Notes — Sarah Tavel on the Hierarchy of Engagement and Marketplaces
Four questions [Adler frame]
Q1 — What is this about? Two frameworks for building enduring consumer and marketplace products. The Hierarchy of Engagement sequences the work of building durable retention in consumer products; the Hierarchy of Marketplaces sequences the work of building durable supply-demand dynamics in marketplace businesses. Both are reactions to the same problem: metrics that look like progress but are not.
Q2 — How is it argued? Through a combination of framework articulation and case studies. Pinterest (where Tavel was first PM) provides the richest examples: the bottoms-up + top-down analysis that identified pinning as the core action; the co-evolution of creator and viewer lock-in at YouTube. The marketplace framework emerges from observing founders fixate on GMV milestones that don’t signal lasting value.
Q3 — Is it true? The engagement framework is well-grounded in cohort analytics practice. The core action analysis (regress every action against next-week retention) is a reproducible method. The mounting loss / accruing benefits framing is insightful but partial — it is a description of lock-in mechanics, not a guarantee that lock-in creates durable competitive advantage. L3 (self-perpetuating network effects) is reached by very few products; the framework may over-index on the aspirational end state.
Q4 — What of it? The frameworks provide a diagnostic for where to invest product effort. Most founders who are struggling with retention are failing at L1 (core action not well-defined or not being reliably reached on first visit), not L2 or L3. The market currents metaphor is a useful corrective to market size as the primary investment screen.
Glossary
Core action — The single product behaviour most predictive of long-term retention; the action where completing it both demonstrates user understanding and forecasts return.
Accruing benefits — Value that compounds with product usage: a larger network, more personalised recommendations, a richer personal library. The product is more valuable to a longer-tenured user than a new one.
Mounting loss — The increasing cost of leaving a product as usage accumulates. The inverse of accruing benefits — what you give up by churning grows over time.
Tipping point (marketplace) — The threshold beyond which supply attracts demand and demand attracts supply without active company intervention.
Thimble strategy — Achieving dominant liquidity in a small, focused pool before expanding. The deliberate choice to be the biggest player in a tiny niche as the path to scale.
Hierarchy of Engagement: the three levels
L1 — Core action + activation. Before anything else, identify the action that best predicts return, and make it the organising principle of the product. At Pinterest, the bottoms-up analysis ranked every action (like, follow, click-through, pin, repin) by two criteria: what percentage of users complete this action, and if they complete it, what is their probability of returning next week? Pinning and repinning ranked highest on both. The top-down confirmation: if a user never pins anything, do they really understand Pinterest? No.
L2 — Accruing benefits + mounting loss. Users who have pinned 1,000 items have a personalised, curated experience unavailable to new users. Leaving costs them that curation. The product must be engineered to accumulate and surface this investment — not just store it.
L3 — Self-perpetuating network effects. The product generates its own supply and demand. Pinterest creators who attract followers have a stake in the platform; their departure weakens it for others. At this level, the marketplace (or social product) is self-sustaining.
Core action identification method
Tavel’s two-step:
-
Bottoms-up: list every user action; regress each against next-week DAU for users who performed that action. Rank by: (a) completion rate and (b) retention predictive power.
-
Top-down: ask what it means for a user to have genuinely used the product. If they never did X, did they really show up?
The correct core action satisfies both: high retention prediction from data, and conceptual completeness from the product’s purpose.
The anonymity problem
Pure anonymity blocks L2. Without a persistent identity, there is no accruing benefit and no mounting loss. Users can delete and return to the same experience three months later. Pseudo-anonymity (Twitter handles, Reddit accounts) can work — those identities accumulate followers, history, reputation. But pure anonymity is a structural barrier to the engagement hierarchy.
Hierarchy of Marketplaces in depth
The GMV trap: founders target $1M annualised GMV as a Series A signal. Tavel’s observation: that GMV number says nothing about whether the supply and demand generating it are retained, complementary, or tipping towards self-reinforcement. The quality question — is this GMV the kind that leads to tipping? — matters more than the quantity.
The thimble strategy: be dominant in a tiny pool. A marketplace that is 80% of a $1M GMV pool is in a fundamentally stronger position than one that is 5% of a $20M pool. The former can tip; the latter probably cannot.
Market currents
Tavel’s preferred market assessment question: not ‘how big is this market?’ but ‘is there a current in this market?’ — a structural force that will pull participants forward regardless of what the company does. Market currents include regulatory change, technology transitions, demographic shifts, or collapsing incumbents. In a market with a strong current, a basic product will be carried; in a static market, an excellent product still has to fight for every inch.