Paper presented at the SPE Annual Technical Conference and Exhibition, San Antonio, Texas, USA, October 2023.
Paper Number: SPE-215132-MS
Developing artificial intelligence (AI)-based drilling advisory software is generally straightforward when good quality labeled data are available. However, deploying such systems in the field for use by a rig crew requires careful planning and execution and often fails to provide the value proposed. It is statistically estimated that most AI projects fail, and that most companies that trial AI solutions report minimal to no impact from AI. This paper details the successful deployment and ongoing success of an AI-based drilling advisory system on rigs across an Operator’s fleet, as well as deployment decisions that helped make it a high-value, sustainable, and successful program.
The Operator developing and deploying the AI system focused on five main aspects to anchor this project: setting a realistic long-term vision for automation, choosing the right tools and techniques, implementing a targeted change management plan, careful selection of team members, and planning for sustained management support. For the longer-term automation vision, decisions on where to deploy the AI models – at the rig or managed from the central office, what parts of the solution to develop in-house or out-source to achieve cost objectives, and how soon to scale AI to all rigs in the fleet were key. Finally, a thoughtful change management plan was implemented taking into consideration the company culture and industry best practices.
The project launched in 2015, with the decision to deploy the AI models at the edge/rig site with an ability to push updates from a remote, central support group as needed. The AI model platform was outsourced; AI models were developed /validated one model at a time, and then deployed to all the rigs as soon as possible. The platform and models were modularized to enable rapid prototyping, field deployment, and iterative change. A key Program Sponsor along with other Stakeholders were identified for each rig, and carefully managed to ensure ongoing support, successful adoption, and regular feedback. Transparency on how the model performed calculations was shared readily to ensure acceptance of the results by the drilling engineers and the rig site crew. An agile development and deployment cycle was adopted to maintain rig crew interest to continuously use and improve the system. Over the past eight (8) years, more than ten (10) AI models have been added incrementally to the rig-based system, which has enabled a 10% improvement in drilling performance year over year.
This paper details the decisions and processes that resulted in the successful deployment of an AI-based drilling advisory system for rigs in North America and Europe. The learning and insights from this multi-year (8 years and ongoing) deployment should provide valuable insights to those planning to deploy AI software at scale, at the edge.