The Transformation: Building, Operationalizing, and Evolving Predictive Models

Initial Phase: IBM PCI Implementation

Xerago led the end-to-end deployment of IBM Predictive Customer Intelligence (PCI). This included integrating IBM SPSS Modeler with the insurer’s core policy systems, data lakes, CRM, and campaign platforms.

We built the foundational data pipelines, curated AI-ready datasets, and developed a predictive workbench.

To ensure the solution scaled with the business, we architected it on a robust infrastructure, with high-performance server nodes and a dedicated analytics environment.

Data pipelines were orchestrated across an operational data store (ODS), fed by scheduled ETL jobs that harmonized information from policy, customer, product, agent, and call center systems. This foundation, built using best-in-class technologies, enabled us to deliver high reliability and throughput for all predictive workloads.

On top of this foundation, Xerago built a comprehensive suite of behavioral models:

These models powered downstream journeys across channels like SMS, email, agent alerts, and IVR. They became embedded into business logic, from campaign segmentation to escalation triggers.

Ongoing Model Management

Beyond initial implementation, Xerago also took responsibility for managing and refreshing models to ensure continued precision. We periodically restrained them, monitored for drift, and adjusted features as product constructs and customer behavior evolved.

This meant not just deploying algorithms, but building a data science methodology that included detailed data audits, consistency checks, and business-aligned feature engineering.

Over 30 variables (such as agent tenure, payment patterns, policyholder demographics, contact history, and KPIs like complaint resolution time) were derived and validated. Extensive transformations and aggregations ensured that each model drew from a single source of truth, with robust checks to maximize accuracy and minimize bias. For maximum transparency and business alignment, we selected the CHAID decision tree as our lead algorithm, balancing predictive power with explainability for business stakeholders

This helped the models stay relevant, not just technically accurate, but business-aligned.

The Crossroads: IBM PCI Deprecation

When IBM announced the sunset of Predictive Customer Intelligence, the insurer faced a major inflection point. All existing models risked becoming unusable. To avoid operational disruption, Xerago designed and executed a complete transition strategy, replacing IBM PCI with an open, modular decisioning stack.

We reimagined the predictive workbench using open, modular technologies, leveraging Python, R, AutoML, H2O, and Apache Airflow for orchestration. This new stack didn’t just replicate previous capabilities, it offered greater agility and transparency, empowering business users to adapt models faster and deploy new use cases at scale.

The business didn’t just retain its predictive layer, it gained greater control, speed, and extensibility.

Technical Architecture, Data & Deployment Summary

From Scores to Decisions to Impact

Xerago helped ensure that the predictions were translated into real actions across business teams. Model scores were integrated with customer journeys, CRM systems, and marketing automation tools. Customer service agents used the insights to prioritize engagement.

Campaign teams built dynamic segments based on real-time churn and conversion scores. Retention managers used early claims risk to trigger preemptive communication and interventions.

AI that Grew with the Business

This transformation wasn’t just about models, it was about embedding intelligence into the operating fabric of the organization. From initial implementation to platform sunset and beyond, Xerago ensured that predictive intelligence never became shelfware. It evolved with the business, scaled with the data, and delivered measurable value, year after year.