AI personalization – fact or just more hype on hype?

Wouldn’t it be nice if a web store could propose the exact products you want? Or your online newspaper had news and TV series you are actually interested in? Or a user interface adjusted automatically to your requirements? These have been ideas I have heard many times during my 18 years in the data and analytics business. The problem is that those terms are mainly used by people who don’t know what they are talking about when talking with other people who won’t admit they don’t know what they are talking about.

To simplify, personalization is typically based on one, or a combination of three things:

1. what preferences you indicate (e.g. tick boxes);
2. what you typically buy and how you typically use a service; and
3. what other similar people buy and how they use the service.

But none of these are as simple to realise as one would think.

When you ask preferences from people, most tick all boxes or no boxes, they either don’t concentrate or know what they really want. And if they indicate their preferences today, there is no guarantee they will match tomorrow’s preferences. 

Models can learn from yours and other similar user’s preferences. The system then starts to offer specific offers to you, that you may or may not use to buy those specific things. This in turn reinforces the system believing that you are interested only in those things. It narrows the options and offerings to you and in turn misses many things. The same happens in services like Facebook, and how it selects which people and posts to show in your daily feed.

Another angle is that the system doesn’t even try to serve or help you better. It just tries to maximize sales or keep you engaged in the service. It offers you products and content that you are likely to buy or click. It focuses on maximum, short-term monetization. 

These issues are not new. People who work with personalization, machine learning and analytics have talked about them for over ten years. But it doesn’t stop many people dreaming about personalization, putting it in their business plans and presenting it as a key use case for ML and AI.

It is not impossible that personalization could be more useful and one day we will have really valuable personalization that actually helps users. But it needs much more than what many solutions and business plans offer today. A fundamental starting point is to really understand, what people want to do and achieve in each use case. It is much harder than optimizing some clicks or processes.

Let’s take some simple examples:

• People still like to browse printed papers and online publications, and not simply focus on articles selected for them – a significant part of the browsing experience is to find things you were not specifically looking for or expected.
• Customers also like to walk around in physical stores and browse many products on the Internet, just to see different choices, get new ideas and pass time. Then there are other situations when they just want to make an immediate and specific purchase.
• Your earlier purchases or actions don’t necessarily mean you want to do the same things again.
• Movies and TV series that are recommended to you based on earlier watching behavior may not really be what you want to see, especially when based on a simple categorization of the content that doesn’t really understand your preferences. People can experience similar content in many different ways, which is completely beyond a simple ‘tags’ choice tree. 

These are simple examples, but they illustrate how personalizing an experience is not a simple algorithm achieved by optimizing a few variables. The system should know your preferences now, your state of mind, the real reasons why you have done something earlier, and it should be there to help you, not just to sell you products and services.

Personalization and AI are terms that have been diluted with stupid use and marketing of the terms. Both of them will be very important in the future. But many existing solutions and especially business plans are crap. They are crap produced by people who don’t really understand people’s needs and technology, but love to give the impression that they understand both things.

There is no simple solution to change the situation for the better, but there are certain things that would help, for example:

1. AI and personalization services that work for people that help to make the experience better based on personal data, not just to sell you more or hook you to a service.
2. Models and analytics should be based on richer use of data, not only analyzing actions in one service, but putting them properly in context and anchoring them to your characteristics.
3. Use proper terms for things, e.g. sometimes optimization of the buying process is a more honest and correct term than a fancy AI-based personalization,
4. When people market AI personalization, try to dig deeper into what their system can really do with difficult and concrete questions. Don’t accept statements like, “it is amazing, how great algorithms developed by very smart guys can automatically find you the best options,”
5. And, at least, try to analyze your own behavior in different situations, and see, what kinds of personalization could really help you in daily situations.

​Of course, there are smarter and smarter systems all the time. People are getting worried that AI knows everything about them and can utilize all that data. A system can have too much of your sensitive data, but often systems are more stupid than people expect. Real development happens with solutions that offer specific solutions for specific needs, not with those big plans that claim to solve all needs with big data and general personalization algorithms. And if it is your data in a system you can manage and that works for you, then at least you are represented and know the incentives the ‘intelligence’ works toward.

​