Our client’s existing process had previously been handled through Excel and VBA. This relied on many manual interventions and a time investment from Subject Matter Experts to build forecasts.
It created a very slow and user-heavy process to generate even a single forecast for a field. It was criticised for providing insufficient insight into the most important measurements to deriving quality of the field. And it was was unsympathetic to small changes in those measurements too.
Modern Data Science techniques and a cloud-native architecture to provide rapid business value.
By using Python and a range of Dimensionality Reduction and Regression tools – such as Principal Component Analysis and Random Forests – we validated a new forecasting algorithm against known historic data.
This algorithm was demonstrably better than the existing process.
In the time taken for an analyst to execute a single forecast, approximately 1,000 Monte Carlo simulations aggregated the most likely value, and distribution of values.
We built a Cloud Native solution using Azure ML Studio to create the model and corresponding Web-services. A web application Front-end was developed for analysts, with a corresponding Back-end that can scale up and down dynamically.
Our solution reduced the execution time from several hours (including manual processes) for a single run, to under one minute to generate 1,000 runs. Crucially, it also provides a unified and versionable tool that ensures consistency across the company.
With the Cloud Native architecture of the application, the business has now developed the capability to substantially expand the number of oil fields it is considering for purchase or disposal. It has also improved reliability, and has enabled a team of over 250 analysts to focus on the revenue-enhancing aspects of their roles.