CGI set up a energy trading and risk practice to align our resources across the globe and asked me to lead it. We had multiple projects and talented people around the world delivering energy trading solutions. We have a huge finance practice as well so we have been able to leverage their insights as well from block chain, regulatory compliance to setting up entire markets.
When did you first learn about prescriptive analytics? What got you interested in it?
Funny thing. Early on in my career I realized that prescriptive analytics could be applied to energy trading — I just didn’t know the term for it. One of the projects I called NBT (Next Big Thing), because I sincerely thought we were going to revolutionize energy trading. I would read up on genetic algorithms and other areas of metaheuristics to figure out how to do this. I was involved in several projects where we attempted to build an optimization engine from scratch. It was painful. Everyone saw the possibilities but at the end of the day three things would sink us:
1) Energy markets change all the time.
Your application better be able to handle acquisitions, divestments, new markets, new regulations, new prices…
If you need a developer to spend a week responding to the change: Forget it. Commodity operations don’t have that time and they will revert back to spreadsheets.
2) Managing multiple data sets so that the model reflected reality.
You only need one data set to be wrong and the credibility of the model evaporates. Energy traders are dealing with millions of dollars. These models can’t be right “most of the time”. They have to be right “all the time”. Energy trading is the only industry I have been in where everything has to be perfect every day.
3) Application design.
We never thought of constraint based optimization. We did everything else. Constraint based optimization helps find the TRUE optimizations. It helps to keep your current biases out of the model. CBO has been the model that adopts well to changes in physical and financial energy markets.
We have been watching the field of prescriptive analytics emerge and started to develop proof of concepts from several vendors but their solutions would only be viable if our business clients could write R and Python. We tried to explain to them that business users want to run optimizations and simulations themselves and make adjustments to the models on the fly, because that’s what they are used to with Excel. A lot of people in trading are reacting to market events or client situations.
How would you describe prescriptive analytics and its value to someone (in Energy/Oil & Gas) who doesn't have a Data Science background?
It is like building a GPS for your business.
You know where you are. You know where you need to get to. The question that everyone struggles with is:
What is the best way to get there?
Energy trading is like trying to get from your home to the airport in record time. You have a lot of dynamic data sets and constraints to correlate in order to determine the best path. And based on the time of day and year and events...the most optimal path at 5:00 am…may not be the optimal path at 5:00 pm.
How do we expect originators, risk managers and traders to make the most optimal decisions without a GPS in a very dynamic market? Currently most people in trading operations are using Excel to optimize their decision making. What is the best way to position your company to take advantage of these changes and forecasts? What if a new well comes online? What if we purchase new assets? What if a large oil producing country implodes? How do we balance the emergence of renewables with hydrocarbon electricity? What if we are changing from Brent index to WTI or a LOOP based index due to the ability to now export crude oil from the US?
The optimal decision today…might not be the optimal decision tomorrow. The world keeps changing. It’s annoying. Luckily, there is prescriptive analytics that keeps track of all those changes and recalibrates the situation each time you run the model. You can actually set it up to have the model compare itself to what’s actually happening.