Sarah Tavel
General Partner at Benchmark, former first product manager at Pinterest and partner at Greylock
Where to hear and read about Sarah Tavel1 episode or test
Sarah Tavel is a General Partner at Benchmark who invests in consumer and marketplace startups, and was Pinterest's first product manager and later a partner at Greylock. The lesson most worth taking from her is to measure real engagement and happiness instead of MAUs or GMV, and to win one small, constrained market before expanding.
What the best, ambitious founders do is they focus that ambition like a laser beam on a small market, the thimble.
The hierarchy of engagement | Sarah Tavel (Benchmark, Greylock, Pinterest)
Sarah Tavel: the story and the lessons
1Sarah Tavel's path from Pinterest to Benchmark
Tavel led discovery at Pinterest as its first product manager. She worked on the home feed, search and recommendations, and the job was to raise engagement. Engagement meant helping people find something they loved enough to pin to a board.
She later became a partner at Greylock and is now a Benchmark partner investing in consumer and marketplace startups. Benchmark is early stage and aims to be the first board member, usually at Series A.
Founders' growth charts bothered her, so she wrote the Hierarchy of Engagement. Consumer founders in the growth hacking era showed up with up-and-to-the-right graphs of sign-ups, downloads and MAUs. Her Pinterest work made her distrust those charts.
The gap bothered her like a pebble in my shoe, so she wrote it up and distilled it into the hierarchy.
She watched profit climb with market dominance as a board observer at OLX. As a board observer under Fabrice Grinda, she saw the OLX graph showing profit rising with the lead over the number two player.
That picture shaped her view that marketplaces should be built backwards from dominance, and that Airbnb, Amazon, eBay and Google became very profitable once they held dominant share.
2The Hierarchy of Engagement: core action, retention, self-perpetuation
Tavel says every product has one core action that makes someone a real user. Early Facebook had friending, Pinterest had pinning, Evernote had writing a note, Snapchat had sending a snap. Measure users completing it as a cohort by week or day, and make the whole product lead to it.
Early Pinterest tracked follows, click-throughs, likes, pins and time on site. Some went up and some went down, so the team did not know what to optimize.
She found Pinterest's core action with a bottoms-up and a top-down analysis. Bottoms-up, the team listed every action and checked what share of users did it and how likely they were to return the next week. Users who pinned, repinned or clicked through came back with more than 90% probability, so weekly active pinners became the North Star.
Top-down, a user who never saves anything to a board may not understand what Pinterest is. After that, an experiment that did not raise the share of pinners or pins per cohort counted as a failure.
She treats the core action as a bet that can be revised. Tavel first assumed YouTube's core action was watching. Shishir Mehrotra, then YouTube's CPO, told her later analysis changed it to subscribing, because loss builds up as subscribers accrue.
She says the best core action serves both sides of a network at once, since a viewer subscribes only after finding creators who resonate. Teams should expect to try something for six months and then adjust.
Retention means the product improves with use and costs more to leave. Pinterest's picked-for-you feed turned every pin into recommendations in the home feed, and may have been the first algorithmic feed in a social product. Boards full of books, recipes and holiday plans gave users more to lose by leaving.
People did not follow each other much on Pinterest, so personalization was the way to make it improve with use. Evernote locks users in the same way with thousands of saved, searchable documents. "The product should get better the more you use it, and you'll have more to lose by leaving it."
Self-perpetuation turns every user action into a better product and more users. Every pin adds an edge to Pinterest's graph that improves recommendations for everyone. Early Pinterest also sent a push such as 'Sarah just pinned your pin to her art board' that could pull a dormant user back with no work from the company.
Evernote retained users but had no loops, so it had to pay to acquire them and tapped out. TikTok spent over $1B on paid acquisition only after the first two levels worked.
She warns against loops that depend on push notifications. Houseparty and Clubhouse relied on pushes to signal when to join. The more people users followed, the more pushes they got, until they ignored them and the flywheel broke.
Realtime products need pushes to catch the moment, which is what makes the trap hard to escape.
