My product management career has been shaped by my experiences in developing personalised products. In this talk for #MTPEngage Hamburg, I walk through my experiences, failures, and learnings, and share what personalisation in product has taught me so far.
Content-based Recommendations Versus Demographics
Personalisation only takes you so far on an ad-driven platform. Your safest bet is to provide recommendations on what to watch or read next.
In advertising, we risk putting people into demographic buckets to fit clients’ needs and create distrust with our users. Why not focus on the content itself? Even while volunteering for ProductCamp Berlin, I learned that the best sponsor deals were the ones that matched the brand and delivered delight to attendees.
Collaborative Filtering: When you Have Plenty of User Data
A hybrid recommender system works well when – as in the case of Deezer – you have 14 million users who consume a lot of content daily. Together with the user research department, a product designer, and our data science and development team, we revamped the user onboarding to match both the product experience better and to improve the first Flow experience. Flow is a personalised radio feature; however, there’s not enough data for effective personalisation for a newly registered user, so onboarding plays a vital role.
With personalisation, a first experience sets the tone.
To improve the listening experience, we interviewed users in different cities, asked for their feedback, and came up with personalisation concepts inspired by “artist communities”. These artist communities are clusters of artists who are often listened to together by similar sorts of users. We used a combination of collaborative filtering and content-based tags to improve the experience. Today, Flow is more than just a radio feature and provides personalised tracklists for any mood.
Build Trust and Engagement: Recommendations as a Service
Great recommendations start with content and user knowledge. We all appreciate discovering something new, but how much data are we willing to share? The answer is “only what is necessary”. I believe that to make recommendations more trustworthy you should use natural language to explain the why behind them, and allow users to share feedback. At Spideo we specialise in recommendations as a service, working with creative industries and developing the tools that can build such a dialogue with semantic data.
Personalisation at the Core of the Product Experience
Explanations in the name of transparency should always be helpful. For example, 10 pages of terms and conditions do not help me to understand how my data is being used, whereas being able to provide explicit feedback does. As product people, we should take the transparency test: am I withholding information to maximise understanding, or would more information help the user improve their experience? It’s about good product ethics and “kind technology”.
We need to keep in mind who we are building products for: humans.
The word personalisation means tailored to an individual, so let’s be “human-driven” first.