View Point
I
Realizing the Value of Recommendation Engines
Abstract Recommendations engines have become ubiquitous in the digital marketplace and are getting smarter in matching customers with the products they may like. These self-learning engines have a complex in-built algorithm that analyzes available information and provides personalized and real-time recommendations. Several third-party engines report that their clients benefit from a significant improvement in online sales owing to recommendations from the engine enabling consumer conversion. However, there are several customer privacy concerns surrounding the operation of these engines. This paper analyzes how recommendation engines operate, how effective they are, how customers perceive them, and how they can evolve to become more relevant.
www.infosys.com
Introduction The business of e-commerce is based on the fundamental goal of customer conversion i.e., transforming visitors to the site into customers. This often depends on whether e-vendors have robust online engines to assist visitors in their product searches, thereby increasing the average order value of each transaction. Over the last decade, recommendations engines have become ubiquitous at all leading e-commerce stores. These engines offer realtime, relevant and personalized recommendations to shoppers, increasing the conversion rates as well as the average order value of each transaction. According to a Forrester report, 62% of online customers that notice these recommendations purchase the recommended product1.
Mechanics of Recommendations Engines Today’s e-commerce merchants can choose from several leading third-party recommendations engines that offer ready-to-use Application Programming Interfaces (APIs) hosted on a Softwareas-a-Service (SaaS) Cloud model. While different engines run on different in-built algorithms, essentially each engine operates on the same information set pertaining to the customer, his/ her social circle, and the behavior of all shoppers. Since the lack of social commerce features minimizes access to information on a customer’s social circle, recommendations are predominantly based on the latter two parameters.
Figure 1: How the engines provide personalized recommendations
The algorithm in the engine processes the available information and provides real-time and personalized recommendations for each customer. These recommendations are tailored to respond dynamically to each user and differ in real time based on the user’s online store activities.
Customer Information
Personalized recommendation
Social Information
Collective Information
WHAT IS THIS?
Who is the customer
What are the customer's friends browsing, purchasing, sharing
What are all the customers browsing, purchasing, sharing
WHY IS THIS NEEDED?
Customers expect stores to know them and recommendations need to be tailored to the customer
90% of consumers trust recommendations provided by friends . This in uences the customer’s outlook
Store-wide trends may appeal to rst-time visitors as well as repeat customers
WHAT INFO IS USED?
• • • •
EXAMPLES
Recommendations based on your browsing history
Demography Past purchases Browsing history Current shopping basket
• • •
Friends' recommendations Friends' opinions Friends' past purchases
Items purchased or recommended by friends
Table 1: What the engines use for providing recommendations
02 | Infosys
Collective information on purchases
Best-selling trends and information snippets such as ‘customers who bought this also bought’
Let us understand the mechanics of how recommendation engines work by illustrating examples using some common e-buying scenarios:
Registered customers
John visits an online store to buy an accessory for his Blackberry and conducts an online product search. Since John is a registered customer, the recommendations engine draws up information of his previous purchases - a Blackberry phone bought a few months ago - as soon as he logs in and queries for accessories. The engine also draws up in-store trends and recognizes that a majority of customers that buy the same/ similar Blackberry also purchase a particular kind of accessory. The engine collates information about John and the collective and recommends the same accessory to John. An online store with advanced social commerce features can also display comments based on the buying behavior of John’s social circle such as ‘Your friend George bought this’ or ‘Your friend Jane reviewed this product’ to influence his decision. If John adds the accessory to his online shopping cart, the engine will continue to offer real-time recommendations of products that complement his Blackberry and/or the new accessory. Thus the engine is constantly aware of John’s digital actions and refines its recommendations to suit him.
New customers Taking the context of the above example let us say John is a new visitor to the online store, seeking to make the same type of purchase. Despite having no information about John the engine can offer recommendations about collective preferences in the form of ‘Best Sellers’. As John begins to browse a few pages, the engine determines John’s preferences and leverages this information to offer recommendations that may interest him.
Online Stockout
A customer named Robert visits an online store to buy a sofa and finds that it is currently out of stock. Stockout situations at brick-and-mortar outlets are often deftly handled by salespersons who provide alternative options for purchase. In an e-commerce store, the recommendations engine performs this role by making suggestions based on collective buying behavior such as ‘People who liked this sofa also liked…’, thereby enabling the customer to explore similar options without exiting the store. Thus a timely recommendation can transform a potential loss due to a stockout situation into a winning sale.
