Artificial Intelligence Startups: An Empirical Study of Business Model Innovation in Digital Entrepreneurship

Prevailing digital innovation theory assumes fast go to-market, rapid, and near-frictionless scaling of Software as a Service ventures. However, Artificial Intelligence (AI) ventures developing data-dependent solutions in Business to Business (B2B) niche markets tend to deviate from these assumptions. Drawing on a comparative multiple case study, we investigate how data dependency affects Business Model Innovation processes and trajectories in AI ventures. We contrast the early stage business model experimentation of three AI digital ventures with that of three non-AI digital ventures operating in B2B niche markets. Our data corpus comprises 11 semi-structured interviews with 119 pages of transcripts, and is complemented by informal interviews and archival data. Our analysis followed a transparent, process-oriented coding approach. The findings reveal three recurring limitations on business model experimentation from data-dependent solutions - data availability, epistemic predictability, and data path-dependency – which condition how Business Model Innovation inputs are acquired, experimentation unfolds, and how scaling trajectories emerge. The venture AI ventures in our sample deviated from classical Software as a Service scaling patterns, instead relying on service-intensive launch strategies and staged automation to build a data corpus for AI training and expansion. By specifying data dependency as a structural constraint on business model experimentation and growth, this study refines core assumptions in digital innovation and BMI literature. It provides process-level explanations for slower and staged growth trajectories in AI ventures.