Spotify
algorithmic discovery playlist
Discover Weekly, Spotify's personalized weekly playlist launched in 2015, has accumulated more than 100 billion tracks streamed and gets its users listening more than twice as long.
Key points
- Weekly playlist of 30 discovery tracks, regenerated every Monday with no action.
- In-house collaborative filtering and audio analysis on listening history.
- More than 100 billion tracks streamed, listening more than 2x longer.
- Evidence B, confirmed status, feature active since 2015 (10-year press release).
Objective
Make music discovery automatic and effortless, to anchor the habit of opening Spotify every week and increase listening time and loyalty against competing platforms.
The deployment
Every Monday, Spotify delivers a playlist of 30 tracks the listener has not yet heard but is likely to enjoy. The system combines collaborative filtering (what similar profiles listen to), audio content analysis, and listening history. The playlist regenerates every week, with no action from the user. Since 2025, the listener can filter Discover Weekly by genre.
Results Proof B
Figures published by Spotify (10-year newsroom, and Spotify advertising unit) on a feature in production since 2015. Direct figures from the subject brand, cumulative volume verifiable over time, without going through financial results.
How it works
Documented architectureThe stack in detail
- outil Filtrage collaboratif in-house matching profiles by co-listening to predict tracks never heard but likely
- outil Analyse de contenu audio in-house audio features of tracks, complementing collaborative filtering where co-listening is missing (new releases, niches)
- infra Pipeline batch hebdomadaire regeneration of the 30-track playlist every Monday for each listener, with no action on their part
How it runs, concretely
For ops teams-
1Listening aggregation data team
The week's listens, followed playlists, and skipped tracks are consolidated per profile.
-
2Candidate generation AI
Collaborative filtering and audio analysis produce a list of tracks never heard but likely.
-
3Playlist assembly AI
The system selects 30 tracks, ordered for the listener, delivered on Monday.
-
4Feedback loop human / AI
What the listener plays, saves, or skips in the playlist refines the following week.
The listener's listening history and co-listening patterns between similar profiles. Without recent listening data, collaborative filtering has nothing to match the listener to other profiles.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- per-user listening / consumption history
- catalog with audio features or metadata
- sufficient user volume for collaborative filtering
Org prerequisites
- data science team
- reliable batch pipeline
- product surface to expose the playlist
Possible stack
- custom/in-house
- managed recommendation engine
- weekly batch job
The plan, step by step
- Step 1Check the volume of signal: enough cross usage (active users, co-consumption) for collaborative filtering to produce candidatesDeliverable: Quantified feasibility study, go/no-go
- Step 2Consolidate histories: listens, saves, and skips per profile in a reliable pipelineDeliverable: Per-user dataset, refreshed every week
- Step 3Build the candidate model: collaborative filtering + content features to score items never consumedDeliverable: Scored candidate list per user, evaluated offline
- Step 4Assemble the playlist (30 ordered items) and deliver it in a dedicated product surface, on a fixed dayDeliverable: Weekly playlist in beta on a segment
- Step 5Close the feedback loop: re-inject listens, saves, and skips, measure retention and listening time against a controlDeliverable: Active learning loop and impact reading
First step: Confirm there is enough cross listening volume for collaborative filtering to produce relevant candidates.
The same pattern, proven in other industries
Boots
machine learning scoring of offer-usage propensity, applied to mail and the app
Michaels
genAI personalization of marketing messaging (copy) at scale, tuned to the brand voice
Nike
personalized exclusive offers and access based on member engagement and history
Sources
- S1 Discover Weekly Turns 10: Celebrating 100 Billion+ Tracks Streamed (Spotify Newsroom) Interested party archive pending
- S2 Five Years of Discover Weekly (Spotify Advertising) Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.