Process mining for agile software process assessment and improvement

Abstract

Context:

Agile software processes, designed for flexibility and continuous improvement, pose challenges in extracting actionable insights from event logs due to their inherent unstructured nature.

Objective:

The study evaluates whether existing process mining techniques can effectively uncover reliable and insightful information on software development processes adopting agile methodologies.

Method:

The work uses various algorithms to analyze procedural flows and business rules within an event log containing data from 3,418 agile software development projects at a company with over 1,500 employees. By categorizing processes according to project size, our analysis aimed to determine the kind of insights these algorithms could reveal. We specifically focused on algorithms that produced high-quality insights for a deeper examination of aspects like effort rate, frequency of activities, and relationships between activities. Subsequently, technical and managerial staff reviewed the results to assess the quality and relevance of the insights generated. Validation involved a semi-structured interview with managers and technicians to ensure the relevance and applicability of the findings.

Results:

The analysis demonstrates the efficacy of declarative business process techniques in extracting actionable insights from agile development teams’ data. Such techniques accurately capture the daily routines and documented processes of the teams. High-performing teams typically followed fewer rules, had less job rotation, involved fewer individuals, and engaged in a more limited range of activities. Domain experts and team managers found these insights to be coherent and potentially valuable for enhancing the performance of software development processes.

Conclusions:

Declarative modeling is particularly adept at revealing the patterns of flexible software development workflows, presenting initial support for teams, managers, and decision-makers through both descriptive and prescriptive analysis.