Diagnosing Conversion Failure in Short-Video Commerce: A Stage-Based Framework for Localising Drop-off Across Reach, Content, Credibility and Purchase

When short-form video attracts views but not purchases, practitioners usually respond with a single undifferentiated intervention – more content, a new offer, a redesigned landing page – without first establishing where in the customer journey the failure occurs. This paper argues that low conversion from short video is not one problem but a sequence of separately caused breakdowns, and develops a stage-based framework for locating the responsible stage before any intervention is chosen. Drawing on customer journey theory (Lemon & Verhoef, 2016), the hierarchy-of-effects tradition and its critics (Lavidge & Steiner, 1961; Vakratsas & Ambler, 1999), research on algorithmic visibility (Bhandari & Bimo, 2022; Karizat et al., 2021; Zannettou et al., 2024), source credibility (Fogg, 2003; Lou & Yuan, 2019; Meng et al., 2024) and online retail atmospherics (Eroglu et al., 2001), the paper distinguishes four failure types – reach, content, credibility and purchase-experience failure – each with its own observable signature, governing mechanism and remediation literature. Three refinements make the framework usable under current market conditions. First, it recognises that the journey no longer runs only through the creator's profile: in 2025, half of TikTok Shop's US sales were generated directly from videos and a further 14% from live streams (Momentum Works, 2026), so the credibility stage must be defined by function rather than by surface. Second, it treats credibility as distributed across the video and the intermediate surface rather than confined to the profile. Third, it replaces unfounded rules about "enough videos" with a statistically derived rule of thumb – approximately sixty successful transitions at the stage being judged – and shows, with a worked example, how confidence intervals determine which stages can and cannot yet be diagnosed. It also outlines how a diagnosis translates into intervention priorities. The paper closes with testable propositions, a programme for empirical validation and the limits of observational attribution.