Act-On

Solution

Act-On and Intel built the AI Audience Insights feature as an unsupervised machine-learning pipeline deployed in Act-On’s production environment. Customer profile records synchronized from CRMs and high-volume behavioral signals such as email opens, clicks, page visits, downloads, webinar activity, and form submissions are standardized, cleansed, and aggregated into time-windowed features, then prepared from Snowflake for modeling. Intel Distribution of Modin accelerates data import and handling, while Intel Extension for Scikit-learn speeds supported model fitting on Intel hardware; Intel customized the workflow for predominantly categorical marketing data and used K-Modes clustering with automatic cluster-count search. The containerized system runs on AWS EC2 R7iz with Spring Boot, Spring Cloud Data Flow, and Kubernetes coordinating staging, parallel clustering, and processor tasks. Results are interpreted with statistical and business logic, posted back to the Audience Service, and surfaced in the Act-On UI as recommended audience segments marketers can accept, discard, or rerun with different feature selections.

Data flow

Customer profile records from synced CRM systems and behavioral event histories are stored in Snowflake, where Act-On standardizes fields and aggregates engagement windows. A staging task fetches and preprocesses the dataset, Intel-accelerated clustering jobs run in parallel on AWS EC2 R7iz, and a processor task applies statistical and business logic before posting recommended segments to the Audience Service for display in the AI Audience Insights UI.