How a customized Machine Learning solution transformed decision making in this Retail Fuels market

Overview of Results

• Fuels margins increased nearly $500 million dollars in 18 months of engagement (establishing new record performances)
• Client gained an enduring, game-changing insight about the price elasticity of their market
• Client market share increased, despite overall market volume weakening at approximately 2 to 3% yearly
• Custom management dashboards created to ensure that business results will be sustainable

When we entered our Assessment phase with this global integrated oil company, we discovered a pricing protocol that trailed daily spot price changes by market area.

We also identified an organization that focused on volume first, earnings second. Their pricing group worked independently from their sales organization, and recognized the need for a more sophisticated approach.

Assessment Phase

The initial critical finding

After analyzing millions of daily sales transactions, we discovered that our client had greatly overestimated market elasticity.

This meant they were consistently underpricing their product, out of concerns over losing volume and market share. The data showed that although some periods did show high degrees of elasticity, those periods were far less frequent than our client had believed.

This crucial finding meant that if we could build a model that could reliably predict periods of low elasticity, our client could adjust daily to those situations. 

We recommended a machine learning (ML) model that would accurately forecast daily volume elasticity at the “ship-to” customer level for 90 days out, for more than 3,800 customers driving nearly $20 billion in annual revenue.

Delivery Phase

Building the Pricing Decision Support system

We followed our systematic Delivery process, building a machine learning (ML) model robust enough to support the new approach.

Our first steps included gathering and consolidating ten years of daily sales transactions down to the “ship-to” level. We also built an initial DataFrame for the AI engineering process. And we formed client teams to examine exogenous data, which enabled us to produce a much richer forecast with even better accuracy.

We then developed machine learning experiments to produce a high-level neural net design. Multiple architectures were used to drive highest-level “out of sample” accuracy. Those experiments were performed in sets, to optimize the balance between training accuracy and testing accuracy. An implementation-grade AI model was launched in a secure location, to evaluate accuracy and performance.

Working with our client teams, we were then able to combine implementation-grade strategies, policies, and processes into an integrated Pricing Decision Support (PDS) System.

The client’s teams were now able to use recommendations from the PDS system to set daily price moves against published spot prices based on sophisticated data analysis rather than guesswork.

Managing for sustained results: PowerBI managerial dashboards

Now we had a machine learning-driven system to recommend daily price points without negatively affecting market elasticity.

But we still needed to “close the loop” with a window on how well client’s team actually followed the system’s advice.

To that end, we built Power BI dashboards that management could access on their computers — and even their phones. These allowed management to see at a glance if their organization were using the recommendations generated by the PDS system. The dashboards also allowed management to track planned performance against actual results.

Follow Up Phase

Retaining the profitability gains we built

In our Follow-Up phase, we continue to train the ML model to benefit from the most current data available. This is particularly pertinent as the client adapts to the “Black Swan” global events surrounding COVID-19.

The dramatic success of the project was greatly facilitated by leadership on our client’s team.  The leads of these teams were individuals who understood the potential of AI, and had a strong commitment to achieving the highest possible margins for his organization. After all, systems are only as good as the people who manage them.

We’re continuing to work shoulder-to-shoulder with this client, to make sure they retain the business advantages they gained in this engagement.

Results summary

Following a key insight about elasticity in this client’s market, it became clear that being able to predict those periods would create a dramatic opportunity for improved profitability.

Using machine learning-driven AI to pinpoint periods of low elasticity allowed this global integrated oil company to increase their fuel margins by nearly $500 million dollars in 18 months of engagement, establishing new record performances.

They were able to go from literally being a follower to deeply understanding the complex dynamics of their market’s supply and demand issues by using robust ML and AI to generate deep insights about their environment. Remarkably, the firm’s increased margins and market share came in a period of diminishing overall market volume of approximately 2-3% yearly.

The client’s commitment to implementing the program effectively also drove a change from a narrow volume-focused vision to a larger, more profitable earnings-focused one without negatively impacting market share. The firm is now well-placed to implement additional projects to drive important profitability gains, even in challenging market conditions.