the model, substantiate what you just suggested to me. Show me the steps that you used to get to that. Because now as a trader or as a market risk professional, I can look at that and say, yep, I agree with each step. Or no, I disagree with it. Run the model again with a different view.

Ramakrishna Manne:

The AI models, basically you can run several different scenarios and come and the AI can actually come up with the recommendation. So before the trader gets in, before the markets open, they are ready, they already reviewed all of the recommendations, what happened before the market and they are ready to execute the trades.

Chad Watt:

Welcome to the Infosys Knowledge Institute podcast, where business leaders share what they've learned on their technology journey. I'm Chad Watt. Today I'm joined by Keith Farris, Managing Director at MRE Consulting, and Ramakrishna Manne, Energy and Utilities Global Managing Partner at Infosys Consulting. Welcome guys.

Keith Farris:

Thanks.

Chad Watt:

Keith, let's talk about the current market reality for energy trading. What is the single biggest challenge for an energy trader today trying to manage risk?

Keith Farris:

Yeah, it's a great question. I think the rate of change that's going on in the marketplace, the speed at which assets come online and markets change is probably the biggest driver. And then what that does to uncorrelated markets is probably the thing we see most.

Chad Watt:

Rama, give me your thoughts on that rate of change, uncorrelated markets, what else?

Ramakrishna Manne:

So a lot of internal data, right? You have all of the operational data, weather data, you have geopolitical events that are happening, and you have supply chain shocks that can happen. And all of this causes a lot of noise. And you have to be able to separate the noise from and actually make a decision that matters.

Chad Watt:

I'm going to ask you expand on that little bit, Rama. Trading desks have more data than ever. How are traditional human-led analysis methods managing this volume and this data?

Ramakrishna Manne:

It takes a lot longer to kind of comprehend this data and to be able to make a decision that they're comfortable with. What AI can do is it can help accelerate. It can run multiple scenarios at once and give a recommendation. And then the trader can use that data and be ready to execute the strategy.

Chad Watt:

You're in the field with customers grappling with AI. What does early success look like?

Keith Farris:

I think it starts with getting started, like finding a place to use AI and often that looks like just use it in your daily task, right? How can AI make me faster and better today? And failure's okay, right? Like if you start and you use, you can use Claude, you can use ChatGPT, but use it. Ask it a question, figure out how it impacts what you're doing today. And sometimes that's a mistake, but as long as you're learning right now, in this space, you're getting ahead. There are no experts in energy AI yet.

Chad Watt:

Great. Rama, your thoughts?

Ramakrishna Manne:

There are a lot of clients doing proof of concepts, a smaller scale. But I think there are foundational things like governance and data that we talked about. Those are things that companies still need to work on to benefit from AI at scale. So I think as we move towards more scaling, that's when doing the hard work and setting your foundations right will help.

Chad Watt:

Rama, I'm going to ask you to say more about that. Getting your data right within an energy firm and an energy trading firm, what's the first step there?

Ramakrishna Manne:

So you need to have traceability. Where did you get this data from? And who curated it? What is the level of confidence I have in leveraging this data? And when you publish your analysis, this is your opinion of the data, not the source data. So when people are using your data to do another analysis, so they know exactly where the data came from and how confident they can be on that data. So knowing that and making sure that if you have gaps in data, let AI fill in the data, but knowing that this data got filled in by AI, having that transparency, having that lineage helps.

Keith Farris:

One of the things that makes energy trading maybe a little bit unique is just the vocabulary. And using energy trading vocabulary that your front office would understand to describe your data is important because then as models learn, as junior people come along, they're using your industry lingo. Too often we translate that data or we don't translate it from the vendor specific terminology. And so it's always some random integer number that nobody really understands.

Use your domain vocabulary and make your models understand that. Label your data that way. It opens your models up to what the rest of the trading world outside of energy does and what your competitors do.

Chad Watt:

How can a trader use AI to capture margin and optimize risk management?

Keith Farris:

It's an important thing. I think traders have to make a lot of decisions with less than perfect information. And so the more they can leverage AI, for example, like Rama said, with multiple scenarios to evaluate what's the impact of weather and changing market prices and changing capacity all at the same time to figure out is this a good strategy or not. It lets them potentially make mistakes on paper rather than in their book, which should lead to faster decision making and better P&L.

Chad Watt:

Keith, let me stay with you for a moment. AI models are only as good as the data that they consume. What types of data are best for generating insights?

Keith Farris:

Yeah, I think data that matters to the business. So for example, weather data, market data where you can kind of decompose that data into pricing data, ford marks, wind, solar, all of those elements become data and then turning that data into something meaningful and being able to cipher out what's not meaningful, right? And kind of to the idea we mentioned earlier about uncorrelated markets, prices of LNG in the US previously had no impact on data centers being built in Asia. Today they do. And so being able to take data that maybe wasn't related, two, three, ten years ago and say now it is or maybe it is, can AI help me find that?

Chad Watt:

Rama, can I ask you to reflect on that? How do you use AI and what's the importance of the AI models?

