For the last couple of months, I have devoured science fiction books – Asimov, Banks, you name it. I’m a big fan of e-books and audiobooks, but I'm annoyed by the fact that I never seem to get any good recommendations for my next read or receive recommendations for other genres, like fantasy.
Sci-fi or centaurs? It’s a question of revenue
By now, we’re used to receiving spot-on content recommendations from a few different entertainment providers. Spotify, for example, is becoming a user’s musical soulmate with features such as Discover Weekly. But, great recommendations are not to be taken for granted – even if our expectations are on the rise. How have entertainment providers like Spotify managed this, while others desperately lag behind?
The short answer is data.
Spotify was quick to implement and improve its recommendation system, allowing them to deliver amazing experiences for customers while getting ahead of their competition.
Implementing recommendation engines is becoming a necessity for content providers wishing to remain competitive in this day and age.
1/ Back to basics: Think customer-first
Business is about giving the customer what they need. Guess what? Recommender engines have the power to do so automatically. And the big tech organizations have taken notice. According to a McKinsey report, the best recommender systems are true revenue generators. For example, 35% of what consumers purchased on Amazon and 75% of what was watched on Netflix in 2013 came from users following the organization’s product recommendations.
These systems are generating revenue since they provide the customer with what they need – before they know it themselves. Fancy that, you no longer have to spend time and energy thinking about what you might want or what your next action is. Instead, they’ve done this for you. This has the potential to significantly increase the overall customer experience of your product , creating customers who are not only more engaged with the content but are more likely to keep using the product, more often and for longer periods of time.
2/ Abundance: the new normal in society
We live in a world of abundance. While this may sound ideal, it can be impractical in our day-to-day lives. We all know the feeling: you walk by a bookstore and decide you must go in to buy a book. While in the store, you see a book your friend recommended for you. As you pick the book up, you realize that next to it is a book on a topic you so passionately read about last weekend. As you choose the new book instead, on your way to the cashier you walk by a shelf with a 3-for-2 promotional offer. Now, you see this as an opportunity, “Why pay more for less?” But as you skim through the shelf, you realize it’s close to impossible to choose only three books out of the 15 available. All of a sudden this whole process has taken up 15 minutes of your time, and in the end, the only thing you leave the bookstore with is a headache.
The amount of choices available to consumers is only increasing. An average bookstore carries approximately 3,000 books in total, with eight operating hours per day. Compare this to the digital environment, where Amazon offers 30+ million unique books, and you have access to it 24/7 from your mobile device. Daunting, I know. This has been coined the paradox of choice for customers. Interestingly, the term arose 15 years ago, a time before smartphones.
And books are just one example. Abundance is the new black in our modern world. A pivotal question to business leaders is how they will assist their customers as they navigate through it.
3/ Tools to get your organization started have never been easier to use
Recommendations to customers can be based on either their past actions or from the content itself. If large amounts of usage data about past actions are available, then it is rather straightforward to build collaborative filtering system similar to what Spotify and Netflix have. However, most companies do not have large amounts of usage data, especially for new users and new, or bespoke, content. As an alternative, a content-based recommender system does not require any usage data, just data about the content itself. To build a content-based book recommender system, we need an AI-model that can tell the difference between different types of books. At Peltarion, our platform makes it easy to build AI-systems like this, for data types including texts, images, and audio. For instance, we have been classifying images of clothes for large retailers and sorting music according to mood. In response to the poor recommendations I received for sci-fi books, I built my own book genre classification model with data taken from book summaries (and get this, the model actually does a great job distinguishing science fiction from fantasy). You’re welcome.
There are some technical steps needed to turn the book classifier into a recommender system, but in principle, it is rather simple. We can make use of the classifier to obtain a so-called embedding, i.e. a vector of numbers describing the characteristics of each book. We can use this embedding to get an understanding of what books the user has previously been interested in, and then recommend similar books according to these embeddings.
Today’s customers want and need recommendations and deep learning can be an essential tool for helping businesses acquire the proper data.
Successfully implementing a recommendation engine can lead to:
- Improved customer satisfaction
- Increased customer engagement
- Reduced labor costs/repetitive tasks from employees
- New moat opportunities
02/ More on Applied AI & AI in business
Build a no-code app to start using your AI model
What do you do once you’ve built an AI model in the Peltarion Platform? Build an app around it of course!
Defect detection in mass production using images
Can a pen from the future change lives for children with eczema?
10% of the world’s population. That is the number of human beings who have atopic dermatitis, commonly known as atopic eczema. Simon Grant, CEO of Swedish SciBase, believes that medical technology and deep learning can change this in the near future.
Three ways in which a machine learning project differs from your regular software project
In your traditional software project it's much easier to make statements of what is doable. Machine learning projects require experimentation. Much more so than you might expect.
From idea to AI powered applications, the challenges (part 2)