Uses of AI and automation technologies in the retail sector: benefits and challenges make decisions that could affect staff with little or no input from a human manager. For employers, there are obvious benefits for work planning and being able to ensure staff are working in the ways in­tended. But surveillance and monitoring systems can also potentially undermine employee trust and engagement. For workers, the deployment of increasingly intensive monitor­ing and surveillance systems represent serious concerns over work intensity and job quality. Dynamically-adjusted targets, such as those discussed above, can impose intense pressure and act as a source stress upon workers, especially when imposed without recourse or human managerial oversight. Further concerns abound in terms of both robustness/quality and transparency. In many cases, algorithmic management systems represent ›black boxes‹, with little information avail­able either to those being managed or even to line managers over how decisions are arrived at and can make it harder to challenge decisions as they appear to be objective, based on science and facts not human biases. This can lead to an element of distrust, increase arbitrary managerial over­rulings, and also make fixing problems more challenging. The rollout of such systems further raises the possibility of attempting to ›game‹ algorithms in ways which do not ben­efit firms or workers. Monitoring of particular metrics, such as sales targets, time spent performing certain tasks, etc, can lead employees to focus on working to those targets at the expense of other tasks. This can lead to unanticipated patterns of behaviour that may not always have the intended results for employers. This may be a particular problem in ›black box‹ automated surveillance systems that use machine learning as it is not always clear which variables carry the most weight(and these may be quite arbitrary). Finally, such systems can further intensify work, and raise questions about pernicious or illegitimate forms of worker surveillance. This can erode worker-employer trust, particularly if monitoring leads to sanctions for workers, with negative consequences for employee engagement, motivation and potentially staff absence and turnover. The Covid-19 pandemic has highlighted particular concerns around homeworking. Some worry that AI is(or could) be used to monitor ingoing and outgoing communications in the context of remote working. Systems are already widely used to monitor whether remote workers are working by looking at log-in/log-out times, keystrokes and screen cap­tures. However, AI makes monitoring a far greater number of inputs – raising concerns about surveillance overreach pri­vacy, and fostering distrust between managers and workers. ›Refractive Surveillance‹ Academics Karen Levy and Solon Barocas at Cornell University have examined the huge increase in surveil­lance technologies developed for tracking customers in-store. 26 These include computer vision to determine customers’ race, gender or emotional states; facial rec­ognition systems to identify individuals; and smartphone identification devices, among others. These systems are used to collect customer data for sales purposes and to mitigate against shoplifting. They explore how data gathered for one purpose – monitoring customer behaviour and reducing theft – is increasingly leveraged for another: controlling and dis­ciplining workers. This includes(1) using footfall data and predictions to impose dynamic and unpredictable work schedules(see above),(2) monitoring customer interactions with staff through surveillance footage, kinetic mapping, natural language processing and sentiment analysis to evaluate staff performance(e.g., whether a sale resulted from a conversation);(3) reduc­ing ›clienteling‹ practices of sales workers by digitalising customer data; and(4) replacing workers altogether through advanced customer tracking and self-checkout technologies(as pioneered in Amazon Go stores). Levy and Barocas develop the term ›refractive sur­veillance‹ to account for this effect, whereby managers ›piggyback‹ on customer data to enhance their power and control over workers. They suggest that worker advocates should aim to strictly limit firms’ ability to use data for cross-purposes, seek greater worker input into programs like automated scheduling software, and enlist the support of consumer groups in challenging retailers’ widespread data collection on privacy grounds. https://ijoc.org/index.php/ijoc/article/view­File/7041/2302 2.7 PREDICTIVE MARKETING AND PERSONALISATION Advances in machine learning combined with the availability of vast amounts of consumer data and the processing power needed to process mean that automated systems can market products and offers to consumers in increasingly sophisti­cated ways. In e-commerce, the widespread use of tracking cookies alongside site-specific accounts makes customer monitoring and profile-building relatively simple(although the success of such systems depends on the right selection of measures and methods). Data can be collected and ana­lysed for a variety of measures and purposes such as time spent looking at an item, click-throughs to related items, repeat purchases and so on. This data is used to produce personal profiles of customers in order both to target them with special offers(emails offering discounts on items left in shopping carts unpurchased) and on a macro-level to mon­itor customer demographics and popular items. However, increasingly, technologies deployed in-store can replicate such internet-based data-gathering and personal profiling of customers. While loyalty cards have long enabled personal­9