analytical · hard
"Engagement is up but revenue is down: what do you do?"
Engagement is rising but revenue is falling. What do you do?
This question has replaced the funnel-drop question as the dominant analytical format at Meta and Stripe PM loops. The interviewer is watching whether you diagnose before you prescribe, whether you can name the four mechanisms that produce this pattern, and whether you land on a decision rather than a hedge.
Clarify before hypothesizing
Ask: Which engagement metric specifically? Sessions, DAU, feature interactions, and time in product each point to different causes. Which revenue metric? Ads, subscription conversion, and transactional GMV diverge from engagement for completely different reasons. Gradual or sudden? Has anything shipped recently? Across all users or one segment? This isn’t stalling. It cuts the hypothesis space in half before you start.
The four root causes
State all four, then rank by prior probability given what you just learned.
1. The engagement metric is measuring the wrong thing. Asana’s north star (tasks completed per week) is designed so engagement and value delivery can’t come apart. If your metric would rise even when the product is failing, it’s measuring activity, not value.
2. User mix shifted. A viral moment or broad top-of-funnel campaign brings in high-engagement, low-LTV users. Aggregate engagement rises; revenue per user falls; aggregate revenue falls. The product didn’t break. The acquisition strategy brought in the wrong segment.
3. A product change removed friction that monetization depended on. Airbnb: more browsing sessions don’t equal more bookings. Better filters help users decide faster, raising sessions while bookings stay flat. The product got better in a way that reduced monetization touchpoints in the session.
4. Engagement and purchase intent are genuinely decoupled for this cohort. Users like it. They won’t pay. This is a viability problem, not a product problem.
The 2026 AI case
Cause three has a specific and increasingly common form in 2026: the AI-driven engagement loop. When a recommendation model, copilot, or agentic workflow gets genuinely better, it does more of the job in fewer steps. Fewer steps means fewer monetizable touchpoints. Engagement rises because the product is more useful. Revenue falls because the monetization model was taxing friction, not value.
A PM who diagnoses this correctly is not solving a metrics problem. They are diagnosing whether the current pricing model taxes friction (replace it with value-based or usage-based pricing) or taxes value (healthy). The fix isn’t to make the AI worse. It’s to redesign value capture before the next AI improvement ships and the gap widens further.
Minimum data pull
Segment engagement and revenue by cohort (new vs. retained, free vs. paid). If the divergence is largest in recent new-user cohorts, it’s a mix shift. Check whether the engagement metric has historically correlated with conversion. If the correlation was always weak, the metric was always a vanity metric. Trace the specific feature driving engagement growth and see whether it touches the conversion path. Check if the revenue decline is per-user or only in aggregate: per-user decline points to a pricing problem, not an engagement problem.
weak
"Engagement is a leading indicator of revenue, so this is probably fine. We should keep investing in engagement and trust that revenue will follow." This fails because it assumes a causal relationship that may not exist, is non-falsifiable, and does not investigate the mechanism. In ad-supported or SaaS products, engagement and revenue should move together within weeks, so "just wait" is not a strategy. Interviewers at Meta and Stripe call this the hope-based answer. A close second: proposing to A/B test a new paywall before determining whether the divergence is real, a data artifact, or a segment shift.
strong
"Before I hypothesize: which engagement metric and which revenue metric? And is this gradual or a step-change? Has anything shipped recently? [After clarifying.] Given DAU rising and subscription conversion falling over six weeks with a feature launch in that window, I'd rank cause three highest: a product change improved the experience in a way that broke a conversion touchpoint. To test it: segment by cohort to rule out mix shift, then trace whether the feature driving engagement growth actually touches the conversion path. If it doesn't, that's a pre-conversion feature and it was never going to move subscriptions. If the engagement driver is an AI feature, I'd name the 'AI removes friction that monetization depended on' pattern explicitly and recommend a pricing review before the next AI improvement ships. My recommendation: if it's cause three, this is a packaging problem. Redesign the monetization touchpoint that was removed, don't roll back the feature."
Name the decision
Hedging with “it depends on which scenario we find” is partial credit at senior level. The interviewer’s hidden question: “Would you ship something that grows engagement but hurts revenue?” There is no universal right answer. The test is whether you know when each answer is correct.
Kill the feature if it creates users who won’t convert and crowds out paid capacity. Redesign monetization if the feature creates real value the pricing model can’t capture. Accept the tradeoff only with cohort data showing the segment converts later.
Counter-metric discipline: a PM who ships an engagement feature should have pre-committed to watching revenue, LTV, and subscription conversion, not discovered the conflict six weeks later. Engagement without willingness to pay is not product-market fit. It is usage without viability.
For the metric framework underlying this, see north-star metric. For related diagnostics, see notifications up, time on site flat and DAU dropped: find the root cause.