Time is a scarce resource in the life of the modern consumer. People want to minimize search effort and time in their buying process, but on the contrary they want to be sure that they make the right decisions and maximize the value they get from the products and services they consume. Since the world is becoming increasingly digitally connected and transparent, innumerous amounts of data are generated and subsequently stored. This data can be used to help the consumer to quickly find what he or she wants. By giving personal recommendations you will reduce the search time and effort that consumers usually need to find the products that match their needs, and increase the probability that they will buy certain products.
Nowadays many companies use a recommendation system in their business. In some companies it is part of their core business (e.g. Facebook) and for others it just supports their customer experience (e.g. Spotify). With a good recommendation engine you help your consumer in fulfilling his or her needs and at the same time you increase your up-sell and cross-sell opportunities and customer life-time valu (and thereby your profits and turnover). Some well-known examples of businesses in which recommendations are an important part of the business model are subscription-based services like Spotify and Netflix, online retailers like Amazon and Bol.com and social networks like Facebook and LinkedIn.
For subscription-based services recommendations are very important to increase customer life-time value. Good recommendations make sure that their users will continue using their services, and thus keep paying for it. Therefore, they have to make sure that the customer never stops exploring on their platform. Spotify does this quite nicely with Spotify Weekly, which is a personalized playlist that is weekly generated by Spotify for every individual user. By giving accurate and yet unexplored recommendations that match with the user’s music preferences, they make sure that their subscribers never get bored.
Spotify weekly generates a personalized playlist for all their users with new music for them to discover
RETAILERS: A TRULY PERSONALIZED ADVICE
For online retailers, cross- and up-sell opportunities are another important reason for giving their customers recommendations, next to increasing customer life-time value. If retailers are able to identify what other products are complementary to the products in the basket, or are purchased by other people with the same profile, they will be able to actively offer these products to this particular customer. By applying complex data science techniques, it is possible to go further than classic product category recommendations and make a really personalized offer!
Example of a recommendation at Amazon.com
SOCIAL MEDIA: MATCHING CONTENT WITH CONSUMERS
For social media networks recommendations play an even more pivotal role in their business model, since the entire timeline principle consists of recommendations. It matches user- or company generated content to people that are likely to be interested in it. If Facebook couldn’t select the relevant posts for their users, and thus would be unable to give them what they are looking for, their users will stop using their services. From the advertiser’s point of view; why would they continue to invest in Facebook advertising if Facebook would be unable to match their advertisements with the right target group?
These examples are just a few out of many cases that show the business value of an accurate recommendation system. I can imagine many more companies for which such a system would enhance their results, for example; wouldn’t it be great for a telecom provider to know what type of smartphone and subscription specifications a particular consumer wants, before they even know it themselves? This reduces the search time and effort of the consumer and at the same time the consumer feels that the provider really understands his needs.
However, recommending might sound great and extremely valuable, but it is extremely important to give correct recommendations to your consumers. Do not offer them things they will perceive as useless, things that do not fit to their needs or personality, or only recommendations that are perceived as obvious. Wrong recommendations could result in missed opportunities and in the worst case have a great downside to (the image of) your business. Therefore, it is important that recommendations are unique and surprising, as well as accurate. However, giving an obvious recommendation now and then might increase the trust one has in the other recommendations, because the consumer knows that the recommendations he knows are correct. This positive feeling might reflect on the other recommendations and increases their reliability.
When giving recommendations you’ll have to combine the art of data science with psychology. You’ll have to extract from the data if a consumer likes a product or not. Sometimes this data is explicit (e.g. Facebook likes or ratings on Takeaway.com). But when this data is not available you have to base your conclusions on other implicit indicators. For example, the number of times that a Spotify-user repeats a particular song might indicate that he/she likes it, but when you stop a two hour Netflix movie after 12 minutes and never continue watching, it might be a sign of a dislike.
An example of a too obvious recommendation
Most of today’s recommendation systems are based on categorical inputs such as movie theme or actor, music style or artist and product category or brand. For giving such a recommendation not much data science is needed. It becomes more complex when the recommendations are based on user-profiles, product-profiles and a consumer’s past behavior. Practice shows that the latter type of recommending (user-based collaborative filtering or item-based collaborative filtering) has a greater predictive accuracy, therefore these types of recommendation that are based on data science techniques add more value to your business and are perceived as more valuable by your consumers.