Organizations increasingly want faster, more intelligent business decisions. For enterprises running critical workloads on CICS, artificial intelligence (AI) offers a practical way to improve transaction processing, automate decisions, and uncover patterns hidden within operational data.

Rather than sending data off platform for analysis, modern IBM Z systems can run AI models close to the transactions themselves, enabling real-time predictions with minimal latency.

A Real-World Use Case: Insurance Fraud Detection

The Challenge

Fraudulent insurance claims impose costs on insurers and slow legitimate claims processing. Traditional auditing often requires significant manual effort, making fraud prevention difficult to scale.

The Solution

A machine learning model trained to identify fraudulent claims can automatically analyze transactions and flag suspicious activity for review.

The Business Impact

• Reduce manual auditing hours 
• Accelerate claims processing 
• Improve customer satisfaction 
• Redirect staff toward higher-value work

Why Use AI in CICS Applications?

AI systems learn from historical data and use those patterns to make predictions or recommendations. Machine learning models can identify trends that help organizations make more accurate decisions, while deep learning models can uncover complex relationships in large datasets.

Recent IBM Z innovations, including the IBM Integrated Accelerator for AI, the Telum II processor, and the IBM Spyre Accelerator, help organizations run AI workloads efficiently alongside traditional transaction processing.

The result: business applications can incorporate AI-driven insights directly into transaction flows without introducing delays caused by external processing.

Three Common Approaches to AI Integration in CICS

Organizations typically choose an integration approach based on performance requirements, model type, and existing infrastructure.

1. IBM Watson Machine Learning for z/OS (WMLz)

Applications can invoke deployed AI models through:

• EXEC CICS LINK calls 
• REST APIs 

This approach enables online scoring of SparkML, PMML, and ONNX models directly from COBOL applications.

2. IBM Operational Decision Manager (ODM)

ODM rules can incorporate AI predictions generated by WMLz, allowing business rules and machine learning models to work together in the same decision process.

3. Open-Source AI Frameworks

Organizations can also deploy models built with:

• TensorFlow 
• PyTorch 
• Snap ML 

These frameworks can connect to CICS through REST APIs and run on zCX or Linux on IBM Z environments.

Five Steps to Deploy AI in a CICS Environment

Step 1: Collect and Prepare Data

Every successful AI initiative begins with quality data. Gather relevant information from operational systems, then clean, structure, and standardize it before training begins.

Step 2: Build and Train the Model

Use historical data to train a machine learning model capable of recognizing patterns and making predictions. Depending on the use case, organizations may use anything from simple regression models to deep neural networks.

Step 3: Validate and Test

Evaluate the model against previously unseen data to verify accuracy and ensure it performs reliably in real-world scenarios.

Step 4: Deploy for Real-Time Predictions

Once validated, deploy the model into production where it can analyze live transactions and provide immediate recommendations or predictions.

Step 5: Continuously Improve

AI models should evolve as new data becomes available. Ongoing retraining helps maintain accuracy and keeps predictions aligned with changing business conditions.

Industry Applications

Banking

AI can support:

• Real-time fraud detection
• Credit risk assessment
• Personalized customer interactions

These capabilities help improve operational efficiency while enhancing customer experiences.

Insurance

Predictive analytics can improve:

• Claims processing
• Risk assessment
• Policy pricing
• Underwriting accuracy

The result is faster service and better decision-making.

Retail

Retail organizations can use AI to:

• Forecast demand
• Optimize inventory
• Personalize marketing
• Support dynamic pricing

These capabilities can improve both revenue and operational performance.

The Bottom Line

AI is no longer limited to cloud-native applications. Modern IBM Z platforms allow organizations to bring predictive analytics directly into CICS transaction processing, enabling faster decisions, improved efficiency, and more responsive customer experiences.

For organizations already running mission-critical workloads on CICS, AI represents an opportunity to make those transactions smarter—not by moving data elsewhere, but by infusing intelligence directly into the systems that already run the business.

Next Steps

Looking for more CICS content? Check out the CICS virtual workgroups. The next presentation is June 11. Register today.