3Tavel's rules for measuring retention and onboarding
She tells founders to track weekly cohorts, not MAUs. Group users by sign-up week and track both how many return and how many complete the core action. The goal is a smile curve where cohorts become more retained over time, not a leaky bucket.
Many founders had never tracked cohorts before she suggested it. "Until you reach a point with your cohorts where there is a plateau, you have more work to do on figuring out the retention of your users."
She calls copying the finished product the biggest founder mistake. Founders compare themselves to Roblox, Instagram or TikTok and try to launch that on day one. A new user with no context sees too much and never finds the one action to take.
Tavel says it comes from not getting the activation moment right. Pinterest once tested an onboarding that never explained pinning, and learned that users must understand they are there to discover something and save it to a board.
Tavel argues pure anonymity kills retention. With pure anonymity nothing accrues to a user and nothing is lost by leaving, so you can delete the app and return months later to an unchanged experience. It also invites bad behavior, which pulls down the community.
Pseudo-anonymous accounts on Twitter and Reddit keep a persistent identity that gains followers. Founders keep trying pure anonymity because early growth is fun, but products like Secret cannot persist.
She constrains geography so social products can become retentive. A friend-making app grew its user numbers through TikTok, which has no geographic limit. Without a focused geography, she says, it is hard to make that kind of product highly retentive.
Growing user numbers can hide the lack of retention, and a social product needs enough nearby users for the experience to work.
4The Hierarchy of Marketplaces: focus, tip, dominate
Tavel says GMV misleads founders the same way MAUs do. Chasing a $1M GMV milestone pushes founders toward a big market where it is easier to skim the cream. Many founders believed $1M of annualized GMV meant they were ready for a Series A.
Reaching $1M in a constrained market is harder and slower, but it is the path to something that endures.
She tells founders to aim their ambition at a thimble. Capital and founder attention are the two scarce resources. DoorDash began in the suburbs with little competition and lost a lot of money on long drives, while Postmates spread across big cities and categories at once.
Tavel says the only way to get these companies off the ground is myopic focus on a segment. Constrain by location, niche or both, since constraining early gives you the option to grow later.
She tracks happy GMV and a threshold of minimum viable happiness. Founders define the happy path, measure GMV along it, and do unscalable work until enough buyers and sellers retain after a transaction. Only then do they look for scaling levers.
"Customers don't care how big you are, they don't care how many transactions you've accumulated." Companies with huge scale have still been disrupted by startups.
She uses the Sean Ellis test, not NPS, to judge readiness. Ask how disappointed users would be if the product disappeared. At least 40% answering very disappointed means the focused core is working and it is time to expand.
Tavel dislikes NPS for this and has a blog post on why. Unprompted praise and returning users are the other signs the center is white hot.
Tipping needs growth loops and happiness loops working together. Growth loops add participants: Hipcamp lets a booker invite friends at checkout, Etsy gave sellers business cards linking to their store, and Uber paid driver referral bonuses. Happiness loops use search ranking and reputation to reward good suppliers and churn out bad ones.
Early Uber Eats ranked restaurants by prep speed, so a great restaurant taking 40 minutes ranked low. She calls these loops the kidneys of the marketplace.
Rekki showed her what tipping feels like. Rekki first sent a sales force door to door in London to corner chefs in kitchens. Suppliers then preferred the clean order forms to voicemails and sent Rekki CSVs of their customers to onboard.
Founder Ronan did not predict this. Close relationships with both sides let him notice it and build momentum. "All of a sudden something goes from being really freaking hard to so much easier."
She checks whether a market can tip before backing it. Tipping needs saturation, which comes far faster in Des Moines than in New York. Concentrated supply, entrenched rivals like Grubhub in New York, and homogeneous supply like Mechanical Turk block it.
Incumbent suppliers are fat and happy and won't lean in, so a flywheel needs hungry suppliers. Her post lists six conditions.
After tipping, Tavel says put more wood behind fewer arrows. Grow in three steps: strengthen the tipped market, add use cases there, then add more markets. Uber went from black cars to UberX to UberPool in cities it already dominated.
Spread $1M over 10 cities and a rival who puts $1M into one city wins it, then raises more. "You always want to put more wood behind fewer arrows."
