*Franklin Horn | 12 Jan 24*

Welcome to the Personalization Glossary! At Skillshare, I was a founding member of the personalization team, and I started communicating with stakeholders about the vision and roadmap for personalizing the home and search experiences. I realized that I was speaking a foreign language with some key concepts (e.g. content vs. collaborative filtering) and so I started defining these terms in a specialized Personalization Glossary. My engineering, product, and design counterparts and my stakeholders alike really enjoyed visiting and reading the glossary. One product peer wrote:

This is one of my favorite pages on the internet at this point.

My users had spoken, and clearly a small but passionate audience might enjoy this knowledge, so I wanted to share this information more broadly. Let’s take a dive into the fascinating world of tailoring experiences to fit your unique preferences. I’ll keep it light hearted and touch upon familiar experiences like how Spotify customizes playlists to receiving tailored movie recommendations on Netflix. Personalization is all about making things just right for you. In this glossary, we'll explore concepts like microgenres, cold-start recommendations, taxonomy, and the intricate art of understanding user needs.

Personalization

The act of customizing an experience based upon knowledge of the user. A simple example is asking for a user’s name in a sign up form. Then later, displaying the user’s name inside the product:

Screenshot 2023-12-01 at 10.02.11 AM.png

The user sees their name, they immediately become attracted to the section, and they recognize that Spotify has created some specific playlists for them. A user is probably thinking to themselves, “Ah, Spotify knows who I am. That’s good, I’m logged in correctly. What have you created for me today?”

Personalization can also be described as a two-part feedback loop. The user provides the system with explicit and implicit signals. The system takes these signals and provides a relevant and meaningful output inside the experience, completing the loop. The explicit signal is my name, “Franklin Horn” and it’s explicit because I directly articulated the information to the product. I have been listening to a lot of jazz lately, so my listening history is an implicit signal that Spotify personalization identified and created a Daily Mix playlist of jazz greats like Oscar Peterson.

We can power up personalization experiences with simple variables and mail merge features, to using advanced machine learning techniques. Both types qualify as personalization it’s just that one of them required a significant recommendation engine to identify the user, their interests, and generate the output. It is critical to know what personalization experience you seek and consider the simplest applicable personalization technique.

Recommendation

A suggestion to the best course of action, particularly when there’s a large set of possible choices. When you ask your friend for the best places to eat in Los Angeles, you are explicitly asking them for a recommendation. When a subscriber arrives on Netflix’s homepage, they are implicitly asking for help to find the right class. The larger the set of possible choices, the more likely someone could use a recommendation. In Los Angeles there are tens of thousands of restaurants, though you can only eat at one for dinner time. On Netflix, there are thousands of titles, though you can only watch one at a time.

Cold-Start Problem

Personalization systems rely upon reference points, like user interests, to make recommendations. When a new user registers for a product, there is typically little to zero reference points about them, making it challenging to provide a recommendation. This is known as the cold-start problem.

Let’s continue using the restaurant analogy to illustrate the case. Imagine you’re visiting a close friend in Los Angeles, and you ask them for the best places to eat. Your friend knows something about where you’ve eaten before, perhaps you’ve both shared a delicious meal together. They have a sense of your tastes in food, what cuisines you particularly enjoy, your preferred ambience, and whether if you have any dietary restrictions. Your friend takes a moment, ponders these past experiences, and calculates all these factors together so they can provide a recommendation.

What if we removed your friend from the scenario. Let’s re-imagine that you’re visiting Los Angeles, just that it’s a business trip for the first time, and you ask the same question of a hotel concierge. When you ask a friendly-yet-complete stranger for the best places to eat, they know nothing about your tastes or past eating patterns. They have a “cold-start” problem.

The stranger has a few strategies. They can simply give you a popular restaurant or a nearby restaurant. They can look at you and make a guess about the kind of person you are (judging your appearance, deciding if you are wealthy or not, and giving you a restaurant between $$$ and $$$$$). They might also ask you a series of questions to better understand you.

What kind of cuisine are you looking for? Do you have any dietary restrictions? Are you dining alone or with someone?

Your answers to these questions give the concierge some reference points. This conversation is akin to an onboarding flow, like this one from Spotify: