The Past | The Present | The Future
Product recommendations is a widely used personalization technique in today’s online shopping environment. As eCommerce websites began to develop, a pressing need emerged for providing product recommendations derived from filtering the whole range of available options. The broad portfolio of products on eCommerce sites made it hard for users to filter the alternatives and come up with the most appropriate decision for themselves. In the mid-1990s, recommender systems (software tools and techniques) appeared as the answer to this problem by providing suggestions for products/services to be of use to a user (Recommender Systems Handbook – F. Ricci, et al., (Springer, 2011) BBS)
Amazon.com is the first player in eCommerce to invest heavily on product recommendations therefore recommender systems. Their intention is to personalize the online store for each customer. Since recommendations are usually personalized, different users or user groups are receiving diverse suggestions. The store radically changes based on customer interests, showing programming titles to a software engineer and baby toys to a new mother. The technical approach pre-dominantly used by Amazon while recommending products is called “item-to-item collaborative filtering” and has been developed in-house by Amazon engineers. (https://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf)
YouTube, Netflix, Yahoo, Tripadvisor, Last.fm, Spotify and IMDb are other famous sites taking advantage of the recommender systems. For example Netflix, the online movie rental service, awarded a million dollar prize to the team that first succeeded in improving substantially the performance of its recommender system. Many skilled computer engineers and data scientists competed for the prize and although most did not get awarded, they had the chance to work on Netflix’s data set. After the competition, many of the entrants established their own companies and started to act as ‘recommender system vendors’ to help websites that are not in a position to develop their own recommender systems.
The main objectives of a recommender system is to:
- Increase the number of items sold
- Increase sales revenue
- Sell more diverse items
- Increase the user satisfaction
- Increase user loyalty
- Better understand what the user wants
eCommerce and content sites have traditionally been lucrative grounds to test recommender systems as these types of sites attract a high number of visitors who view many pages and take actions such as giving orders or consuming content. All this clickstream and action behaviours are then analysed and translated into ‘historical user data’ by the recommender systems. The recommendations are aimed at supporting their users in various decision-making processes, such as what items to buy, what music to listen, or what news to read.
Today, some big online retailers or content sites – that could afford to employ engineers and data scientists for the sake of developing a recommender system or a personalization project – apply their own in-house methods and algorithms. The other sites use recommender system vendors that offer enterprise software or SaaS solutions. These solutions usually function either as stand-alone engines which could be attached to a website in the form of a plugin (widget) or in the form of a rest API that feeds the site with relevant information.
Product Recommendations in eCommerce
Asos began testing its personalized recommendation feature in 2015. CEO Nick Beighton says; “The recommendations piece, that’s been deployed all over mobile now and all over our key channels and we’ve now put it into You May Like piece, we’ve had that for a long, long while, but the improved algorithms have given it a much better experience based on what you browse, what you’ve liked, what you’ve added to bag and may not have purchased, based on your behavior on other parts of the channel.
In terms of personalization, it’s a never-ending approach; there’s no start, middle and end to it. So personalization can go from here’s an edited choice of 900 brands that we’ve chosen for you, for customers like you, endorsed by customers through social media. Our content is personalized through many channels. Customers upload pictures of themselves so they’re part of the conversation, they’re part of shaping the offer. So it’s a continual journey that we’ll never stop on.” (1)
The company has also announced that it is investing “significant” cash in developing personalisation software as it optimises how it uses its data and has taken a major step in personalization with the launch of its fully personalised homepage at Very.co.uk. (2)
The cosmetics retailer has a loyalty program called Beauty Insider, which enables the brand to convert anonymous visitors to recognizable entities. Through its ColorIQ feature, which defines skin pantone, the company can match existing products to individual skin shades.
Russian online fashion retailer Lamoda saw a year-over-year (YoY) increase of $11 million in revenue after implementing a personalization software across its online platform in January 2015. They track consumer data through various forms, so that they can better target individual shoppers and deliver product recommendations.
