Redefining Retail with Artificial Intelligence: Tesco
Technological and industrial advances that have brought about economies of scale, have forced retail firms to evolve into a race for achieving the most optimal, efficient and lean operation, i.e. the ability to deliver the promise of the right product at the right place at the right time at the right price. Online shopping in the mid-90s brought convenience and increased expectations of immediacy. Since then, there has been a sea of change in advancements and expectations alike, on account of a number of factors like analytics, business intelligence, artificial intelligence and machine learning.
Since the 90s, decision support systems have helped planners, buyers, marketers, finance professionals and warehousing and retail operators with analysis and decision-making. However, these required the underlying data to be clean and reliable and that the right set of metrics be defined to allow the right decisions to be made. Back then, AI was naive at best, inflexible and needed to be hand-coded as the lack of libraries, native technical support and raw compute power made AI difficult to build. According to a McKinsey report, the top 10 global retailers from 1990 to 2010 changed significantly, with only 4 of the top 20 in 1990 even being on the list in 2010. With changing demographics, new business models emerged as an outcome of it. It became clear that only those who were serious about data, analytics and AI for decision-making had survived. Over time, with AI being accepted as the norm, data has become reliable while compute and storage have become readily available, more powerful, cheaper and elastic. Retailers who had invested in fragmented applications of AI, started staring at bigger questions and answers for AI to solve.
Perhaps the first application of AI in retail is demand forecasting at a product, location and time level which enhances customer experience and facilitates the efficient use of company funds. Many retailers employ a layered approach to disaggregate the component parts and then aggregate them together. Advanced AI algorithms can be leveraged to correlate climate and weather patterns to arrive at product mix optimization which determines which products should sell at which locations and at what time. At a macro-product level, even before the forecasting, assortment planning can leverage AI to optimize the top-down, bottom-up and middle-out projections of the product assortment to reconcile these. The granular demand forecasting also defines where the product should be made available, i.e. in a store or a fulfilment center. Shelf space optimization solves the more granular problem of where the product should be placed within the store or fulfilment center and is typically based upon the sales velocity of the item, how conveniently is the shelf located, etc. At a macro-business planning level, where to locate the store or fulfilment center in itself is an optimization technique that retailers solve based upon population, target customer demographics, competition, access to ports and store footfall.
Transporting products between locations to locations to be made available in the right place at the right time is the biggest concern for every retail logistics operation. This involves optimizing primary and secondary routes at a planning level as well as optimizing primary and secondary trips at a daily scheduling and operational level. The ability to adjust to business or non-business changes on a real-time basis makes this problem more complex to deal with. Resource capacity optimization is used especially during peak seasons like holidays and festivals when contract staff is required to allow the business to stay flexible when the demand spikes.
In addition to these traditional applications of AI to increase margins and improve customer experience, retailers are exploring, building and even productionizing new ways to apply AI to broader, deeper and granular areas.
Automation and robots have started playing a role in back-office, warehousing and fulfilment center operations, such as storing and retrieving products from warehouse locations. Every step further in this direction requires more information to be available to the algorithms to figure out the shortest path, most efficient place to store the products, etc. Therefore, some traditional problem statements such as shelf space optimization are eliminated as a result of warehouse automation. Customer experience is being enriched as a result of personalization and learning of customer behavior from both offline and online channels, paving the way for a true omni-channel experience. Conversational AI-enabled customer service agents are gaining popularity and are positioned to serve even fully capable experiences by employing big data, machine learning and natural language processing. These are built upon the personal and digital assistants that have been in existence for a few years now. Magic mirrors and virtual mirrors that provide visualizations applying augmented reality, are being piloted. Computer vision will be used to identify customers, automatically create basket as well as optimize checkout lines.
It is clear that AI is being and will continue to be applied to every problem that can be optimized in isolation. That the global optima are desirable than local optima is well known. How AI crosses the chasm to arrive at global organizational optimization is something that remains to be seen.
The article is contributed by Mandar Amdekar, Head of Engineering & Technology - Supply Chain, Tesco Bengaluru