Abstract
Recommendation engines decide most of what an online shopper sees. This paper asks what that does to the shopper's capacity to decide anything. Two strands, carrying unequal weight, and I would rather say so here than bury it in the methodology. The first assembles published evidence on how much of online retail runs through algorithmic curation. The second specifies a Stimulus–Organism–Response model linking four recommendation attributes to impulse purchase through pleasure, arousal and cognitive absorption, and reports results from an illustrative dataset rather than original fieldwork. The secondary evidence holds up well. Recommenders are credited with roughly a third of Amazon's sales and four-fifths of Netflix viewing hours, though the Amazon figure has thinner provenance than its repetition suggests. Between a fifth and two-fifths of online retail spending is unplanned, a range reflecting definitional disagreement rather than sampling error. 86% of online purchase decisions close within a day. More than half of online impulse purchases are regretted, and fewer than half of those regrets end in a return. Modelled results put the heaviest stimulus path on scarcity signalling rather than match quality (β 0.41 versus 0.34), with the urge to buy impulsively carrying most of the load to purchase (β 0.57). If that ordering survives testing, it is uncomfortable for the industry's account of what personalisation is for. I conclude, without enthusiasm, that personalisation is a real service and a real hazard at once, delivered by a system with no way of telling which sort of shopper it is serving.
