The MPPA design focuses on scalability and flexibility. The architecture processes large and diverse volumes of data while being flexible enough to use different learning patterns and business processes. Data is acquired by various sources and stored in a master database inside our process-mining tool.

Following acquisition, data is processed and prepared before beginning ML model development. After that, two pipelines are created: Training, which is used to retrain the model when needed and produce results, and Production, used at a specific cadence (1 x 24 hours) to generate predictions and recommendations.

Finally, the MPPA predictions and Next Best Actions (NBAs) are distributed to our end users using email alerts and CRM tool integration. A monitoring system is also set in place to measure model performance and data quality.

Figure 3.              MPPA model architecture