Recommender Systems: A Machine Learning Solution for the Oil and Gas Industry

Editor’s Note: The SCM thesis Reducing Oil Well Downtime with a Machine Learning Recommender System was authored by Jesús Madrid and Andrew Min and supervised by Dr. Cansu Tayaksi ([email protected]). For more information on the research, please contact the thesis supervisor.

The oil and gas industry has seen many ups and downs in the last 20 years. Since 2014 in particular, the industry has been feeling the crunch, as prices are well below all-time highs of over $100 per barrel. Most recently, the Covid-19 pandemic has wiped out a third of the global demand for oil, sending prices below $20 per barrel, and at one point reaching unprecedented negative values.

This capital-intensive industry is plagued by unplanned downtime at its refineries, offshore platforms, and oil wells, resulting in unnecessary high maintenance costs and risk exposure.

In this current low-price environment, oil and gas companies must prioritize operational efficiency. To do so, companies are beginning to lean on advanced analytics utilizing Big Data and machine learning to reduce downtime and its associated costs.

In today’s global and digital world, recommender systems help us make decisions in our everyday lives. Our choice of what product to buy, what movie to watch, and what website to visit can be influenced by the right recommendation.

Recommender systems are present at nearly every step of the purchase process on e-commerce platforms. Companies like Amazon, Netflix, and Google use machine learning algorithms to predict users’ preferences for certain products or services. These predictions are used as recommendations, leading to a better customer experience and a significant increase in revenues.

The implementation of such systems so far has been limited to top-tech companies, but there is no reason why they cannot be implemented by less sophisticated players. So we asked the question, Can the legacy oil and gas industry benefit from the use of recommender systems?

The sponsoring company of this project is Baker Hughes, an energy technology company offering digital solutions to oil and gas companies around the world. It is common for Baker Hughes’s customers to shut down an oil well to perform maintenance and then do it again a few months later for a similar service or product installation. A recommender system can anticipate their customers’ needs and combine products or services together; by purchasing such bundled products and services, customers can reduce the number of times oil facilities need to be serviced.

Using historical sales data, we built a user-based collaborative filtering recommender system powered by machine learning. Our model uses similarities among customers’ purchase behavior to make product and service recommendations. Two customers with similar past purchases are considered “neighbor customers,” and products that have been acquired by one but not the other are used as recommendations.

Initial results proved the system can accurately predict around 20% of all product purchases, resulting in a reduction of unplanned downtime and cost savings of 6.44%, or $2.45 million per customer per year. Incorporating useful customer features, product features, and product data can increase the number of accurate recommendations. Additionally, the use of more complex machine learning algorithms such as matrix factorization and deep learning could result in more robust recommender systems.

It is never too late to embark in the journey of digital innovation. As more sales history data becomes readily available, machine learning recommender systems will only continue to increase in prediction accuracy, allowing companies in the oil and gas industry to benefit from the use of advance analytics to reduce downtime and remain cost efficient.

Every year, around 80 students in the MIT Center for Transportation & Logistics’s (MIT CTL) Master of Supply Chain Management (SCM) program complete approximately 45 one-year research projects. The students are early-career business professionals from multiple countries, with two to 10 years of experience in the industry. Most of the research projects are chosen, sponsored by, and carried out in collaboration with multinational corporations. Joint teams that include MIT SCM students and MIT CTL faculty work on real-world problems. In this series, they summarize a selection of the latest SCM research.

SC
MR