Causergy —  Integrating Causal AI

Causergy is working with US gas generation. We want to make the systems that are already in place better. We can offer entry products in the form of a causal diagnostics report to gas generation plants that could unlock $300-800k/report for a 500 MW plant without them needing to share any data at all (we have approximately 20M data points per plant).

Causal AI does three things for operators; one is it has a very good ability to filter out signal versus noise. In SCADA systems and predictive AI Suites, there are a lot of alarms that go off  — on average 83 an hour, which is unmanageable for anyone. 80% of these are just noise. The first thing that Causal AI can do is single out the alarms that actually matter for operators, and which they can de-prioritize.

​

Then the second layer is the context layer. Causal AI can provide what are the underlying causal factors that make this problem occur.

The third layer is on the decision side — what can I do as an operator to make sure that I can solve this underlying problem? What are the estimated effects, both on the financial and, on the environmental side? 

We are targeting power generation because it is a sector where there has been historically a lot of correlation versus causation problems that cause these alarms. That's how decision-making is done in terms of operations. We are able to distinguish that they are correlated, but it doesn't mean correlation equals causation. We can also bring much better decision-making in that sense. We believe that what ChatGPT is now for knowledge work, causal AI will be for decision-making in industrial operations, and we are the first one here to apply it to power plants.

That you need a lot of prior energy experience to be able to innovate in energy. I had prior innovation experience in other regulated industries such as life sciences and healthcare that I was able to transfer to a new context that gets combined with domain knowledge in gas generation. 

A challenge has been to try to explain causal AI to utility operators that have barely heard even of more basic AI. We were initially focused on how the technology works, and where it comes from, econometrics, and causal relationships. But the more we interacted, we realized that we probably shouldn't talk too much about the underlying technology. It's more about what type of value or outcomes do you actually, as an operator, get from doing this (up to $5-6M/y for a typical 500 MW CCGT). They care less about the nitty-gritty details of the underlying technology. 

​

(FYI, a great intro to causal AI here.)

Predictive AI for power generation is $11 billion per year, but I think that's a flawed market, given the alarm fatigue and that you don't really know why something is happening and what to do about it. I think creating a new market which is more proper causal decision-making in power for utilities would be a trillion dollar market. So if we improve decision-making, it would have huge impact.

Our long-term vision is to be the causal AI layer for utilities worldwide. That means that we're now starting in U.S. gas generation due to the data availability (Causergy has created a proprietary dataset of 7B+ data points spanning 850 US gas plants over 25 years) but we are able to scale geographically to other parts of the world, we can scale horizontally to other energy sources inside the utilities, and also, vertically, so we can go from power generation to dispatch transmission grid, etc.