From Clickstream to Digital Maturity : Measuring Company-Level Software Proficiency in Enterprise Accounting Systems

Sammanfattning

Enterprise software systems serve users with widely varying levels of digital proficiency, yet most platforms treat all users the same regardless of their actual capabilities. This creates a challenge for companies like Visma Spiris, whose accounting platform Bookkeeping and Invoicing is used by thousands of businesses with fundamentally different usage patterns and levels of expertise. The absence of a structured method for quantifying user maturity means that the system cannot adapt to individual customers, limiting the effectiveness of personalization features including AI-powered assistants. This thesis addresses the problem by designing and implementing a data-mining pipeline that derives continuous, company-level digital maturity scores from large-scale Snowplow clickstream data. The pipeline processesover 100 million user interaction events and produces three composite scores: an activity score, an intra-modal maturity score capturing how efficiently and broadly companies operate within their existing features, and an inter-modal maturity score measuring the degree to which companies adopt digitally efficient workflows as defined by Visma Spiris’s Best Digital Practice guidelines. The three dimensions are shown to be meaningfully distinct through Spearman rank correlation analysis, with both maturity scores correlating more strongly with each other (ρ = 0.53) than with activity (ρ = 0.44 and ρ = 0.32 respectively), providing evidence that the model captures a construct beyond activity alone. The results reveal that high activity does not imply mature behavior, and that a substantial share of companies remain reliant on in efficient workflows despite prolonged system use. The proposed pipeline is deployed in a Linux environment with a Flask API and MongoDB backend (Azure Cosmos), making the scores accessible for downstream applications including adaptive interfaces and large language model integration.