What is Causal?
Causal is an integrated product analytics, feature flags, and A/B testing platform. It is designed to help you write more maintainable code, collect better data, and optimize your product quickly.
Causal is built from the ground up to provide robust, easy to use solutions for the entire feature stack:
- Data governance
- Data infrastructure
- Data schema migration
- Product analytics
- Feature flags
- Experimentation (a/b/n testing, bandits)
- ML feature store
- ML model deployment
- Causal inference (metric impact analysis)
- Retirement of technical debt
Most organizations today use a series of point solutions. These point solutions are not well integrated, are overly complicated, are not reliable, and introduce friction across both technical and organizational boundaries.
All parts of the feature stack are highly interrelated and fundamentally dependent on the structure of data. The key insight that allows Causal to provide elegant solutions across the entire stack is that features need to be properly defined and those definitions need to live in code.
Causal uses GraphQL (with a few extensions) to define a schema for your product features.
This schema resides in your revision control system, so it fits seamlessly in with your existing engineering workflows such as code reviews, branching strategies, and deployment pipelines. Since it lives next to your code, Causal’s versioning matches your versioning.
A product feature's schema consists of:
- Context: the state of the user and application at the time you show the feature
- Outputs: what is being presented to the user
- Events: downstream events that give feedback on how this feature impacts the user
Defining Context and Outputs (in addition to Events), fully specifies the contract between the various producers and consumers of data. Because the contract is fully specified, Causal can act much more intelligently than other systems and provide both broader, but also significantly easier to use solutions.
As a point of contrast, other systems do not properly specify product features and their specifications do not live in code. The result is:
- Feature definitions are scattered across the codebase, product documentation, and project management tickets and comments.
- Tracking plans are out of sync with your versioning, which adds friction and complexity.
- Tools are unable to act intelligently because they lack full context.
- Extensive downstream engineering and coordination efforts are required to stitch together data into a useful format. In addition to being expensive to build and maintain, these efforts are invariably limited and introduce errors.
Causal solves all of these issues. It uses the intelligence your schema supplies to provide best in class tooling and optimize workflows across your organization:
- Software Engineers
- Strongly typed custom APIs that guarantee data accuracy and integrity
- Automatic deprecation for retired features
- Data Engineers
- Automatic ETLs with safe and correct schema migrations
- Automatic marrying of impressions and events
- Product Managers
- Feature viewer, which provides visibility into the past and current state of a feature
- Event viewers to allow teams audit and understand tracking events
- Features that can be configured by non-technical users, including updating values, turning features off and on, and running experiments
- Simple A/B tests, along with more advanced experimentation like bandits
- Metrics evaluation through causal inference, allowing you to understand which metrics drive business outcomes that matter
- Analysts and Data Scientists
- Warehouse tables that are ideally structured for analytics and training ML models (no data wrangling required)
- Metrics with consistent definitions to use across your organization
- Contextual Feature Store with time shifting guarantees
- ML model deployment without front-end engineering
- Ramp traffic to features over time
- Quickly disable features that are broken
Causal can also easily interoperate with existing point solutions. It uses industry standard formats, and can import and export both data and events to and from other platforms.
Causal scales. Feature schemas enable automatic marrying of impression and event data, minimizing the need for joins. Causal uses a hyper optimized data layout, so when joins are needed, they can usually be done entirely in memory, ensuring extremely high performance. Fewer joins and faster joins make Causal tooling very fast, and make it easy for users of data to explore, navigate and create insights.
In summary, Causal is a scalable feature infrastructure platform that provides robust, easy to use solutions for the entire feature stack. It does so by having proper feature definitions, having those definitions live in code, and leveraging the intelligence that those definitions supply to provide best in class tooling and workflows to your entire organization.
Please continue reading to learn how to get started with Causal.