Shoe retailer ShoeDazzle asks consumers a series of questions to gauge what they might be seeking. ShoeDazzle.com’s goal is to steer consumers to the shoes she’s most likely be interested. That’s why when a shopper arrives at the online store, she is welcomed with a style survey. With each answer, the site zeros in on what shoes she might be interested in buying. The retailer, which sells subscriptions to receive new footwear each month, uses that data to send customers shoe recommendations. The consumer then selects which items she’d like to receive. Based on how consumers respond to the selections offered, the retailer fine tunes its algorithm to better predict what its customers might like. (4)
Trends in Product Recommendations
Product recommendations is here since mid-1990s so is the Artificial Intelligence. As the computing power started to become cheaper, AI started to play a leading role in the evolution of the recommender systems. The early algorithms depended heavily on product ratings and a high traffic on product pages whereas today’s AI powered predictive algorithms could function well even in small and medium sized stores. That is usually possible with the help of a more personalized approach in which the relationship between users and products are considered to be a high-value factor in the equation.
In today’s eCommerce environment, there is a broad range of verticals and several business models which create a necessity of transformation for the recommender systems. A few cases are ;
- Seasonality is high : Whether it be the fashion industry or the consumer electronics, there is a great seasonality which is tranlated into a high product turnover on eCommerce sites.
- In C2C eCommerce websites, the item stock is only 1 unit : On the marketplaces where end users list their products, the item stock is 1 and when that particular item is sold, there is a high risk that all the system data related to this product becomes redundant.
- Small sized stored attract a low level of traffic to product pages: For the analysis around products to be meaningful, the sample data size should be statistically significant. In small online stores, the traffic is low making it hard to come up with solid analysis.
- Real time decision making: 1) The Internet shopper in 1990s used a few resources to arrive to a purchasing decision whereas today’s online shopper is bombarded by information flowing from social media, forums, aggregators, price comparison sites, online ads etc. A visitor to an eCommerce website can easily change a decision in real time. 2) Today, there is more than 4M online stores in the world – Amazon is selling 200M products on its US store. An online shopper could as well be viewing different alternatives on 5-6 sites in real time in order to make a simple purchase. The decision making process is therefore subject to change in real time.
Legacy recommender systems are now forced to gain a state-of-the-art make-up to be able to function in today’s eCommerce environment. At this point, the ‘personalization’ approach comes to stage. Personalization is understanding the user’s individual tastes and being able to act, accordingly. The user and her interaction with the product is the starting point in personalization. Shopper’s purchasing intention may change in real time and a state-of-the-art personalization engine is to be able to capture these changes in real time. By this method, the shopper is profiled in real time and product recommendations are delivered with the help of the real time analysis.
Understanding the shopper’s behaviour on different channels is also a must in personalizing the experience. Physical retail, mobile, desktop and e-mails are the main sources of information for the personalization engines. When this data is connected and formed under a single identity, it’ll be easier to understand the shopper’s real intention.
Online shoppers want personalization. The largest share of internet users in North America (81%) said they simply want to find what they’re looking for, see product reviews and recommendations, and ultimately buy the product when and how they want. (5) This is a signal showing that eCommerce recommender systems need to be evolved to match today’s online shoppers’ personalization needs from product recommendations. The investment made in personalization will surely translate into higher sales, higher customer lifetime value and loyalty.
Curious to learn more? Take a look at related posts!
- Product Recommendation Engines: 10 Reasons Why You Should Get One for Your Webshop
- Announcing Our Latest Feature: Rule Based Product Recommendations
- Product Recommendation and Artificial Intelligence in eCommerce
- 18X GROWTH IN RECOMMENDATION REVENUES: SUCCESS STORY OF MORHIPO.COM
- Tips For Personalized Product Recommendations in eCommerce
- How To Use Product Recommendations On Web?
- A Close Look at Product Recommendations