Sound data science management is the foundation for effective fuel pricing.

In 2012, Harvard Business Review published an article entitled, “Data Scientist: The Sexiest Job of the 21stCentury.” While the accuracy of this designation at an organizational level may vary with the organization’s exposure to and dependence upon big data, the insights data scientists can provide to both wholesalers and retailers can ultimately become a “secret ingredient” in the fuel pricing mix.

In this post, we continue our conversation with Ed Cockcroft, head of data science at PDI, for some illumination on the oft-misunderstood discipline of fuel pricing. See Part 1 of this interview for Ed’s background and insights on what makes a good data scientist.

The fuel market is pretty volatile. The fuel industry is quite diverse, and information is constantly evolving. How do you normalize data without losing too much meaningful information? How do you get rid of the noise?

Data scientists have to have a certain degree of business knowledge. The noise typically comes out when you fit a first-generation data model, you naively do that, and you analyze the results. The results may not be great, so you do some diagnostic, investigative plots. You might spot some anomalies. So, you look at the worst anomaly and potentially realize the data isn’t believable from a business perspective. So, you build in more data cleansing, and you run it again.  You can do some naïve removal of outliers, but if you just remove the high and low values, you might be losing your best information.

It takes time to get a good first-time data model for a customer, particularly if they’re in a new market. For example, if someone in a given country is interested in price-volume optimization, my first question is, “Can you tell me a little about that country’s market?” We need that market knowledge is important when we start looking at the data. Otherwise we might not spot something anomalous. it tends to be an iterative process. We’ll build a model. We’ll look at the results. We’ll try to figure out if we can do better. There’s no standard, out-of-the-box approach. You take the data you’ve got, and you restate it in different ways that you know are business-relevant.

As one who has a finger on the pulse of fuel wholesalers and retailers, what should these groups look for in a fuel pricing solution? 

In wholesale, the biggest thing to understand are your margins. After that, you want to understand your margin opportunity and what else is out there in that market. You also have to understand your prospects—the combination of the deals you won and didn’t win. Those are three things over and above just simple price execution.

In retail, price execution is really important because there are fast-moving retail markets. Automation is also important, because it’s part of price execution. Free up your staff to do the hard stuff, and get the machine to do the easy stuff. People are also looking for a competitive advantage in terms of smarter strategy. They’re trying to take advantage of opportunity in the market, be it either identifying patterns in the market and maybe being able to price differently on different days or being able to proactively react to changes in the market. That’s where machine learning can really help to spot those patterns. Speed to market for the active markets and being smart. It’s about price execution and freeing people up to spend the time on making sure that every individual decision is right.

Every site and wholesaler is unique. Cookie-cutter solutions are likely insufficient to support a sophisticated and varied market. How is PDI able to develop site-specific strategies? In what ways are fuel pricing solutions customizable?

The software itself is hugely configurable. On the one hand, we have complete flexibility to model cost margins. That means we can model various types of costs: credit cards, wetstock, loss, temperature effect, however your freight is charged and structured, all sorts of things. That will typically be configured for each customer because people have different requirements. In terms of the actual pricing strategies, the system supports a variety of ways of doing that. We can turn different bits of that on and off. Be it a PVO-driven model or a PVO-motivated, variable, rules-based model. Sometimes it’s driven by a country’s regulations. So, we do run systems with mixed ways of pricing for different products and portions of our customers’ chains. It’s not fixed to be one configuration for the whole system. It’s mix-and-match at the site and product level in terms of how you set the system up.

Fuel price optimization is complicated, and Ed has effectively laid a strong foundation for what one might look for in a fuel pricing data scientist. Not every company is ready or able to hire a full-time data scientist. For many, the wisest decision is to outsource this role. Whatever path your organization takes, the results are worth the investment.

Did You Know: Your Source for PDI News provided by PDI the leader in enterprise management software for the convenience retail and petroleum wholesale markets.