The internet enables companies to move from ‘broadband’ interactions, where everyone receives the same content and message, to 1:1 interactions. With 1:1 interactions, every online experience is dynamically tailored to the unique interests of a user. Using machine learning powering personalization, the consumer experience will dynamically learn, communicate, and customize based on a users actions.
Computing power and new technology is only now beginning to unlock this potential. For example, in the future, a CNN.com or ESPN mobile app visitor will see different content, a different UI layout, different UI color-schemes, and different ad placements. The UI of every site, app, and connected experience will be tailored to each user. Users will have immediate access to the information they care about and the experience will change in real-time, and it’s pretty darn exciting.
The rise of mobiles makes personalization even more salient. Mobile devices always with us, making them extensions of ourselves (pretty darn personal). However, their limited screen real-estate mean that every pixel of real-estate matters even more. Companies better choose the right the experience for each user to keep them engaged.
So what’s holding up the personalization vision? Why aren’t intelligent agents customizing your online experiences across all sites? Personalization requires a couple key components – (1) algorithms to create accurate and actionable user descriptions, and (2) tools which allow anyone (even a non-techy) to access these user descriptions and communicate with the user. Unless they are Facebook, Google, or Netflix, most organizations don’t have the internal engineering resources for (1). And to date, no one has made (2) a reality.
Here are a few more specific challenges (and opportunities) –
Generic Learning Frameworks
There are a few options for using personalization technology based on a generic ‘collaborative filtering’ (CF) algorithm. Generic collaborative filtering just won’t work for most problems. The challenge here is that most personalization problems require domain-specific knowledge. Consider a site with premium video content – both movies and episodic (e.g. series). The machine learning-powered personalization technology in this case would require knowledge of both episodes (and likely associate a different ‘meaning’ with them) and movies. Not taking this information into account will dramatically impact performance and the UI experiences generated. Or, take a site with time-sensitive content. The temporal dynamics of content availability will dictate placement in the UI and frequency of presentation. Finally, consider a site where a user visits hourly vs. a site where a user visits once a month. The visit cadence of a user requires different dynamics around content rotation.
Here, we need generic personalization frameworks which take into account problem-specific nuances – and this is hard.
Easy Personalization Tools
Personalization technologies are tech-heavy and contain many machine learning algorithms. The broad adoption of machine learning will happen when we make products that make machine learning pipelines actionable and accessible for everyone.
Personalization, implicitly, requires an understanding of users, and many organizations jealously guard user data. These businesses are afraid their competitors will leverage it, and information is therefore stored in different schemas making it hard to integrate different sources. The best technologies will both optimize every bit (in the literal sense) of information on users and provide easy mechanisms for integrating 3rd party data.
Dynamic real-time personalization of consumer experiences is one of the most exciting areas of online consumer innovation. There are lots of challenges but smart people are hard at work across academe and industry. It’s going to be a wild ride the next few years.