Applying Leading-Edge Data Science to Push the Bounds of Rocket Science

Data Science at Lockheed Martin

Lockheed Martin is at the forefront of innovation, applying data science to protect our world and explore new frontiers. Their 110,000 employees build advanced technology systems, products and services to help clients in the U.S. Department of Defense (DoD) and international allies complete their missions. To that end, the company leverages artificial intelligence/machine learning (AI/ML) in every area of its organization—from mitigating supply chain risks to analyzing manufacturing defects, reducing unplanned production outages, and building AI-powered pilots.

In alignment with the DoD’s DevSecOps technology strategy, Lockheed Martin is investing in technologies that facilitate rapid development of secure and resilient AI solutions at scale. This is part of Lockheed Martin’s enterprise-wide digital transformation to support customer missions with greater speed and agility.

Lockheed Martin’s investment in data science yields $20 million in value each year and they’ve recognized 8x return on investment from their data science platform. Their competence in AI and data science means data scientists deliver new innovations faster than they could before. For example, Lockheed Martin’s data science prowess allowed them to build an AI-enabled pilot for the Defense Advanced Research Projects Agency’s (DARPA’s) AlphaDogfight competition2, where they were named the runner up based on their AI-enabled pilot’s ability to outmaneuver a human pilot in simulated air combat.

Challenge

When Lockheed Martin set out on their digital transformation journey, they identified siloed efforts across the firm’s broad portfolio. With data scientists working in silos, they were individually struggling with many of the same problems and delays, such as gaining access to the infrastructure (including GPUs) and tools they needed to perform research and model development. Lockheed Martin recognized that establishing an enterprise data science strategy would bring standardization to their people, processes and technologies, ultimately leading to efficiency and productivity gains.

The biggest challenge was creating an organizational structure to align access to tools, technologies and knowledge management. They needed to address the following pain points:

• New data science employees were spending an industry average of five weeks onboarding, delaying productivity.
• Data scientists spent 65% of their time on data engineering and administration of various development environments, rather than model development, which is the true value of their work and fuels innovation for the company.
• Data scientists had limited visibility into what others were working on or what they’d already built, resulting in duplicate efforts.
• It took weeks to get each new model or application into production once built.

Solution

The solution was to strategically partner with leaders in AI/ML to ensure Lockheed Martin remained in a position to deliver ground-breaking innovations with speed and at scale. Longtime partnerships with MathWorks and NVIDIA already allowed Lockheed Martin data scientists to develop models and access high-powered computing... but only if/when they could get access.

They set out to find a data science platform that would:

• Leverage container-first infrastructure
• Avoid software lock-in
• Automate manual activities
• Foster cross-functional collaboration

Domino Data Lab met Lockheed Martin’s criteria. The industrial-grade enterprise data science platform aligned with Lockheed Martin’s technology strategy. The data science platform centralized tooling across the enterprise for streamlined collaboration and knowledge sharing, and automated manual DevOps tasks that had hindered data scientists’ productivity. Collaborating with Domino helped maximize the productivity and output of Lockheed Martin’s data science teams.

In addition to implementing Domino’s data science platform, Lockheed Martin led Domino to partner with MathWorks and NVIDIA for a three-pronged solution that offered a holistic, best-in-class environment for data science.

Cross-functional teams are now unified in the open, hybrid cloud platform. Data scientists quickly onboard and independently access shared tools and compute resources in containerized environments.

Impact: Cost savings: $20+ million

Lockheed Martin has recognized over $20 million in total annual cost savings from supporting 300 data scientists on its centralized data science platform (Domino), and those savings will incrementally grow as Lockheed continues to support additional users.

These cost savings are attributed to:

• $16 million in data science efficiency savings.
  • Data scientists are 10x more productive, thanks to self-serve access to resources (including NVIDIA GPUs) and automatic reproducibility of model development work. This adds up to thousands of hours of data scientists’ time over the course of a year. Data scientists have capacity to help the business pursue and win new contracts.
  • $0.7 million of the data science efficiency savings are directly tied to data scientist onboarding and offboarding. Today, new hires are productive in a day vs. the industry average of five-plus weeks, because they can easily access the tools they prefer right within Domino. Lockheed Martin estimated that the average employee saves nine days during onboarding.
• $4.7M in IT savings through automation and centralization of hundreds of users who have migrated to the platform so far. Over 90% of DevOps engineers previously assigned to supporting data science workflows have been able to take on new or different business-critical projects because data scientists can independently get access to the tools and infrastructure they need. Automated deployments make it easier to manage the deployment environment, translating into a savings of nearly $100,000 per app in many cases.

Additional capacity to innovate and create revenue: Priceless

Beyond cost savings, the business value of Lockheed Martin’s data science platform investment increases exponentially because of the additional data science capacity. Data scientists can do more. They’ve published 200 web apps. They have bandwidth to solve groundbreaking use cases such as building the aforementioned AI-enabled pilot, improving target recognition capabilities, and creating AI-equipped space capsules3. They’ve also streamlined the development and deployment of deep learning models that mitigate supply chain risk, analyze manufacturing defects and predict maintenance needs.

Lockheed Martin is poised for long-term success. Their data science system is built for scale. They’re "up-leveling" data analysts, software engineers, data engineers and business analysts to become expert data scientists by making the work of the experts more accessible to those learning the discipline. The environment fosters seamless collaboration in a remote-first world spanning teams, lines of business, personas and geographies.