Ramakrishna Manne:

The AI models, basically you can run several different scenarios and come and the AI can actually come up with the recommendation. So before the trader gets in, before the markets open, they are ready, they already reviewed all of the recommendations, what happened before the market and they are ready to execute the trades. They can make the trades that they think are the best, otherwise it would take some time for them to process all of this and they're losing valuable time in the process. Now AI can help speed up that execution.

Chad Watt:

Keith, what is the most critical first step to ensure a successful value-driven AI implementation?

Keith Farris:

I think to start out you've got to have a definition of success. What is it you're trying to accomplish? How do you know when you're done with that? I think too often we rush in and say, I need to start, but I don't know where I'm headed. And that's a recipe for spending tens of millions of dollars and years chasing something that you couldn't exactly define.

Chad Watt:

Rama, as AI takes on more autonomous roles, how does the role of the human energy trader evolve?

Ramakrishna Manne:

Human is basically doing more value-added work. There still needs to be that accountability for that decision. That accountability still lies with the trader. And what AI is doing is enabling you to be the best yourself so that you can maximize the margin on any trade.

Chad Watt:

What is required to maintain accountability and trust in AI-driven trading? Rama?

Ramakrishna Manne:

So I think proper governance and proper data foundation so that you can have trust in the data. So both are very, very important. So the governance should be based on what are the roles? Who is responsible? What is the framework that allows you to make that decision? And the data, then you need to have the right data so that the foundation of that data is good so that you can make a sound decision.

Keith Farris:

Yeah, I think it's a little bit like high school algebra. You've got to be able to show your work along the way. If you can ask the model, substantiate what you just suggested to me. Show me the steps that you used to get to that. Because now as a trader or as a market risk professional, I can look at that and say, yep, I agree with each step. Or no, I disagree with it. Run the model again with a different view.

Chad Watt:

So AI can't get away with partial credit is what you're saying.

Keith Farris:

No, I think that's the thing you've got to be able to do is say, how do I get partial credit? I'm not going to let you go beginning to end and execute a trade on my behalf, but I will ask you to suggest a trade and tell me why you suggested that trade. And if I agree with all your thesis, then great, let's execute it.

Chad Watt:

Rama and then Keith. If we look five years down the line, how will the day-to-day operations of an energy trading firm be fundamentally different because of AI systems?

Ramakrishna Manne:

If you look down the line, right now you have front office, middle office, and back office. They're like separate systems that do different functions. So there are desks that deal with the customer or the persons, and then all of them, everything else is facing the backend. So you need to think about this as a unified desk with operating along with AI agents. There may be specialist agents like somebody is an agent can be weather forecasting expert. An agent can be a market scenario, market data expert. So different agents that specialize in different things that then able to, there'll be an agent that's able to do all of this analysis and come up with this training suggestion. And then it will really enhance the trader's ability to execute this. They can do more trades. Create more profitable trades with their time.

Chad Watt:

Great. Keith, what did you see in your five-year time machine?

Keith Farris:

The basis of your question, what's operationally going to be different? Nothing. What should be different is the number of people it takes to do that operational process and the number of errors that come out of that process. I think the better your process is, AI is not going to change that. It's going to augment or extend it. So what should happen is it's not that it'll take less humans overall, but those humans ought to be focused on things that add more value.

So how can AI find the flaw in my process, maybe that's getting rid of a reconciliation that happens every month that wasn't necessary. Maybe it's eliminating day one P&L problems that don't show up until month end, right? So it enables the front office to make more and faster decisions and allows the rest of the organization to not have to check it as frequently so that they can do things that actually add more value rather than finding problems.

Chad Watt:

So this is the excitement and promise. You've got the humans doing the valuable work and all the process and standardized practices can be automated and handled in an automated kind of error-free way?

Keith Farris:

The challenge of all of that is quite frankly going to be the change management, right? Getting people comfortable with doing that. It's all possible. AI will do that, but we got to be willing to let it do that and say, great, now what else can I do to supplement or augment that, not fight it?

Chad Watt:

There's AI literacy, there's culture change, there's a lot that goes into this. Rama, do you have any final thought there?

Ramakrishna Manne:

For it to be successful, there's always like we talked about the data debt, right? So the data foundation needs to be correct. The process foundation needs to be correct. And then Keith touched upon the people debt, right? So people or an organization need to be ready for AI to welcome AI coworkers among, to treat, to understand how to relate to them and how to leverage them and be successful with them. And so this is a huge change management effort and organizations that really put emphasis on people and really bring them up to how to work with AI and AI-led processes and agents will be successful.

Chad Watt:

Keith, Rama, thank you so much for your time today.

Keith Farris:

Thanks for having us.

Chad Watt:

We recorded this interview at the Infosys Knowledge Institute Studios in Richardson, Texas. Christine Calhoun and Yulia De Bari are our producers, and Dode Bigley is our production engineer. I'm Chad Watt with the Infosys Knowledge Institute. Until next time, keep learning and keep sharing.