In all these examples, the recommendations are based on mountains of data and are hard to replicate. They offer a competitive advantage to online brands seeking to attract and retain consumers. Recommendation engines are self-learning mechanisms that can be tuned to provide more relevant recommendations based on growing information sets. Further, the engine can offer real-time recommendations based on the customer’s actions. By instantly matching the shopper with his desired product, the store increases the chances of conversion as well as the average order value of the transaction.
Infosys | 03
Where can Personalized Recommendations be Found? Besides providing real-time, in-store and personalized recommendations the recommendations engine can be put to use at all customer touch points that previously contained static cross-sell and up-sell recommendations such as bills, mailers, loyalty programs, email campaigns, and search.
Figure 2: Where the engines provide personalized recommendations
In E-store
Search
Product Dicovery
Recomm. Engines
Bills and mailers
Loyalty Programs
Email Campaigns
Matches varying individual customers taste with a huge inventory
Recommendations for new products, up-sell, cross-sell suggestions and product discovery
Order of search results on the store can be varied for each customer to aid product discovery
Whitespace can be used for cross-sell and up-sell. Usage can be analyzed for recommending plans.
Recommendations pertaining to speci c program, suggestions for redemption and program upgrades
E ectively targeting abandoned shopping carts with personalized recommendation
Infosys | 04
Email campaigns are one of the most popular personalization techniques used by companies. While setting up new email campaigns, the content of emails and product recommendations can be personalized to be relevant and compelling for each recipient. Companies can combat the high rate of abandoned shopping carts by targeting those users with personalized messages and enticing them to return and complete the purchase. Product discovery plays a key role in improving the user experience for products such as books, movies and music where individual tastes vary. Stores use recommendations engines to assist customers in discovering products that match their tastes.
Video-on-demand provider in North America and UK - Matches 23 million customers with a huge inventory of movies according to their tastes - 60 -70% of views result from the recommendations
Gold standard of e-commerce. Pioneer in using recommendations - Sits on a huge volume of collective information of its customers - Customers can view what people with similar tastes viewed or purchased - Customers can ask the recommendations engine to ignore selected purchases
Social and professional networking sites - Sits on a huge volume of collective information of its customers - Customers can view what people with similar tastes viewed or purchased - Customers can ask the recommendations engine to ignore selected purchases
Music station. O ers music suggestions based on ratings - Sits on a huge volume of collective information of its customers - Customers can view what people with similar tastes viewed or purchased - Customers can ask the recommendations engine to ignore selected subscriptions
What do Customers Think of Personalized Recommendations? Customers value personalized recommendations and expect online stores to know their requirements and offer products based on their tastes. For example, customers were offended when a shopping goods retailer sent them recommendations for a bike after they had purchased the same or a similar bike4. The recommendations appeared stale and irrelevant to the targeted consumer base. A Forrester study1 on third-party recommendation engines indicated: 1.
15% of customers admit to buying recommended products
2.
62% of the customers that notice recommendations on websites follow them as it helps them find suitable products or accessories
3.
Vendors claim that recommendations can increase online sales by 2% to 20%
Companies that have leveraged third-party recommendations engines have reported significant improvements after deploying them.
PredictiveIntent
Richrelevance
a leading provider of behavioral personalization technology and services for digital businesses, helped retailer Nuwear increase consumer conversions by 93% by leveraging information about online visitors and matching them with products from a large catalog5. a company that enables personalized shopping experiences, helped UK-based apparel retailer John Lewis increase online sales by 27.9% in December 2011 by using site and email personalization6.
Barilliance an Israel-based company that creates personalized shopping experiences on a Software as a Service (SaaS) model, created a personalization service that helped increase online sales by 20% for Acuista.com7.
Infosys | 05
Despite the efficacy and the proven benefits of recommendation engines there are several data privacy concerns. To present effective recommendations, online stores need to analyze all available customer information. This includes pages browsed, products purchased, etc., which can make the customer uncomfortable about giving away too much information while the retailer tracks their in-store online activities. According to a survey by Pew Internet in March 2012, an overwhelming 68% of the respondents stated that they dislike targeted advertisements as they do not like to have their online behavior tracked and analyzed8. Figure 3: Personalized recommendations vs. Data privacy concerns
Personalized Recommendations Data Privacy Concerns Thus e-commerce stores need to address concerns of data privacy before offering tailored recommendations. While customers appreciate the benefits and ease provided by recommendations, they also have rights as consumers to know what information the engine accesses.
