Intelligent recommendations, or recommendation engines as they are also called, are everywhere. As digital consumers, we experience them when we buy merchandise on Amazon, pick a show to watch on Netflix or scan curated content feeds on social media. They are algorithms that analyze data about users, items, and interactions between users and items and suggest recommendations about what users might consume next. Such systems seek to predict the rating or preference a user would give to an item and their propensity to act on it.
Financial services organizations, including banks, insurance companies, and wealth-management firms, have intricate access to their clients’ finances, investments, and spend through various product offerings. As such, these companies are uniquely positioned to take advantage of intelligent recommendation systems. These systems offer great benefits to the industry by combining artificial intelligence, machine learning and data-monetization strategies for creating cross sell/upsell opportunities, running targeted marketing campaigns, and identifying incremental revenue streams.
How do they work?
Recommendation engines apply various data-mining techniques such as collaborative filtering, content-based filtering, knowledge-based and case-based methods depending on the characteristics of the domain, the quality of available data and the business goals. Let’s look at the two most common ones in greater detail.
Collaborative filtering (CF) – This is one of the most often used recommendation techniques because it’s not dependent on any data other than the transaction-level data itself. These techniques find correlations in the consumption patterns. There are several types of collaborative filtering algorithms:
- User-based CF: Measures the similarity between target users and other users and offers products that these similar users have chosen in the past.
- Item-based CF: Measures the similarity between the items that target users interact with and other items. It is quite similar to user-based CF, but instead of finding users who look alike, it tries to find items that look alike.
- Context-aware CF: Adds another dimension to the user-based and item-based CF in the form of contextual information such as time, location, and social information, thus enhancing a recommendation engine with context.
Although very effective, user-based CF is time- and resource-intensive, since it needs to compute every customer pair information (similarity score). As such, for large user bases, this algorithm is hard to implement without a very strong, parallelizable system. Item-based CF does not need to compute similarity scores between customers. And with fixed set of products, the similarity scores between products is fixed over time.
Content-based filtering (CBF) – This approach makes recommendations based on the metadata of items in a user’s history and items you would like to suggest. It is a personalization technique that builds a profile of a user’s personal interests with the goal of finding look-alike items and recommending them. The process involves two main tasks:
- Identifying metadata attributes to be associated with each item.
- Building user profiles of items a user has interacted with, giving more weight metadata attributes that appear more often. User feedback is critical here to fine tune the profile and subsequent recommendations.
The accuracy of these models relies heavily on the quality of metadata and are usually less accurate than collaborative filtering methods. Other simpler techniques like popularity-based methods or market basket analysis generally do not have high predictive power or accuracy.
How is performance measured?
In the context of financial services, recommendation engines are typically deployed with the goals of identifying and raising effectiveness of cross channel/cross product sales campaigns, identifying a customer’s affinity for specific financial investments and products, reducing customer churn and boosting revenue. So, how do we assess how well our recommendations for the various next-best offerings are?
One technique is cross-validation and a common metric is the so-called Area Under Curve. This measures how likely items consumed by the user come out on top in the final ranking matrix.
Once the intelligent recommendations systems go online, there are two KPIs that you should track and optimize:
- click-through rate – tracks how often the user explores the recommended offerings
- conversion rate – tracks how often the user engages in a transaction based on the recommended offerings
Other business metrics to consider include return on investment and customer lifetime value. However, these metrics depend on a number of unknowns and thus can be difficult to measure at times.
OpenText Magellan accelerates deployment of intelligent recommendations
Built on an open product stack, OpenText™ Magellan™ lets you quickly build, deploy and monetize intelligent recommendations for various use cases by bundling technologies for advanced analytics, machine learning, data modeling and preparation, and enterprise-grade BI into a single integrated infrastructure. Its Magellan Text Mining component allows you to augment and enrich these recommendation solutions by marrying your structured data with metadata from unstructured sources of information like websites, Twitter, document repositories, emails, etc.
With all the necessary tools, components and technologies pre-integrated and configured out of the box, Magellan simplifies, accelerates and greatly reduces total cost of ownership and time to market for rapid development and deployment of intelligent recommendations. Learn more about financial service providers can benefit from intelligent recommendations powered by Magellan.
OpenText’s AI & Analytics Services Practice can guide you and your team through this entire process and help define and implement a successful intelligent recommendation solution for your organization.