5What Tavel looks for when investing and what she warns about
Tavel looks for a current in the market, not a big market. She thinks of a market as a current, not a body of water, so even a plank of wood gets pulled forward. A market with no momentum forces you to build something big and fancy to make progress.
Benchmark asks about the dynamics of change, the why now question. She sees stablecoins as having a strong current because of inflation and many local currencies, and DeFi as having lost its own.
She leans toward markets that look small from the outside. Airbnb and Etsy were underestimated for years, and small markets have no competition early. Her pitch to founders is that if others say the market is too small, she wants to meet them.
The market must not be a cul-de-sac. Hipcamp began with hardcore campers, then hosts built bathrooms, tree houses and yurts, which opened it to glampers like Tavel.
Three signals make a marketplace pitch exciting to her. An earned secret about an overlooked opportunity with something changing now, focus like a laser beam, and intellectual honesty about what works instead of vanity metrics.
She says honest metrics are how you build long-term, enduring value, and that about 95% of marketplace hurdles are ordinary startup problems.
She saw Etsy lose trust by chasing GMV before its IPO. Etsy let mass-produced goods onto a handmade marketplace, and handmade sellers and buyers reacted as trust eroded. Tavel saw one seller spamming Pinterest with pins of clearly mass-produced scarves.
She thinks it may have precipitated the CEO change, though she isn't sure.
She says marketplaces get disrupted if they stop improving. Grubhub was a public company that couldn't access capital like private markets and was too slow to change its supply to delivery it fulfilled itself. DoorDash and Uber Eats overtook it.
Consumer expectations constantly rise, as Bezos and Amazon argue. "You can never rest on your laurels when you're building a marketplace."
She hunts for new marketplaces by changing the atomic unit of supply. Three routes: find the low-NPS pocket of a horizontal player (sneakers on eBay led to GOAT), find a niche no one watches, or change the unit of supply. DoorDash, Uber Eats and Postmates took any restaurant and supplied the delivery.
She thinks LLMs may help most with the third route by automating outreach and onboarding of long-tail supply, but hasn't seen an example yet and is on the hunt.
The hierarchy of engagement | Sarah Tavel (Benchmark, Greylock, Pinterest)
Two frameworks for building a consumer product and a marketplace: the Hierarchy of Engagement (core action, retention, self-perpetuation) and the Hierarchy of Marketplaces (focus, tip, dominate). The lesson is to measure real engagement and happiness instead of MAUs or GMV, and to win a small constrained market before expanding.
Tavel led discovery at Pinterest as its first product manager. She worked on the home feed, search and recommendations, and the job was to raise engagement. Engagement meant helping people find something they loved enough to pin to a board.
She later became a partner at Greylock and is now a Benchmark partner investing in consumer and marketplace startups. Benchmark is early stage and aims to be the first board member, usually at Series A.
Founders' growth charts bothered her, so she wrote the Hierarchy of Engagement. Consumer founders in the growth hacking era showed up with up-and-to-the-right graphs of sign-ups, downloads and MAUs. Her Pinterest work made her distrust those charts.
The gap bothered her like a pebble in my shoe, so she wrote it up and distilled it into the hierarchy.
She watched profit climb with market dominance as a board observer at OLX. As a board observer under Fabrice Grinda, she saw the OLX graph showing profit rising with the lead over the number two player.
That picture shaped her view that marketplaces should be built backwards from dominance, and that Airbnb, Amazon, eBay and Google became very profitable once they held dominant share.
Tavel says every product has one core action that makes someone a real user. Early Facebook had friending, Pinterest had pinning, Evernote had writing a note, Snapchat had sending a snap. Measure users completing it as a cohort by week or day, and make the whole product lead to it.
Early Pinterest tracked follows, click-throughs, likes, pins and time on site. Some went up and some went down, so the team did not know what to optimize.
She found Pinterest's core action with a bottoms-up and a top-down analysis. Bottoms-up, the team listed every action and checked what share of users did it and how likely they were to return the next week. Users who pinned, repinned or clicked through came back with more than 90% probability, so weekly active pinners became the North Star.
Top-down, a user who never saves anything to a board may not understand what Pinterest is. After that, an experiment that did not raise the share of pinners or pins per cohort counted as a failure.
