Energy AI Consulting

Mosaic brings cutting-edge energy AI consulting to firms all across the energy spectrum, whether it be exploration and development of oil or gas reserves, oil and gas drilling and refining, or integrated power utility companies including renewable energy and coal.

Data science and decision support is nothing new to the energy industry. Many industry giants have long relied on large quantities of data to make decisions. Oil and gas producers now capture more detailed data in real time at lower costs and from previously inaccessible areas, to improve oilfield and plant performance. Utility providers are able to mine data to improve the customer experience, increase operational efficiencies, and maximize revenues.

Every part of an organization can be optimized with data science. Upstream producers and asset managers have many decisions to make, whether it be which field to drill into or when to turn the drills off, or whether to run scheduled maintenance at a specific time. Using predictive analytics helps oil producers and energy providers save significant dollars on the bottom line and increase operational efficiency.

Mosaic can provide expert energy AI consulting to all sectors of the energy world. Mosaic’s unique blend of machine learning, data engineering, and analytic expertise helps us collaborate to deliver immediate value. Our data scientists have experience working for both utilities and oil companies, but they can also bring techniques and expertise from different verticals, truly designing and deploying innovative, yet appropriate solutions.

Leading Oil & Gas Company – Fuel Price Forecasting

  • Mosaic, a leading oil & gas data science consulting firm, built automated prediction models to support forecasting process, allowing advisers to make better informed downstream decisions.
    • Manual, time-consuming processes were predominant.
    • Our data science consultants designed data infrastructure to support predictive models that can process hundreds of variables.
  • Machine learning models evaluated to maximize model performance for each of forecasted target prices.
    • ARIMA, Random Forest, gradient boosted (GBM) models.
    • Reduced forecasting error by more than 50%, increased productivity.

Link to full case study.

Leading Midwestern Regional Utility – Predicting Customer Churn

  • Objective: identify leading indicators of customer attrition and predicting customers at risk of leaving.
    • Mosaic utilized data mining techniques to identify the most statistically significant factors that lead to customer attrition.
  • The predictive model was developed to consider the business value of being able to predict the target variable, the feasibility of developing a sufficiently accurate prediction based on insights from the data mining phase of work.
    • Resulting model identifies customers with highest risk of leaving, along with explanations as to probable causes, to allow customer service intervention.
  • Model built in R and deployed to interact with Tableau for decision support.

World’s Largest Oil & Gas Manufacturer and Distributor – Preventative Fraud Analytics

  • Accurately quantify expected fuel consumption of marine vessels in company’s fleet.
  • The business needed innovative energy AI consulting experts to come in and provide immediate value.
  • Machine learning model estimates daily fuel consumption for over 1,000 ships based on activity reports from captains’ logs, position and speed records, and other data.
    • The model allows analysts to quickly compare expected and reported fuel consumption to identify cases where they diverge significantly, potentially indicating fraud or maintenance issues.
  • Result: Within a week of announcing that behaviors are now predicted and measured, reported consumption from fleet captains in one geographic area decreased by 50%.

Link to full case study.

World’s Largest Oil & Gas Manufacturer and Distributor – Terminal Imbalance Prediction

  • Objective: Reduce instances of excess and insufficient product (gasoline and diesel) at pipeline terminal locations.
    • Predictive machine learning model looking 15 days into the future.
  • The model allows distribution decision makers to have current situational awareness of inventory risks across the distribution network, improving productivity and facilitating quicker and better decision making.
  • Combines a time series model and classification trees to predict the risk of high/low imbalance for each product at each of about a hundred distribution terminals in the continental US.  Predictions are generated multiple times each day with the most up-to-date data and presented to schedulers in a simple dashboard that fits into their workflow process.

Link to full case study.

Leading Northeastern Regional Utility – Competitive Transmission Build

  • Use analytics to identify future congestion problems in the transmission grid, proposing projects to regional transmission organizations, and bidding on the execution of those projects.
    • Mosaic designed a predictive analysis model to provide data insights on whether they should pursue various network builds.
  •  Mosaic helped this utility automate a very manual process, complete with custom built-in analytics.

Leading Midwestern Regional Utility – Workload Planning Optimization

  • Objective: predict when customers will turn on their natural gas during colder months
  • The machine learning model pulls data from weather, energy usage, historic light up requests/jobs, and other sources to validate hypotheses, identify patterns/clusters in the data, and defines workload planning processes for the light up season.
    • Mosaic, energy AI consulting experts, helped deploy the production model, delivering predictions on the volume of turn-on requests during the light up season based upon weather patterns and other parameters.
    • Throughout the project, Mosaic provided data science and advanced analytics mentoring to a younger data science team across the analytics lifecycle.
  • Resulting model provided 5x increase in lead time for crew scheduling.
  • Model built in R and deployed to interact with Tableau for decision support.