Note: Deniz Oktar is the partner founder of iletken recommendation technologies and the article of his has been published in Webrazzi as guest author.
Every product has a buyer. However, it is difficult to bring right buyer and right seller together. Especially most of the online sales (ecommerce and music) are done for the most popular products and this actually is a crucial problem. The main reason to this is being able to put the bestseller product on the main page and advertise it from there. For instance, advertising Megadeth in Turkey while Madonna is selling a lot would be a mistake. If a product sells a lot, its profit margin is usually low. All competitors have to sell the same product for the same price. On the other hand, non-popular products occupy space in the storage and cause loss every second. In real life, this problem is much less. Sales representatives try to guess what you’ll buy and have you buy them.
According to results of many analysis, companies like Amazon that apply the Long Tail effect successfully make most of their profit not from the best selling products, but from the long tail part of the chart. If, non-popular products are brought to the interest of right buyers with a successful mechanism, profitability increases drastically. As seen on the graphic, depending on the number of the products, the volume that non-popular products provide can be much more than popular products.
In addition to Long Tail, Cross Sales counts here, too. Selling more than one product in a package is called Cross Sales. This can be done in two ways in ecommerce web sites. 1- A moderator decides on what products to sell in the package, 2- You build a smart recommendation system that can do this job for the moderator.
For these two options, smart mechanisms like the ones used for personalization are used. It becomes an important task to analyze each user’s buying activity and guess what products s/he would go for. Thanks to this, products that are normally not advertised because of their unpopularity are introduced to buyers that might buy those products. Calculated recommendations might be used to design customized pages for users or for email marketing. According to the experiences of companies that use recommendation system, sales are expected to increase between 10 to 35 percent. According to 2006 sales figures, 35% of Amazon’s sales are done through recommendation system.
Recommendation systems, thanks to their learnable mechanism, are able to recommend a product to the buyer each time s/he logs in. With Machine Learning based algorithms, they educate themselves on targeting the designated goal. This target is usually built to sell the most in ecommerce. With real-time work, the pages user encounters are the pages that are designed instantly only for that particular user.
Some examples to usage areas of recommendation systems in ecommerce:
- Personalized main page recommendation: building a main page for users according to their previous activity,
- Recommending similar product or new products for the user: recommending new products based on user’s instant intent,
- Cross Sale: Demonstrating the products that are sold together, applying bulk discounts,
- Personalized Campaign/ advertisement / e-mail recommendations: Coming up automatically with the campaigns user might be interested in and customizing advertisement e-mails.
Recommendation Systems are subject of a academic research. They are considered a difficult area of study and that is why various competitions are organized. Neflix’s competition which lasted 3 years and ended last month granted the team that would improve the existing recommendation system by 10% 1 million dollars (official results are yet to be announced). The fact that this team was made up of 23 other teams is another interesting point.
Recommendation Systems need to be configured continuously as user figures and website structure tend to alter. To reach the best result, a new engine needs to be built for each company. Scalability and technical difficulties come along with algorithmic difficulties. Systems that have a big number of users and products need high memory and processing power.
When the recommendation systems do not work properly, it is observed that they cause unwanted results for the company. One of the de discussion topics on Friendfeed is the event when Hepsiburada’s recommendation system advised Sule Ozmen to buy kamasutra play set dices to go with her recent purchase of ecommerce book. My personal experience with Hepsiburada is that having purchased computer parts for two years, I have been recommended to go with women’s shoes.
For these reason, except for the big players in the world, ecommerce companies get this service from recommendation companies. These companies work on as Software as a System method and they provide hosting, algorithm development services. Hepsiburada is the only ecommerce website that we know of that uses recommendation system in Turkey.
P.S.: Visuals are courtesy of www.strands.com.