Limitations of Recommendations 1. The algorithms of recommendations engines are based on the following assumptions: -
A customer’s preferences are often static
-
Customers that purchased the same product may have similar likes
Usually, recommendations engines tend to offer more of the same. However, practical experience suggests that customers are unpredictable and prefer variety - both these factors go against the assumptions. 2.
Customer ratings are subjective as customers can either be generous or cautious with their feedback. Hence, algorithms that create recommendations based on customer ratings tend to be only an approximate fit.
Conclusion Recommendations engines have transformed into dynamic mechanisms that can integrate online information to offer relevant and personalized recommendations. The key driver for success is to provide more relevant recommendations by incorporating implicit and explicit information across all customer touch points. In the future:
1.
Recommendations must incorporate a richer social context - With 90% of consumers trusting recommendations provided by friends, stores will need to enable advanced social commerce capabilities. These capabilities will be able to leverage data on a customer’s friends’ purchases, opinions, ratings, reviews, and shares to provide more influential recommendations.
2.
Recommendations need to be more contextual - Customers are logging in through multiple devices from multiple locations. Recommendations engines need to be aware of the user’s context and provide device-aware and location-aware recommendations. For instance, based on the user’s IP address, the engine can provide recommendations for shoes depending on the climatic conditions of the user’s location.
3.
E-commerce stores need to offer customers the option of turning on/off recommendations - Customers have real and valid data privacy concerns regarding recommendation engines. Stores must explicitly offer customers the option of choosing to receive personalized recommendations. Customers that opt for recommendations will allow stores to track their page views, clicks, and social interactions and recommend products and services based on these. This presents a win-win situation for the customers as well as the e-commerce merchants.
Infosys | 06
References 1. What You Need To Know About Third-Party Recommendation Engines: Sucharita Mulpuru, Forrester Research http://www.forrester.com/What+You+Need+To+Know+About+ThirdParty+Recommendation+Engines/fulltext/-/E-RES57914 ) 2. Global Advertising: Consumers Trust Real Friends and Virtual Strangers the Most: Nielsen Wire blog http://blog.nielsen.com/nielsenwire/consumer/global-advertising-consumers-trust-real-friends-and-virtual-strangers-the-most/ 3. How computers know what we want- before we do: Lev Grossman http://www.time.com/time/magazine/article/0,9171,1992403-1,00.html 4. Prioritizing personalization for growth: Lauran Freedman, e-tailing group 5. Driving sales and revenue for Nuwear: PredictiveIntent case study http://www.predictiveintent.com/wp-content/uploads/2011/10/Nuwear.pdf 6. John Lewis unveils new personalization strategy with RichRelevance: Richrelevance press release http://www.richrelevance.com/blog/2012/02/john-lewis-unveils-new-personalisation-strategy-with-richrelevance/ 7. Customer testimonials: Barriliance.com http://www.barilliance.com/customers 8. Targeted advertising: 59% have noticed it but most don’t like it: Pew Internet survey http://pewinternet.org/Reports/2012/Search-Engine-Use-2012/Main-findings/Targeted-advertising.aspx
About The Author
9. Recommendations as a conversation with the user: Daniel Tunkelang http://www.slideshare.net/dtunkelang/recommendations-as-a-conversation-with-the-user
Vikas Kasturi is a Lead Consultant with the Business Platforms unit at Infosys Limited. He has over 8 years of experience in the domains of e-commerce and Customer Relationship Management (CRM). He has managed and participated in several transformational CRM programs for Fortune 100 companies that involved defining sales processes, identifying operational improvements, solution rollout, and user adoption.
About Infosys Many of the world's most successful organizations rely on Infosys to deliver measurable business value. Infosys provides business consulting, technology, engineering and outsourcing services to help clients in over 30 countries build tomorrow's enterprise.
For more information, contact askus@infosys.com
www.infosys.com
Š 2012 Infosys Limited, Bangalore, India. Infosys believes the information in this publication is accurate as of its publication date; suchinformation is subject to change without notice. Infosys acknowledges the proprietary rights of the trademarks and product names of other companies mentioned in this document.