This page describes what Cortex Fabric Profiles are and how they work. First read the Introduction to Fabric if you are familiarizing yourself with the Fabric toolset for the first time.
For details about how to build Profile schemas go to Build Schemas in the Build Profiles section.
For information and links to the Cortex Profiles SDK go to Cortex Profiles SDK
Introduction to Profiles
This section provides an introduction to Profiles including what they are, who uses them, and how they work.
The Profiles feature allows users to create versioned profiles that represent entities such as people, roles, or organizations with attributes and schemas for use in agent processing.
The Profiles tool suite includes a graphical interface in the Console, CLI commands, and a Python Library that Cortex Fabric users can employ to create a focused set of attributes and schemas for an entity that can then be applied to AI applications serving entity-specific results (e.g. suggestion engines).
To set up Profiles you must first configure the Data Sources, data stores from which the schemas and attributes are derived, integrated, and managed. Data Sources provide the data input for Profiles.
What are Profiles?
A capability that:
- Stores and visualizes the attributes and relationships of entities and attributes
- Contains attributes that are declared, observed, or inferred about an entity
- Is temporal in nature; attributes are fluid (added, removed, changed as new data is available)
- Provides personalized insights about entities and attributes through learning and interactions
- Enables business outcomes and KPIs using metrics
- Relies on data, models, analytics, and a set of Cortex capabilities to provide insights
Profiles are not:
- Just a database with schemas and attributes
- Master Data Management software
Profiles feature description
The Cortex Fabric Profiles feature allows users to:
- Build profiles for entities
- Consume entity profiles to generate insights
- Learn about the profile entities in the process
The primary goal of building Profiles is to learn as much as possible about the entities being profiled.
The secondary goal of Profiles is to generate the best possible insights for profiled entities.
- Entity: A person, group, organization (or whatever) being profiled
- Profile: A consolidation of different pieces of information about a specific entity at different points in time
- Attributes: The different pieces of information captured within profiles
- Declared Attributes: Attributes that are set for an entity (e.g. Name, Identifiers, Age)
- Inferred Attributes: Attributes that are mathematically or algorithmically calculated about the entity (e.g. preferred television genres based on past views)
- Observed Attributes: Information gathered based on the users interaction with the application (e.g. home location selection, product selections)
- Assigned Attributes: Attributes that are assigned to the entity by the profile builder. (e.g. Blue Ribbon Customer)
- Schemas: A set of attributes used to model a class of profiles.
- Entity Events: Updates to the profile attributes and attribute values.
- Versions: Every modification to a profile's attributes results in a new version of the profile. Versions help track how entities evolve over time.
- Event-based modeling: Backed by an event sourcing model
- Schema: Records facts, interactions and insights about an “Entity”
- Versioning: Provides versioned views of entities and attribute values over time
- Feature store: Supports training and inferences based on new ML Models
- Auditing: Tracks changes for audit and compliance purposes
- Visualization: Entity and attribute relationships can be visualized in Cortex Fabric or other BI tools
- Outcome tracking: Helps track business KPIs
Who might use Cortex Profiles?
- Architects and Developers can use Profiles to help create a catalog of attributes that can be populated for entities of a specific type, and organize the attributes for said entities. This is done by creating a Profile Schema.
- Interested Stakeholders can use the Profiles interface to view attributes of a specific profile.
- Developers can use either the
cortex-python-profileslibraries or the Cortex CLI to programmatically create profile schemas, load profiles into the system, and download profiles.
Profiles value proposition
Business users are able to:
Explore relationships and data visualizations that are integral to generating the best possible insights for more focussed marketing or services
Technical users are able to:
- Generate normalized representations of the many attributes that can be associated with profiles.
- Identify list the differences between profiles
- Classify and cluster profiles by attribute with minimal configuration
- Simulate evolution of profiles based on historic trends
- Infer attributes for certain profiles based on similar profiles
- Observe how different profiles are being used
- Identify patterns in the profiles
Profiles use cases
Example use cases for Profiles are broken down into categories below.
Predicting behavior based on one or more classes that the entities fit into
- Customer service experience
- Service enrollment
- Invoice optimization
- Data retention and compliance
- Property management
- Risk management
Example: Predicting the likelihood of people charged for certain medical procedures calling customer support about the billing.
Providing better recommendations based on past behavior and preferences
- Product recommendations
- Intelligent interventions
- Action recommendations
Example: Provide movie recommendations based on user preferences and past viewing history
Identifying similarities between entities
- Fraud detection
- Profile similarity and differences
Example: Identifying fraudulent insurance claims based on similar patterns of information in claims forms.
Identifying and grouping similar entities
- Targeted marketing
- Profile segmentation
Example: Identifying profiles of individuals who have recently purchased homes or vehicles to target for insurance marketing.
Forecasting and Simulation
Forecasting the behavior or outcomes of entities or simulating ”What if” scenarios using synthetic data.
- Early warning systems
- Disaster simulation and forecasting
- Financial forecasting
Example: Time series forecasting of the number of new admissions of Covid-19 patients.
Generating insights and evidence-based learnings about an entity
- Customer insights
- Event insights
Example: Detecting agent fraud based on prior transactions and policy enrollments
Powering different ML model training and inference
Example: Classification models for fraud, waste, and abuse in medical claims