Skip to content
Analytics  •  B2C  •  Customer Experience  •  Customer Lifecycle Value  •  Growth Hacking  •  Marketing Technology

Analytics Tactic Spotlight: Recommendation Engine

Recommendation engines have gained increasing popularity since Netflix showed how to successfully recommend new movies to their entire user base on a large scale. However, knowing when a recommendation engine is a good fit for your purpose and when it is not can be non-trivial. Also, a recommendation engine is often confused with a Next Best Offer (NBO) model, which is not surprising since they share multiple properties. This first post, in a three-part series, dives deeper into the world of recommendation engines and how you can utilise them to drive business results.


Typically, the purpose of a recommendation engine is to improve the experience of your product or service by providing good and relevant recommendations to your customers. The goal of your recommendation engine is not to sell anything short term; instead, the goal is to improve the customer experience and thereby improve customer loyalty to your product and brand. This will add long-term positive value to your company indirectly. In contrast, the goal of the NBO model is to upsell a product or service to your customer, yielding direct value to your company. On the other hand, since they both identify products that in some way are relevant to the customer, although, with different goals in mind, they can sometimes be used interchangeably with great success.


To get a deeper understanding of how a recommendation engine can be utilised to its full potential, we need to take a quick review of the logic at its core. Say we want to invite our friend Jane on vacation, and we would like to pick a destination she will enjoy. We know that Jane previously visited Mallorca, which she enjoyed, and she also visited Thailand, which she did not appreciate as much. We have some other friends who thought the same way about these destinations, so we asked them what destinations they like that Jane has not yet visited. Many of them said they enjoyed their stay in Gran Canaria, and since they like and dislike similar destinations to Jane, it is likely she will also enjoy them. We have decided to invite Jane to vacation in Gran Canaria. This process is often referred to as Collaborative Filtering.



Jane’s enjoyment of Gran Canaria is approximated using data from people with similar opinions

Exhibit 1: Jane’s enjoyment of Gran Canaria is approximated using data from people with similar opinions


Now, let’s review the example of a subscription-based music streaming service where customers pay a monthly fee to listen to songs without limitation. Recommending songs or artists to exist customers does not add any short-term value for the company by direct sales unless the service charges for individual plays of songs or for “unlocking” new artists. It is more likely that recommending a song will add short term licensing costs since it increases the number of plays for the song. However, by providing recommendations that users appreciate, the experience of the music streaming service is improved, and users are more likely to keep using this service instead of switching to a competitor’s services. Over time, this will build up loyalty for the service and the brand, which in turn will help in retaining customers and securing revenue from future subscription fees. Thus, the recommendation engine has added long-term value to the company indirectly.


By comparison, the NBO model would identify that multiple devices are using the same subscription recurrently and suggest that upselling a family package to the customer is the best way to increase revenue directly. The upsell to a family package adds value to the company more directly, while the recommendation engine indirectly makes sure that long-term revenue is secured. In combination, the two models complement each other and serve as a compelling and automated method of driving both positive short and long-term results.


An often overlooked application of the recommendation engine is to use it as a reviewer of your product portfolio. How often a product or service gets recommended is a good indicator of your customers’ overall sentiment towards it. This information enables you to increase funding on advertising for the products that your customers are more likely to engage with and to decrease spending advertising money on products less likely to be enjoyed. Also, this can indicate which products you should stop selling and maintaining due to their lack of popularity.


Read Analytics Tactic Spotlight: Customer Lifetime Value


In summary, a recommendation engine is great for improving the customer experience and thereby adding long-term positive value to your company. One challenge is then to measure the impact of your recommendation engine on actual monetary value. One way to do this is by modelling and measuring the change in your customers’ lifetime value when exposed to your recommendation engine. The Customer Lifetime Value (CLV) model will be the topic of the next instalment of this three-part series.


If you want to know more about recommendation engines or how we can help you with the development of one, don’t hesitate to contact us!


Contact us


Written by Fredrik Kihlberg and Henrik Nordström



Eric Hörberg

Senior Data Scientist



Möchtest du lieber auf Deutsch weiterlesen? Kein Problem, du kannst auch weiterhin auf der englischen Hauptseite stöbern. Wähle unten einfach deine Präferenz: