FRIEDRICH-EBERT-STIFTUNG – Artificial Intelligence and Auto­mation in Retail ›When I started working in retail forty years ago, if I was asked by a client:»have you got such and such item?« If I didn’t have it, I called the other branches of my company to see whether the product was available and how long it would take to be sent to our store… Today this is not happening because people don’t want to wait around for you to make those enquiries(which can take time) anymore. What I need now is to use software to locate the item instantly: Where is it? How long will it take to reach my store? Can it be sent directly to the consumer? This necessitates data on stock to be shared between both stores and warehouses… and for the worker to have substantial digital skills and to be able to adapt quickly to new digital systems and ways of working.‹ – Spanish retail sector trade unionist 2.4 WORK/STAFF PLANNING AND SCHEDULING Algorithmic work planning and staff scheduling is increas­ingly common among larger retailers. Benefits for employers include more precisely forecasting and calibrating staffing rotas with peaks of demand – whether in-store or for home­based e-commerce deliveries. Systems like Kronos and Per­colata combine multiple data points like customer foot/web traffic, delivery arrivals, workforce skills data, and weather forecasting to optimise staffing numbers over any given pe­riod. 21 Use of such responsive digital scheduling systems can help ensure customer demand can be more readily absorbed, providing a smoother shopping experience at peak times. Automated shift planning software can represent a powerful negative downward force in workers’ conditions and in work life balance. This is particularly true when combined with flexible(e.g. ›zero hours‹) contracts. Used together, these permit employers to deploy labour during demand peaks and reduce labour costs during troughs – shifting the burden of risk for reductions in consumer demand from the firm to the worker. Inside the workplace, moreover, scheduling systems can lead to work intensification since it is intended to minimise ›downtime‹ when workers could traditionally relax on the job. They further can lead to a sense of being ›always on‹ when combined with little advanced scheduling. The kinds of work intensification that you see in e-commerce were felt to be permeating bricks and mortar stores as well. The automation of work planning and scheduling can de­humanise the employment relationship by limiting workers’ abilities to change or contest schedules(for example, taking time off for family reasons or swapping shifts) as the system appears ›objective‹, while even managers often have less autonomy to question such systems. Algorithmic scheduling could in principle act to delimit managerial use of scheduling as a system of discipline, reward and favouritism. However, there is evidence that such ›automatic‹ scheduling systems can in fact have the opposite effect and be used to favour and discipline workers. 22 2.5 TASK ALLOCATION, TARGETS AND REWARD Alongside planning and scheduling applications, retailers are also deploying digital technologies on a large scale for purposes of ›algorithmic management‹. 23 Tasks, targets, rewards and bonuses are increasingly being algorithmically allocated to workers by variously sophisticated software analytics packages(including scheduling systems like Kronos identified in section 2.4 above). Machine learning and AI enable employers to gather, clean and analyse ›big data‹ gathered across a very wide range of sales functions – such as sales calls, webinars, and prepara­tion, alongside customer interactions – in order to pinpoint the most valuable activities and strategies and develop new comparative metrics for staff. Data for such systems is gathered from a diverse set of sources, including kinetic and heat-mapping software, in-store cameras, beacons, sensors, RFID tags, wearable devices, image and language gathering and processing, and browser tracking cookies and apps(in e-commerce). This takes place through sales and customer interaction data, gathered via digital and physical sensors including payment records, cameras and wearable devices (wearables). For firms, such systems promise substantial benefits of being able to develop highly refined customer data, and to closely monitor and improve staff performance more than is possible through human management. 24 For example, a multinational jewellery retailer was reportedly using detectors to monitor footfall and sales data, with this data being used to adjust staff sales targets in real time. Such systems can be used to encourage staff productivity by stim­ulating sales competition between workers while appearing to legitimise algorithmic control. 25 Concerns were raised among focus group participants that such systems could lead to work intensification, and that creating such intense competition between workers could undermine cooperation and trust between staff and lead to unintended consequences. Furthermore, such systems necessarily involve increased data gathering on staff and cus­tomers and feed in to surveillance and monitoring systems (discussed in the following section), raising ethical and legal questions about privacy and control, alongside whether such procedures can be said to be in line with existing collective agreements. 2.6 SURVEILLANCE AND MONITORING SYSTEMS The task allocation and workforce planning systems outlined in sections 3.4 and 3.5 necessarily involve the gathering and use of sales, staff and customer data on an unprecedented scale, often in real-time. This can include data from wearable technology and in-store(or warehouse) cameras and sensors, as well as data from sales and stock information and RFID tags. While workplace surveillance and monitoring is nothing new, there were concerns among focus group participants that AI and automation were enabling staff surveillance and monitoring on a scale not seen before and could be used to 8