Freeport-McMoRan (Freeport) has a reputation as a savvy operator in the mining industry. The company operates a fleet of relatively mature, large-scale copper mines in the Americas. Its performance is tightly connected to global copper prices: in a high-price environment, the mines generate significant cash, but at the bottom of the price cycle, some mines struggle to break even. The company’s expectations for growth required significant capital and lengthy permitting and construction efforts. Seeking another path, Freeport turned to AI to see if it was possible to get more out of its existing assets.
Over a five-year AI journey, the company successfully designed and executed what it called the “Americas’ Concentrator” program, aiming to unlock the equivalent of an entire new processing facility’s worth of incremental annual copper production through the use of big data, AI, and agile working methods. No new capital deployment was required.
Such an ambitious program required full-fledged commitment from the leadership team. The head of Freeport North American operations was convinced that the company would need to evolve to survive and thrive and wanted to learn from cutting-edge practices used in other industries. This “continuous improvement” leader drove the team to be as ambitious as possible. The chief information and innovation officer had the foresight to establish a common data infrastructure and architecture to support all processing operations, as well as to enable swift deployment of AI tools across sites with modest tailoring. This allowed much of the site-level focus to be on agile practices, training, capability building, and change management. And the CEO and CFO championed the program to external audiences, energizing and buoying the team as it pushed forward with the effort.
To get started, Freeport selected a mature mine, with an enthusiastic and entrepreneurial general manager, as its test case for the AI transformation program. By demonstrating the value of AI in Bagdad, Arizona, the company sought to learn how machine learning (ML)/AI could enhance its existing systems.
Over the course of about six months, a small team of metallurgists, site operators, and engineers worked to develop and train an AI model to recommend changes in settings to safely increase the mill processing rate. After sufficient testing and development, the operators ran the model and deployed the AI-generated recommendations. During the next several months, copper production increased by 5 percent. In one quarter, the Bagdad site’s throughput exceeded 85,000 metric tons of ore per day—10 percent more than the previous quarter—while its copper recovery rate rose by one percentage point and its operations became more stable. Improving throughput and recovery is an elusive goal in metallurgical processing, and Freeport achieved this in an asset that had been in operation for more than 50 years. The gains allowed Freeport’s leaders to cut by half the capital that they had planned to spend on a series of improvements.
In one quarter, the Bagdad site’s throughput exceeded 85,000 metric tons of ore per day—10 percent more than the previous quarter.
The company’s leadership recognized that scaling the potential of ML/AI across its mines in the Americas could unlock a systemwide production increase of 125,000 metric tons per day, which could yield 200 million pounds of copper per year, representing $350 million to $500 million in EBITDA.1 This would be comparable to bringing a new concentrator on line (a concentrator mills rocks containing around 0.4 percent copper into a fine-ground mix of 25 percent copper and 75 percent rock) but without spending $2 billion or waiting the eight to ten years that such major capital projects typically require.
With leadership aligned on the opportunity, Freeport launched the Americas’ Concentrator program to roll out the AI capability to its mines. The key challenge in this effort was to industrialize the capabilities developed at the Bagdad site so they could be scaled. Freeport had a strong understanding of where to focus, based on a recently completed operating-performance benchmark. The company also had a head start on data; it had previously standardized data on mine performance measurement and reporting. It had enriched the data by installing additional network equipment and performance sensors on the company’s trucks, shovels, and stationary machines. The company also built a central data warehouse to store the data, allowing it to capture and correlate second-by-second performance readings in real time.
Developing AI models at pace required a change in how the company worked. A culture of planning and development built around a set of safeguards had served the company well, but it had its drawbacks, chiefly with pace. For the Bagdad AI pilot, the mine shifted to an operating model that emphasized agility, continuous improvement, and quick, low-risk tests without compromising safety. The key to the success of this change was assembling a cross-functional mix of experts from the mine and a central data science group that could evaluate and execute on change initiatives.
Company leaders made the crucial decision to add metallurgists and plant operators to the development team at every site. When each new set of recommendations came out during the testing phase, the AI developers, operators, and metallurgists on the team would assess the recommendations: Why were they made? Did they make sense? Would they work? In this way, the teams uncovered flaws that the AI developers then quickly fixed, which in turn helped the agile team learn more quickly. The team trained the AI tool and, in doing so, increased the metallurgists’ and operators’ trust in it. By the time the new tool was fully ready, they were much more willing to use it.
The new AI model and interactions facilitated a dialogue and deeper understanding of the process among operators and metallurgists. The initial team had developed an ML model that they called “TROI” (throughput, recovery, optimization, and intelligence). This product helped predict how the processing plant would behave and how much copper could be recovered under any set of conditions. The optimization algorithm evolved settings that would produce the most copper given a particular type of ore and issued recommendations every one to three hours, depending on the operation.
To make TROI work at other sites, however, Freeport had to “assetize” the models. That essentially meant refactoring and repackaging them for easy adoption at other plants. The modular way that the tool was built allowed 60 percent of its core code to be reused easily, while the remaining 40 percent would be customized for the new site. To further simplify these localization efforts, the company invested in developing a centralized code base that site-specific modules could call on rather than having to re-create the necessary code for each specific module.
Running and scaling these models efficiently was possible because Freeport had migrated its data architecture to the cloud. The company was able to further take advantage of the cloud by automating many processes, such as developing the data pipeline, which had previously been a laborious process of pulling data from dozens of manually updated spreadsheets.
As the company’s agile teams proliferated, management of the overall process had to evolve. With multiple agile teams running in parallel, for example, obtaining resources became difficult. Freeport solved this issue by putting a senior product manager in charge to help coordinate teams and improve allocations. It assigned a finance director to manage impact tracking and reporting and to help sites manage their funding requests and progress measurement. And finally, it instituted a quarterly planning system (similar to quarterly business reviews) in which top leaders from the company came together to set objectives and key results and to focus resources on high-priority areas.
With a kitchen-tested transformation recipe in hand and most of the Americas’ Concentrator program vision achieved, Freeport then turned to other areas of its business where these AI capabilities could be applied. In multiple domains, including capital-project execution, maintenance, and leaching operations, the company is now using new iterations of the playbook that made the Americas’ Concentrator program a success.