Online retailers such as Amazon have changed our shopping habits. Besides providing conveniences such as shopping in one's pajamas, they also offer an infinitude of choices with lower prices. Brick-and-mortar retailers still are indispensable to those shoppers for whom instant gratification is a must. Still, online shopping retailers became a staple for the rest of us, from buying books to groceries, from music to movies.
One of the exciting novelties that came with the online shopping experience does not apply to brick-and-mortar retailers. Almost all online retailers recommend products to their customers to purchase depending on their previous purchases. For example, Spotify recommends certain artists depending on what you have listened to, or Netflix offers you a specific list of movies to watch depending on what you have watched or liked so far. In this post, we are going to attempt to explain in broad stroked how these systems work.
Online retailers have the luxury to collect detailed data on customers: what they buy and what they have looked at. They do this individually and in large numbers to measure and identify statistically verifiable delineated subgroups. They may even derive detailed socio-economic data on their customers by peeking into their friendship networks if they have little to no scruples about online privacy. Such heaps of data open new avenues hitherto not possible or even thought of. In an age where the best minds of our generation try to make people click on ads and buy stuff, data has become the new gold, and the race for developing applications of machine learning and artificial intelligence for marketing has become a new gold rush.
One of the new applications of such heaps of customer data is a simple but intelligent enumeration of goods in the same shopping carts. By recording which products are purchased simultaneously, we can reliably predict which products are purchased together. In this way, we can expect with the precision that someone who buys a power tool often also buys spare batteries for these power tools. Or that someone who watched "Interstellar" on Netflix will most likely watch "Tenet" once it becomes available. Such systems are called recommendation systems.
Not to be outdone by their online counterparts, the brick-and-mortar retailers also jumped on the bandwagon by collecting their data by offering loyalty programs for their customers by offering discounts in exchange for recording their purchase history. If you are known to your grocery store or supermarket (online or otherwise) via your telephone number, they may know your purchasing behavior to-a-T, perhaps even better than your significant other. They may claim that you are not just a number, but you may be a uniquely identifiable customer ID.
If we look closely, we see that recommender systems supplanted the eminent class of salespersons (mind you, I mean a salesperson and not a clerk) who knew their customers intimately through years of cultivated business relationships. Fully automated recommender systems may be the natural result of the evolution of relationships between large companies and small customers. However, for small and mid-size businesses, an intelligent mixture of recommender systems and customized salesforce appears to be a more effective solution. Small companies would find it almost impossible to compete with large online retailers regarding price and convenience. However, a personalized shopping experience through a live salesperson who may understand and predict a customer's needs and recommend products and services accordingly would probably beat any automated recommender system hands down when buying underwear. Not oranges. Probably.
Prof. Dr. Atabey Kaygun