Return to site

Recommendation systems within music platforms strategy

After investigating music recommendation from the perspective of music platforms product, listeners behaviour, recommendation systems design, we are going to consider where recommendation fits within music platforms overall strategy.

  • Beyond an initial strategic positioning, the ultimate shape of music service product is the result of various factors, to name a few:

    • Platforms international expansion race: localizing content means not only sourcing catalogues, but also clearing all the rights, getting metadata relating to the content, and culture specific curation, hence the concept of MVL (minimum viable localization)

    • The cost of content: a smart radio format costs less than on demand format. For example Deezer Flow is smart radio (it is not possible to get more than 6 skips, to change position within the track or see the next songs in advance), whereas Spotify Discover weekly is an "on demand" format.

    • Niche versus global product: platforms can target a worldwide presence, all types of listeners, extended range of products (not only audio music content but also podcast, live concerts,...), or niche focused market segments or stand alone apps.

    • The way to measure success: the type of KPIs used to measure success (user base growth, subscription conversion, longest sessions, user satisfaction (short and long term), net promoter score (NPS), customer effort score (CES)), and the way alternative scenarii are tested (focus groups, A/B tests,...),....are key drivers of the evolution of platform products. Deezer claims to look at maximizing songs addition to user library, Spotify more looking at longest sessions, two different ways to measure satisfaction.

For listeners it is difficult to get a sense of the depth of a music catalogue rich of tens of million tracks. Providing a wide area of interface helps them perceive this value.

As seen in an earlier post, the main recommendation system today has the following caracterics : not personalized playlists, updated once a week, to which people subscribe. Those playlists are numerous, covers most day-to-day situations, and are to a significant extent redundant (same songs in different playlists, same mood for different playlist names). Interfaces have become overloaded and less readable.

It is interesting to try to explain why the market has come to this situation. For listeners it is difficult to get a sense of the depth of a music catalogue rich of tens of million tracks. Providing a wide area of interface helps them perceive this value. In our view, the race to convert rapidly the maximum of listeners from various regions at an acceptable cost led platforms to provide an extensive UI, contextualised to mundane data-to-day situations, satisfying a large share of the potential audience. It may have even led people usually indifferent to music to subscribe, which by itself is an achievement.

Hence integrating sophisticated recommendation systems were not platform top priority. Differentiating listeners and providing them with discovery and diversity were only elements of user satisfaction, which itself bears a loose relationship with high level KPIs like user base growth.

Going forward, there will be less room for many players pursuing global strategies. As competition heightens, there will be a need for differentiation, and filling the recommendation gap will be one response

Going forward, there will be less room for many players pursuing global strategies. As competition heightens, there will be a need for differentiation, and filling the recommendation gap will be one response:

  • Personalisation: poor recsys at adapting to each types of listener (repetition need, diversity need,...), and adapting to the appropriation cycle (discovery, high frequency, saturation), popular/little known artists and expert/mainstream

  • Recommendation of contents other than music recordings: lyrics (based on text similarities), artist news, new releases, geolocalised and personalized recommendation of concerts,...

  • Less redundant and more simple UI

  • Better exploration UI : helping listeners to explain where they stand in their exploration navigation (summary of historic exploration versus to be explored,...)

  • Better onboarding UX

  • Features allowing to get surprised, to measure the extent of narrowness/adventureness

  • Providing playlist both with a context (activities, mood) and personalized to listener music universe

  • Recommendations of curators, of playlists

  • A proper dialogue with the listeners, with a narrative progression

  • If use collaborative filtering, mitigation of its majors drawbacks

  • More immersive fan / artist interaction UX

  • If use of content other than music recordings (podcast,...), not just adding a new category but Integrating it in the listening experience

  • Dashboard for editorial teams: emerging artists, song properties (moods, genres), what people say about artists, songs, genres,...

Were one of these suggestions relevant to your situation, feel free to contact us for advice.

All Posts
×

Almost done…

We just sent you an email. Please click the link in the email to confirm your subscription!

OK