Contextualizing Knowledge for Minerva-AI

Contextualizing knowledge for Minerva AI bridges the critical gap between raw process data and meaningful business insight. While process mining reveals what happened, true root cause understanding requires integrating contextual knowledge—ranging from case-level attributes to external geopolitical influences. By systematically capturing both static (“at-rest”) and dynamic (“in-operations”) context, organizations can empower GenAI-driven analysis to interpret anomalies more accurately, prioritize issues intelligently, and transform descriptive process data into actionable, strategic decision support.

Overview

In today's data-driven organizations, the interpretation of process mining outcomes—such as visualizations and KPIs—remains a persistent challenge. While event logs reveal what occurred within a system, they often tell only half the story. What’s missing is the context surrounding the process—factors that explain why certain patterns or anomalies appear.

This document introduces how incorporating contextual knowledge transforms raw process data into actionable insights and illustrates how GenAI can be leveraged for more accurate, scalable root cause analysis.

Process mining tools visualize process execution data to uncover inefficiencies or compliance issues. However, these tools inherently focus on what is documented in the event data. This creates a blind spot: real-world influences on processes are often not recorded directly in the data. We refer to these influencing factors as context.

Defining Context in Process Mining

To understand how context impacts process interpretation, we refer to a classification widely used in process mining.

Case Context

The most narrowly defined context, related to attributes of a specific process instance.
Example: In a purchasing process, high-value orders for raw materials may require additional checks (e.g., dual approval), while office supply orders might pass through automatically due to low risk and value.

Process Context

This involves surrounding processes and shared resources that influence the analyzed process.
Example: Delays in a sales process may originate in a related purchasing process if materials are ordered too late, impacting delivery timelines.

Social Context

Human-related influences, such as employee availability, labor strikes, seasonal workforce trends, or satisfaction levels.
Example: A production halt caused by a strike results in process delays visible in mining outcomes, which are only understandable with knowledge of the human factor involved.

External Context

External forces, such as geopolitical events or regulatory changes.
Examples:

The Brexit-induced shift in customs regulations led to prolonged shipping times.
New legislation (e.g., supply chain laws) created additional documentation steps.
Global logistics disruptions (e.g., the Suez Canal blockage) impacted delivery chains worldwide.

Context Management at Noreja

We categorize context into two main types:

Context "At-Rest"

Defined during the process design phase or while building the process mining model.

Typically static and includes:

Business rules

Organizational structure

System architecture

SOPs (Standard Operating Procedures)

Context "In-Operations"

Captures dynamic, real-time or event-based context during process execution.

Includes:

Seasonal patterns (e.g., holiday staffing)
Temporary disruptions (e.g., system outages)
Operational findings and reported anomalies

Input Formats for Context

Free text or structured text entries
Speech-to-text records
Documents and reports
Visual models (e.g., BPMN models from tools like Adonis)