product sense · hard
"How would you improve Spotify's engagement?"
How would you improve Spotify's engagement?
The failure mode on this question is proposing a better recommendations engine. Spotify’s Large Taste Model processes 3.4 trillion daily taste signals and is already the company’s primary competitive asset. Recommending it as an improvement signals no product knowledge. Every interviewer at Spotify knows this, and they use it as a filter in the first minute. The question is actually asking something more specific: given that personalization is already working at scale, what is the next engagement surface that makes Spotify stickier, more viable as a business, and genuinely worth returning to on a Tuesday morning with no specific song in mind?
Start with a clarifying question before any framework. “Engagement” is underspecified. There are at least three distinct goals: depth (listening time per session), breadth (DAU/MAU ratio), and stickiness (days listened per month). Each points to a different user problem and a different feature. Naming this distinction is the first signal that you understand engagement as a metric problem, not just a product one.
What Spotify’s engagement actually looks like in 2026
Spotify has 761 million MAU and 293 million premium subscribers as of Q1 2026. The free-tier base is the engagement opportunity: a personalized free experience rollout in the US is already moving days listened per month in that cohort, per the Q1 2026 earnings call. Wrapped 2025 generated 620 million shares (the largest social sharing event in streaming), which tells you something concrete: Spotify wins when users feel seen and self-express, not just when the algorithm surfaces a new artist.
Three engagement failure modes exist and are worth naming because they require different interventions. Cold-start abandonment: new users who never build a taste profile deep enough for recommendations to click, and leave before the product gets good. Lean-back passivity: users with 3-4 years of history who open the app only when they already have a song in mind, use it as a library, and have no reason to return on low-intent days. Format fragmentation: music listeners who have never touched a podcast or audiobook on Spotify, and vice versa, even though audiobooks grew listening 60% from 2024 to 2025, Audiobooks+ is on track for $100 million ARR by mid-2026, and nearly 50% of audiobook listeners are under 35. If those audiences never cross formats, Spotify is running three parallel products in one app.
Pick one failure mode and commit to it with a rationale. Strong choice for this question: lean-back passivity, measured by days listened per month among free-tier users, because Spotify’s own data shows the lever already exists and the global base is largely untouched.
Structure a strong answer
strong
"Before I propose anything: when you say engagement, are we optimizing for depth, breadth, or stickiness? I'll commit to stickiness: days listened per month among free-tier users. Spotify's Q1 2026 data confirms this is moving in the US after the personalized free experience rollout, and there's a much larger global base that hasn't seen that intervention yet."
User segment: the lean-back listener. Someone with 3-4 years of Spotify history, a rich taste profile, who opens the app 8-10 times per month but only on days when they already know what they want. They aren't churning. They just don't have a reason to open Spotify on a Tuesday morning when no specific song is top of mind. Their problem: the default app state (Home) is a grid of recommendations that requires an active choice. Low-intent sessions are punished by a UI designed for high-intent ones.
Root cause: Spotify's Home surface optimizes for discovering new content, not for entering a listening state. On days when a user has no specific intent, the Home grid creates choice paralysis rather than removing it.
The feature: a Mood Door. A single-tap entry state, surfaced as the primary card on Home during low-intent hours (commute windows, late evening), that asks one word of context. Four or five options (focused, winding down, moving, wandering) that immediately launch a context-aware DJ-style session: no playlist to choose, no grid to scroll. This extends Spotify's existing DJ feature to the specific moment lean-back listeners drop off. DJ already drives roughly 20% more interactions after personalization improvements, and Jam has 50 million monthly users, which proves that interactive engagement at scale is real on this platform. Mood Door is not a new technical capability: it's a new entry surface that routes the existing taste graph into the session type that fits the moment.
This is directly in line with Spotify's stated 2026 strategic shift from passive to interactive experiences (named explicitly at the 2026 Investor Day by both co-CEOs). It doesn't require new content, new licensing, or new infrastructure. The only build is the entry surface and the routing logic.
Metrics: days listened per month (primary), session start rate on historically low-intent days (secondary). Guardrails: skip rate within the first 90 seconds and session completion rate, to avoid inflating raw opens without listening value. I'd A/B test on free-tier users in one non-US market and measure the 30-day listening day delta before any expansion decision.
Trade-off to name: Mood Door competes for prime Home real estate with the recommendation grid. For power users who already know what they want, the card is noise. Solve this with a context signal: show it only on days with no prior session, not on days the user has already opened the app and selected something. That targeting also makes the A/B signal cleaner.
weak
"I'd improve Spotify's recommendations engine using machine learning to better personalize playlists." This fails for three reasons: (1) the Large Taste Model already processes 3.4 trillion daily signals, so proposing it as an improvement signals the candidate did no homework; (2) it names no user segment, no engagement metric, and no specific behavioral problem; (3) it has no connection to any business outcome. A second weak pattern: listing features without prioritization logic ("better social sharing, offline mode for free users, more podcast discovery") when Spotify has either shipped these or explicitly deprioritized them for business reasons. A third: walking through CIRCLES or GAME step by step without ever landing on a specific feature or metric. Interviewers don't need to hear the framework name. They need to see you apply judgment.
The “Time Well Spent” reframe interviewers are looking for
The Spotify co-CEOs named “Time Well Spent” at the 2026 Investor Day as the company’s explicit engagement philosophy: durable listening that users choose to maintain, not time-on-platform maximization. That distinction matters for your answer. Features that inflate raw session time (auto-play loops, aggressive notifications) are off-strategy. Features that make Spotify the place where audio identity lives are on-strategy.
The evidence runs through every major 2026 product moment. Wrapped 2025 generated 620 million shares because it made users feel represented and gave them something worth sharing. SongDNA launched in March 2026 and drove 265 million interactions in its first weeks because it connected taste to self-expression. The 20th anniversary experience generated 100 million engagements in 6 days and drove the single biggest subscriber intake day in company history. The Taste Profile editor (in beta after its SXSW March 2026 announcement) lets users give natural-language feedback on their taste, which is the identity-salience mechanic made explicit.
Viable engagement features are ones users will pay to keep (Audiobooks+, a future Deluxe tier) or share because they reinforce identity. Lovable means meeting users in the moment they actually have, not the average session. Feasibility is largely not the constraint: Spotify has the taste graph, the catalog, 700,000 audiobook titles, and the creator infrastructure. The real question is which surface produces the next identity moment at scale, and whether the mechanic is a format (cross-format adoption) or an entry state (the lean-back problem).
Candidates who clear the bar at Spotify in 2026 know that “better recommendations” is table stakes, not a direction. They know the difference between engagement depth, breadth, and stickiness before proposing any feature. And they can name a trade-off before the interviewer asks.
See how Spotify compares to other consumer product interviews for the full interview structure. The north star metric framework covers how to choose and defend a single metric before you propose features. For the broader 2026 lens on viable versus lovable, feasibility is free applies directly to every Spotify product-sense question.
Framework
Asked at
Related
- How would you improve LinkedIn engagement? product-sense
- "Engagement is up but revenue is down: what do you do?" analytical
- "How would you improve YouTube's recommendations?" product-sense