Granulify: Continuous Microservice Granularity Management with Saturation Signals from Longitudinal Industrial Evidence

Abstract

Problem: Microservice granularity is typically treated as a one-time design decision, yet service boundaries must evolve with changing technical and business demands. The absence of systematic mechanisms for continuously reassessing granularity often leads to excessive fragmentation, increased maintenance effort, and elevated operational costs. Method: This paper presents Granulify, a novel evidence-based approach for continuous granularity management that addresses critical gaps identified in a systematic mapping of 24 primary studies. The approach integrates qualitative diagnosis and quantitative assessment to support iterative boundary adjustments. The approach introduces the Granularity Classification Spectrum (GCS) for positioning services along a granularity continuum, and the Granularity Saturation Method (GSM), which combines polarity-aligned metrics into two aggregate indices—Net Benefit (NB) and Total Variation (TV)—and derives a bounded Saturation Score (SS) from their ratio for regime classification. Thresholds are calibrated through a lagged rolling window, ensuring that each period is classified only using prior historical observations. We followed an industry-oriented Design Science Research approach, operationalised through 13-month longitudinal action research in a large-scale financial services platform, where the first author served as lead architect actively participating in granularity decisions while systematically collecting evidence. Results: The rolling analysis treated the initial months as calibration warm-up and identified the first actionable over-decomposition signal in February 2024, when negative Net Benefit values indicated architectural degradation. These quantitative signals were interpreted through triangulation with qualitative observations from sprint retrospectives and architecture review meetings. During the prospective intervention phase, Granulify supported the team in identifying efficient decomposition periods, detecting empirical saturation before further fragmentation, and performing a strategic consolidation that was associated with the highest Saturation Score values of the observation period. Conclusion: Granulify provides actionable, evidence-based support for architectural decisions that traditionally relied on subjective judgement, supporting teams in detecting saturation signals and adjusting service boundaries before incurring unnecessary complexity and costs.