Building and Deploying Models at Speed and Scale
Mach One is an advanced web platform designed for Ford Motor Company to improve the workflow of its data scientists. By focusing on reusability, discoverability, and user-friendliness, Mach One enables data scientists to concentrate on solving critical data problems, thereby boosting customer satisfaction and revenue.
Problem
How might we help data scientists in reducing friction, and streamlining processes. So that they can focus on solving data problems and growing Ford’s customer satisfaction and revenue.
Goals
1
Create a structured place for data to live.
— Defined
— Reusable
— Shareable
2
Build human-centered processes.
— Creating a feature
— Building a feature set
— Training data
3
Integrate complex flows into one platform.
— Build internal tools
— Integrate with external tools
Creating Focus
Visualization. A user-friendly layer of UI helps users, from novices to experts, visualize workflows intuitively.
Guidance. Process indicators and simplified language assist novice users, while providing guardrails for advanced users.
Discovery
Clarity in data relationships. The platform displays data relationships clearly, ensuring that complex data points like CreditScore are meaningful in context, e.g., Vehicle > CurrentOwner > CreditScore"
Nested information trees. Clean, nested information trees allow for quick visualization of relationships and feature discovery.
Feature Sets
Data shopping cart. Feature Sets act as blueprints for generating data, making abstract data points like VIN andodometer_change meaningful within user-defined archetypes.
Data Sets
Experimenting with real data. Users can instantiate Feature Sets by querying the database and can repeat this process to test different variations.
Projects & Experiments
Organized problem solving. Projects serve as containers for experiments, helping users stay focused on specific customer problems.
Experimentation. Experiments represent the process of training data sets, encapsulating the tools and methods used by data scientists.
Impact
Mach One has significantly transformed the workflow of Ford’s data scientists, delivering notable improvements:
1
Greatly improving time to market.
— Reduced engineering time from 3-6 months to less than a week.
2
Discovering data more reliably.
— Simplified access to and discovery of data sources.
3
Training data faster.
— Enabled faster training of feature sets, drastically cutting down time requirements.
Conclusion
By enhancing clarity, organization, and efficiency, Mach One empowers Ford's data scientists to focus on their core objective—training data to inform better decisions—resulting in improved customer satisfaction and increased revenue for the enterprise.