Recommender Systems

Recommendation engines allow the effective, intelligent personalisation of your offerings.

‘Customers who bought this item also bought …’, ‘Discover Weekly’, ‘Recommended for You’ – the personalisation of online services through recommendations is now ubiquitous. Amazon, Spotify, Netflix, YouTube, Zalando and many other services successfully use recommendation engines and machine learning to generate these suggestions. We will put our expertise in designing and implementing recommender systems to work to create effective and intelligent personalisation for your business.

Empfehlungen

30 increase in revenue (Amazon)
80 more Streaming (Netflix)
1 Billion Dollar Added Value (Netflix)

Considerable benefits from personalisation

Personalisation generates considerable benefits for both businesses and users, making it indispensable for today’s online services:

Personalisation is a key factor in online success, and recommender systems or recommendation engines are the technical facilitators behind it. These systems use data to learn and aggregate user needs in order to customise and optimise content and its presentation.

Adding value

across industries

Leading technology companies are demonstrating that intelligent personalisation adds value across industries: from e-commerce to entertainment right through to social networking and online journalism.
When competing for business online, the offerings that prove most valuable to users will win through.

Bild von Marcel

„Be relevant or become irrelevant“

Personalisation improves the user experience by reducing a vast number of options to the most relevant ones. This makes it easier and more enjoyable for users to find the right product, playlist or item. Improving the user experience ultimately increases a business’s success by reducing churn rates and improving click-throughs. Additional benefits include a more precise understanding of the customer base and improved segmentation. Businesses can use these insights to develop and position new campaigns, products and services effectively.

Data-driven machine learning

Machines use data to learn individual user preferences in order to suggest relevant content (such as products, songs or newspaper articles) from the tremendous amount of options offered. Analysing user, inventory, interaction and context data clarifies user preferences and provides a starting point for model-based learning approaches.

Content Information

Content Information

The data and approaches used by recommender systems are as varied as the industries themselves. An online vehicle marketplace, for example, requires structured data such as price, model and mileage, whereas the extraction of information from unstructured data (articles) is essential for online newspapers.

Context Information

Behavioral Information

User behaviour is also a factor. This is expressed through user actions, including clicks, views and likes. These actions connect users to items (such as products, videos and articles) and implicitly or explicitly show how users feel about them. Skipping a suggested song after 10 seconds, for example, indicates something completely different from adding it to a favourite playlist. Recommendation engines aggregate this information in order to use similar user actions to generate suggestions.

Context Information

Context Information

Last, but not least, contextual information – which includes details of the time and place of an interaction or the device used to perform it – is what makes recommendations relevant. After all, without information about a user’s current location, it would be impossible to search for nearby cafés for them.

The holistic view – from proof of concept to deployment

Effective domain-specific personalisation relies on different approaches. The most important thing, however, is to take a holistic approach – from PoC to deployment.
When it comes to recommender systems, we distinguish primarily between collaborative, content-based, and hybrid approaches. The distinction is made through the type of data used. Companies typically start with collaborative approaches, then expand to a content-based method to solve what is known as the “cold start problem” before ultimately refining their approach to a hybrid one.
Standardised solutions are limited in their ability to help achieve effective and scalable personalisation, and proof of concept alone does not add value to users or companies. What is needed are holistic, iterative domain-specific solutions which run end-to-end, from concept to productive solution to monitoring and continuous improvement.

inovex has cross-industry experience in the development and operation of recommender systems. Our experts are familiar with the latest developments and can avoid the pitfalls involved in deploying and productively operating models. We maintain regular contact with science and industry through events such as the annual Recommender Systems Conference. In working with you, we will leverage our in-depth knowledge and our many years of practical expertise to develop tailor-made solutions to facilitate personalised, effective user experiences on your platform.

We have a proven track record of success in this area. One of our projects involved working with a retailer to use receipt data to generate customised coupon suggestions. Personalising customer coupons increases the likelihood of them being redeemed – significantly boosting sales. We are also currently working on a recommender system for a video-on-demand platform. This system will increase customer loyalty, thus eliminating the costs involved in winning back lost customers and ensuring revenue stability.

 

Get in touch!

Florian Wilhelm

Head of Data Science, Contact for Data Management & Analytics