Independent Product Teardown
How I approached it
All three options are RICE-scored, not just the one I shipped — and the option that scored highest is the one I rejected. Raising the skip cap scores 8.1 and would work; it also removes the exact constraint that makes Premium feel necessary, a strategic cost the formula doesn't capture. I'd rather state that out loud than quietly score it down to produce the answer I wanted. The fix I did ship has two parts because one part alone would have failed its own primary metric: a better-targeted upgrade prompt converts some users, but almost everyone declines, and for all of them the dead end I logged as F4 would still be a dead end. Every quantitative claim here is traced to a stated assumption or a named source, including one apparent four-fold contradiction in my own baseline that I've reconciled rather than left sitting.
The Job
"When I'm bored with my usual rotation and don't know what I want to hear next, I want Spotify to surface something new that actually fits my mood, so I end the session having found at least one track worth keeping — without spending real effort searching for it myself."
The User & Context
Persona: a bored weeknight listener — mentally checked out of work, under 30 seconds of patience before disengaging, on the mobile Home tab, free ad-supported tier. Alternative if this fails: falls back to a familiar playlist (zero discovery), or opens a competing app instead.
Journey Map — Discover Weekly
8 steps mapped from app open to the skip-limit wall: orient → open Discover Weekly → play → skip tracks 1–5 (low friction, expected cost of exploring) → skip 6 triggers the wall (high friction, task-blocking) → generic limit modal with no path back into discovery (high friction, dead end) → track resumes with a mid-roll ad (medium friction). Six taps total to hit the wall — this number carries through to the before/after comparison.
Friction Log — Four Frictions, Each Scored
Severity
Frequency
F1 — Skip-limit wall stops exploration
4/5
High (~35% of sessions)
F2 — Mid-roll ad covers the player
3/5
Medium (~20–25%)
F3 — No visible reason for a recommendation
2/5
High (most sessions)
F4 — No recovery path when the wall appears
3/5
High (~35% — every session that hits F1)
F1 and F4 share the same ~35% frequency because F4 is by definition the state a user lands in the moment F1 fires. They differ in severity, not in how often they occur.
Root Cause — A Deliberate Friction Ceiling, Not a Bug
Skip limits and ad interruptions fund per-stream royalty costs and make Premium feel necessary — if free listening felt unlimited, there'd be little functional reason to upgrade. The cap is tolerable for casual playlist listening, but noticeably painful for exploratory listening.
The Real Problem Statement
"Users in this moment fail to complete music discovery because Spotify treats exploratory listening the same as playlist listening — which costs the company weaker recommendation signal quality." The user falls back to a familiar playlist or leaves; the business receives weaker exploratory signals, which can degrade recommendation quality over time.
Three Options Considered
- A — Raise the skip cap (Rejected): removes the exact lever that makes Premium feel necessary, eroding the conversion funnel with no revenue offset.
- B — Earn extra skips (Rejected): bonus skips for rewarded ads or saves add engineering and trust complexity, are gameable, and only delay the wall.
- C — Discovery-Context Wall, two parts (Shipped): when the wall fires inside Discover Weekly, Radio, or Release Radar — (i) show a contextual trial prompt instead of the generic limit modal, and (ii) for everyone who declines, offer a one-tap route into a different Radio station or mix tuned to what they were skipping toward. No extra skips, so the cap is untouched, but the session continues inside discovery instead of ending at a dead end.
Why part (ii) exists: part (i) alone converts a slice of users to Premium, but almost everyone declines a trial prompt — and for all of them the dead end I logged as F4 is still a dead end, so Discovery Save Rate wouldn't move. Shipping only the paywall would have meant identifying the recovery-path problem myself and then not fixing it. The two parts together serve the user job and the business lever rather than trading one against the other.
The RICE Math — All Three Scored, Not Just the Winner
Option
Reach
Impact
Conf.
Effort
RICE
A — Raise the skip cap
3
3
90%
1 pm
8.1
B — Earn extra skips
2
1
50%
4 pm
0.25
C — Discovery-context wall + recovery
3
2
80%
3 pm
1.6
Why the highest score loses. Option A scores 8.1 and I still rejected it — because RICE measures effort, not strategic cost. Raising the cap directly removes the constraint that makes Premium feel necessary, and that cost doesn't appear anywhere in the formula. This is the case where the framework has to be overruled and the reason stated out loud, rather than quietly scored down to produce the answer I wanted. Option B is rejected on the arithmetic alone. Option C's effort rises from 2 to 3 person-months once the recovery path is included — a real cost I'd rather carry than ship a fix that doesn't move its own primary metric.
Tradeoffs & Pushback
- Growth / Platform Engineering: scope creep from tagging every discovery entry point → start with Discover Weekly and Radio only.
- Growth / Paywall Owner: a second paywall could create prompt fatigue → run a holdout A/B before rollout.
- Trust & Design: the prompt could feel like penalizing curiosity → frame it around continuity, not restriction.
- Data Science: lift may reflect creative refresh rather than context → isolate the exact discovery-wall trigger.
Phased Investment — De-Risking the Bigger Bet
Phase One
Ship both halves on Discover Weekly and Radio only — contextual prompt plus recovery path — then run a 50/50 holdout A/B.
Phase Two
Only if Discovery Save Rate moves without depressing ad revenue, extend surface tagging to Release Radar and the rest of discovery.
If Phase One Is Flat
If save rate holds flat even with a working recovery path, that's evidence the skip cap itself — not the dead end — is the binding constraint, and the argument moves to Option A with data behind it.
Success Metrics
Primary — Discovery Save Rate: ~15% today (working estimate) → 20%+ target
Secondary — Session continuation rate after the wall fires
Guardrail — Ad Revenue per Free User: must not decline
Counter — Premium Trial Starts: isolated via holdout
The secondary metric exists specifically to measure part (ii). If trial starts rise but session continuation doesn't, I've built a better paywall and not a better product — and I'd want to see that split rather than read a single blended number.
Before / After
Before: 6 taps to reach the skip wall, then a generic dead-end modal — the user must navigate separately to Premium, or abandon the session. After: the same 6 taps, then a modal offering two live routes — one tap to start a free trial, or one tap into a related Radio station. The session ends inside discovery either way.
References & Data Notes
Official Spotify sources: Spotify Q1 2026 earnings (newsroom.spotify.com); Spotify Technology S.A. financials and shareholder reports; "Discover Weekly turns 10" (Spotify Newsroom, Jun 2025); Spotify discovery-playlist coverage (Spotify Newsroom, Jul 2026).
Secondary sources — provenance caveat: Intellijend (2024), AMRA & Elma (2025), Axis Intelligence (2026) and VaaSBlock (2026) are secondary aggregators, not primary research. I've used them for order-of-magnitude context only, never as the basis for a specific claim.
On the 15% baseline — reconciling an apparent gap: Intellijend reports save rates of roughly 2–4%, which looks like it contradicts my 15% figure. It doesn't measure the same thing. Their figure is a catalogue-wide save rate across all listening; mine is a within-session save rate for users who have actively opened a discovery surface — a far smaller and much higher-intent denominator. I've kept 15% as a working assumption on that basis, but it is an assumption, and the first thing I'd replace with real instrumentation. The 20% target, friction frequencies, and session-volume estimates are working assumptions on the same footing, not Spotify-reported internal metrics.
