Transcription is an integral part of much qualitative data analysis, yet rarely has it received close attention in debates over the use (or non-use) of "computer assisted qualitative data analysis software" (CAQDAS). This article draws upon a mixed-methods study that involved transcribing conversational interviews with carers, third sector practitioners and policy-makers, to explore how "computer assisted transcription software" (CATS) can affect data and its analysis in ways unanticipated at the outset by researchers. From an agential realist perspective, the article outlines three steps towards making principled choices over the use (or non-use) of CAQDAS in qualitative data analysis. These steps require navigating extremes associated with technological determinism; that we re-think our understandings of the software-data-researcher relationship; and that we move away from asking how well a given CAQDAS can 'perform' and towards exploring what a given CAQDAS can (and cannot) do.