Artificial Intelligence and Retail: What to Expect
Many industry observers are expecting retailers to incorporate more artificial intelligence and machine learning technologies this year, as they look to improve their merchandising and marketing decision making. These decisions can involve pricing, online product content, product line reviews and personalization, and even efforts to improve mobile user experiences. Here’s a look at the ways some of those functions could evolve.
Retailers have traditionally relied on historic sales-based trends and forecasts for making pricing decisions. However, just analyzing sales data isn’t enough. Retailers need to integrate this data with demand data available outside the firewall (e.g., competitor products and pricing, search, location, product reviews, and social media data) to be able to make more competitive pricing decisions. By letting machines identify price patterns, retailers can easily make smarter, more timely pricing decisions at scale.
Product line reviews across categories require effective taxonomy classification and mapping (i.e., making sure your products are being put in the proper categories so they can be easily found via search and so you can match and compare them with the corresponding products offered by competitors). This can take months if done manually since it involves detailed taxonomy classification of competitor products using many attributes. You can soon expect machine-learning algorithms being used to classify products under various taxonomies at a fraction of the time, helping retailers review a lot more categories and be more relevant to their customers.
Mapping products against the competition, which is currently being done either manually or through rule-based engines, will evolve, as doing so at scale will be crucial. Retailers have started experimenting with sophisticated methods of product clustering combined with self-learning fuzzy logic algorithms to get this done by a click of a button.
Online retailers are striving to include more attributes in their product content to inform confident purchase decisions. A process known as attribute extraction, which includes extracting attributes from feeds or manufacturer websites, is carried out for attributing products at scale. Machine-learning algorithms will be relevant here, as the need for speed, scale and diverse sources for attributes continues to grow.
Product images also form an important part of product content. Deep learning algorithms can be used to classify images based on the retailer’s requirements — e.g., identifying fraudulent images and promotional text on images at scale.
Product discovery requires products to be discoverable in multiple nodes across a catalog or in the taxonomy. Multiple taxonomies or a constantly changing taxonomy for product discovery is an important governance that retailers need to deal with in the future. It’s impossible to tackle this manually, so a combination of classification techniques and constant learning algorithms coupled with human intelligence will be the future need for most online retailers.
Navin Dhananjaya is chief solutions officer at Ugam, a leader in managed analytics.