Recommendation Engines – those systems that use data to make predictions around products and information a user might want – have the potential to change the way brands communicate with customers, and play a pivotal role in customer relationship management. They can help retain customers by providing tailored recommendations specific to their needs. Their use in solving CRM problems has resulted in direct impact on key business metrics like customer engagement, customer retention, and ROI on marketing costs.
One place we see Recommendation Engines shine is streaming services. The global over-the-top (OTT) streaming media services market is expected to grow remarkably and online media consumption is on the rise as more customers are subscribing to these entertainment platforms during the COVID-19 pandemic. Companies have started to recognize relevant approaches to connect with their customers based on the information they can gather on each customer’s preferences and purchases.
A study conducted by McKinsey suggests that 35% of what consumers purchase on Amazon, and 75% of what they watch on Netflix, are based on recommendations from Recommendation Engines. In the over-the-top media streaming service industry, Recommendation Engines have resulted in substantial cost savings in marketing efforts. Netflix estimates that the company can save up to $1 billion in marketing costs each year.
Despite the widespread usage of recommender systems and the positive impact being stated by business leaders, there are critical challenges that hinder organizations from realizing their full potential.
1. Sparsity of data
The number of users of an online store or OTT service can exceed a million, and the products or programs on offer number into the thousands. But unless a user rates what they have purchased or viewed, the Recommendation Engine can not capture these preferences and suggest to a similar audience since the algorithm is based mainly in the rating data.
2. Computational complexity
With increasing complexity, the time and resources required for the recommendation task will rise. Those Recommendation Engines with less complex algorithms will be more beneficial. However, the accuracy and the relevance of recommendations could be negatively affected.
3. “Cold Start” problem
This problem occurs when new users enter the system or new items are added to the catalogue. In such cases, neither the taste of the new users can be predicted nor can the new items be rated or purchased by the users leading to less accurate recommendations.
This problem occurs when new users enter the system or new products are added to the catalogue. In such cases, due to limited information, the taste of the new users cannot be predicted, and the new items can’t be rated (or purchased) by the users, leading to less accurate recommendations.
4. Inability to capture changes in user preferences
Consumer preferences vary throughout the CRM lifecycle. The challenge for the Recommendation Engine is to identify changes – or indications of impending changes – in consumer preferences. If the Recommendation Engine fails to identify these, it will lead to generalization of recommendations thereby hampering personalization.
5. Altering recommendations based on the context
Recommendations are supposed to vary based on the context in which the user is accessing the services. Various external factors like time of day, day of the week and platform used to access the service (mobile, tablet, laptop etc.) affect the choices of a customer. The challenge for the algorithm is to incorporate these external contextual features and render the recommendations.
How can analytics help solve Recommendation Engine challenges?
The above challenges can be addressed with the help of Data Analytics. Advances in Recommendation Engine algorithms and control systems using feedback loops have enabled the industry to ensure greater personalization when using these tools.
Approaches that are currently being taken for recommender systems include:
- New algorithms are built on a dataset of user/product feedback or ratings. It filters out items/products that a user might like based on reactions by similar users.
- New users’ responses to a questionnaire about their preference for a certain set of items or products during their onboarding process is captured and used by algorithms to provide recommendations based on the profile constructed for the user.
- Algorithms capturing features like genre or director in movies, or country of manufacturing in electronics products, or type of ingredients in products provide more nuanced and personalized recommendations.
- Including the context in which the user is consuming a service (e.g. a movie enthusiast may prefer watching long duration horror movies on weekend nights) helps provide more accurate recommendations.
Case study: Harnessing CRM data for more powerful Recommendation Engines
Our client, a North America-based OTT service provider, wanted to improve the number of newly acquired subscribers moving from a 14-day free trial plan to a paid subscription plan. This involved unlimited free access to all video content during the first fourteen days of registration. During this phase it is imperative for the consumer to watch content matching their interests from the video catalogue, in order to identify value in the service and become a paying subscriber.
Since the consumer has only a short time to decide to opt for a paid subscription, the organization decided to provide personalized services for the fourteen-day free trial period. A Recommendation Engine was a logical solution for harnessing and analyzing consumer behavioral data and identifying the ideal content.
The Kantar Recommendation Engine was deployed in the client ecosystem. The solution harnessed data on subscription watch history. An offline recommendation was sent to the subscriber on the 5th and 10th day of signing up for the free trial subscription.
Some of the challenges involved were:
- The total subscriber base consisted of around 1.2 million paid subscription members. The total number of shows available in the catalogue is around 700. With the increasing complexity of algorithms, the efficiency of the Recommendation Engine drops when catering to a large subscriber base.
- Newly acquired subscribers had to be shown personalized recommendations at the earliest opportunity, or customer churn was likely. This meant that the problem of “Cold Start” would be encountered, as very minimal information on new customers was available.
- Our client operated in a niche entertainment segment, so the recommendations administered tend to appear “stereotypical” without any personalization. The above challenges were solved by using a combination of algorithms, as mentioned previously. The output of the algorithm suite was dynamically scored to ensure the best click-through rate. This scoring depended on the context in which the user operated.
The OTT service provider was able to see an uptick in the engagement levels of the free trial subscribers. Consumer Engagement improved by 14% during the free trial period. Because of the personalized recommendations, the subscribers were able to discover titles that suited their interests from the video catalogue. This resulted in free trial subscribers realizing higher value from the services offered by the business. The free trial to paid subscription conversation rate rose from 60% to 68% after the deployment of the recommender systems, resulting in a 13% growth in revenue from subscription services in a year.