Lesser-known artists get a fair first listen. They just don't get a second one.
A liked first listen of a lesser-known artist usually leads nowhere. About 8 in 10 aren't heard again within two weeks, and listeners rarely go back on their own.
Second Listen. After two quiet weeks, bring the artist back once, with a different song, in the radio or mix where the listener first heard them.
Five songs: two of Ena's picks and three lesser-known artists from the sample. Skip one in under 30 seconds and the phone marks it as a skip, the way this study counts one. Phones and logged-out visitors hear a preview, so Full song ↗ opens the whole track in Spotify.
I expected popularity bias: big artists pushed harder, lesser-known ones skipped more. The data pointed somewhere else.
Two terms used throughout: a lean-back listen is a song that played on by itself (radio, autoplay, or a playlist carrying on) rather than one the listener picked. Lesser-known means in the bottom third of artists in the same genre (or under 100,000 Last.fm listeners when the genre isn't known).
Open music databases undercount African artists.
Afrobeats artists one global Last.fm cut-off would call lesser-known. Burna Boy came out "mid".
How much Last.fm undercounts African artists against Western ones, checked by hand against Spotify for 30 artists
Rank correlation: the order is still right within a genre, so ranking artists within their own genre fixes most of it
A lesser-known artist's first listen is skipped more
Not supportedSpotify's lean-back introductions lean towards big artists
Not supportedLesser-known artists stick less after a first listen
Supported · 10 of 10Listeners go back on their own to new artists they liked
Rarely · 1–2%The question changed from "is discovery unfair?" to "why does a fair introduction lead nowhere?"
Add everyone's plays together and lesser-known artists seem to win by 10 points. Look inside each listener and the advantage mostly disappears.
Gap = big minus lesser-known skip rate, in percentage points (pts). A plus means big artists were skipped more.
"Would even give me more reason to go."
An interviewee, on an artist with very few monthly listeners.
"I don't equate quality music to more famous songs."
Another interviewee, asked whether they check an artist's monthly listener count. All four said a small count doesn't put them off.
"Stuck" means still being played, not skipped, 30–90 days after a lean-back first listen. For every single listener, big artists stuck more often.
Gap = big minus lesser-known, in percentage points (pts). For L04, 25.2% of big artists stuck against 8.2% of lesser-known ones: a gap of about 17 points, shown as +16.9 because it uses the unrounded rates. A plus means big artists stuck more, and every listener is a plus.
A bouncy, stripped-back pop track made from the data. Each listener gets one bar: the stabs are their lesser-known stick rate, and the lead bends up into their big-artist rate. Higher stick rate, higher note. Notes snap to a minor scale, so L06's tiny 0.6-point gap sounds level. The other nine all climb.
Lesser-known vs big artists that stuck, everyone's plays added together
Typical gap for a single listener (median)
Listeners show it. Odds of that by chance: about 1 in 500 (p = 0.002)
The gap when any one listener is removed. It never disappears.
Follow 1,000 first listens the listener liked: played through, not skipped. Step through what happens next.
These figures leave out the one listener whose background listening makes "liked" unreliable. Under the proposed trigger rules they are 3.3% against 7.1%. Checked listener by listener, the gap shows up for all 10 (typically 5 points; about a 1-in-500 chance of that by luck). This is correlation, not proof: that's what the experiment is for.
of liked lesser-known first listens got a click back from the listener (1.9% for big)
points of lean-back returns were the same song replaying
Even when lesser-known artists came back, few reached a different song
"Another popular" one.
A self-described loyal listener, on what they want after a popular song. A lead, not evidence: I'd raised popularity bias myself.
"At the very least, I'm likely to click on it."
On a new artist framed as similar to one they love. Also a lead, answering an example I gave.
When a liked lesser-known artist hasn't come back after two weeks, bring them back once, with a different song, in the radio, autoplay or mix where the listener first heard them.
Half of listeners get Second Listen, half don't, chosen at random. Success means more listeners go on to play a different song by the artist 30–90 days later. Today that happens 3.3% of the time.
Listeners needed in each group of the test (one group gets Second Listen, one doesn't)
Extra times a month a listener goes on to a second song by a new artist, per 1 million listeners like the typical one here
These are rough units, not a forecast: the 10 listeners here listen far more than most people. Technical detail: from sizing.py, 1.2 triggers per listener a month, baseline 3.26%, 80% power, intra-class correlation planned at 0.02. The 90 days of waiting, not the number of users, set the timeline.
| Type | Metric |
|---|---|
| Primary the main success measure | A different song by the triggered artist, played and not skipped, in days 30–90. The re-introduced song doesn't count. |
| Leading · 30 days an early sign, before day 90 | Something the re-introduction can't produce by itself: a return the listener starts, a save, or a follow. |
| Guardrails must not get worse | Listening time per session, skip rate on re-introduced tracks, hide rate. Reported separately for loyal listeners. |
| Counter-metric checks for side effects | Whether other artists in the same radio, autoplay and mixes lose out. |
| Decision rule | Ship if the primary metric rises and no guardrail breaks. Iterate on timing or song choice if only the leading metric moves. Stop if re-introductions are skipped more than the surface's normal rate. |
Most ideas for helping small artists act on the first listen. The first listen is already fair, so they add introductions to the same leak.
| Option | Acts on | Impact | Decision |
|---|---|---|---|
| A. Re-introduce liked-but-lost artists with a different song | The gap | High | Build and test |
| B. More lesser-known artists in first exposure | The first listen | Low | No: more introductions, same leak |
| C. Hide monthly listener counts | The first listen | Low | No: interviewees already ignore them |
| D. Genre tags on tracks | The first listen | Low | No: 1 of 4 interviewees wanted them |
| E. Album-based mixes instead of artist radio | Relevance in general | Medium | Backlog: not specific to small artists |
| F. "Follow this artist?" prompt after a liked listen | The gap | Low | No: clicks back are rare |
Spotify already sells artists a way to reach listeners who met them passively. Second Listen does it organically, for every lesser-known artist, in the two weeks that matter.