She treats the core action as a bet that can be revised. Tavel first assumed YouTube's core action was watching. Shishir Mehrotra, then YouTube's CPO, told her later analysis changed it to subscribing, because loss builds up as subscribers accrue.
She says the best core action serves both sides of a network at once, since a viewer subscribes only after finding creators who resonate. Teams should expect to try something for six months and then adjust.
Retention means the product improves with use and costs more to leave. Pinterest's picked-for-you feed turned every pin into recommendations in the home feed, and may have been the first algorithmic feed in a social product. Boards full of books, recipes and holiday plans gave users more to lose by leaving.
People did not follow each other much on Pinterest, so personalization was the way to make it improve with use. Evernote locks users in the same way with thousands of saved, searchable documents. "The product should get better the more you use it, and you'll have more to lose by leaving it."
Self-perpetuation turns every user action into a better product and more users. Every pin adds an edge to Pinterest's graph that improves recommendations for everyone. Early Pinterest also sent a push such as 'Sarah just pinned your pin to her art board' that could pull a dormant user back with no work from the company.
Evernote retained users but had no loops, so it had to pay to acquire them and tapped out. TikTok spent over $1B on paid acquisition only after the first two levels worked.
She warns against loops that depend on push notifications. Houseparty and Clubhouse relied on pushes to signal when to join. The more people users followed, the more pushes they got, until they ignored them and the flywheel broke.
Realtime products need pushes to catch the moment, which is what makes the trap hard to escape.
She tells founders to track weekly cohorts, not MAUs. Group users by sign-up week and track both how many return and how many complete the core action. The goal is a smile curve where cohorts become more retained over time, not a leaky bucket.
Many founders had never tracked cohorts before she suggested it. "Until you reach a point with your cohorts where there is a plateau, you have more work to do on figuring out the retention of your users."
She calls copying the finished product the biggest founder mistake. Founders compare themselves to Roblox, Instagram or TikTok and try to launch that on day one. A new user with no context sees too much and never finds the one action to take.
Tavel says it comes from not getting the activation moment right. Pinterest once tested an onboarding that never explained pinning, and learned that users must understand they are there to discover something and save it to a board.
Tavel argues pure anonymity kills retention. With pure anonymity nothing accrues to a user and nothing is lost by leaving, so you can delete the app and return months later to an unchanged experience. It also invites bad behavior, which pulls down the community.
Pseudo-anonymous accounts on Twitter and Reddit keep a persistent identity that gains followers. Founders keep trying pure anonymity because early growth is fun, but products like Secret cannot persist.
She constrains geography so social products can become retentive. A friend-making app grew its user numbers through TikTok, which has no geographic limit. Without a focused geography, she says, it is hard to make that kind of product highly retentive.
Growing user numbers can hide the lack of retention, and a social product needs enough nearby users for the experience to work.
Tavel says GMV misleads founders the same way MAUs do. Chasing a $1M GMV milestone pushes founders toward a big market where it is easier to skim the cream. Many founders believed $1M of annualized GMV meant they were ready for a Series A.
Reaching $1M in a constrained market is harder and slower, but it is the path to something that endures.
She tells founders to aim their ambition at a thimble. Capital and founder attention are the two scarce resources. DoorDash began in the suburbs with little competition and lost a lot of money on long drives, while Postmates spread across big cities and categories at once.
Tavel says the only way to get these companies off the ground is myopic focus on a segment. Constrain by location, niche or both, since constraining early gives you the option to grow later.
She tracks happy GMV and a threshold of minimum viable happiness. Founders define the happy path, measure GMV along it, and do unscalable work until enough buyers and sellers retain after a transaction. Only then do they look for scaling levers.
"Customers don't care how big you are, they don't care how many transactions you've accumulated." Companies with huge scale have still been disrupted by startups.
She uses the Sean Ellis test, not NPS, to judge readiness. Ask how disappointed users would be if the product disappeared. At least 40% answering very disappointed means the focused core is working and it is time to expand.
Tavel dislikes NPS for this and has a blog post on why. Unprompted praise and returning users are the other signs the center is white hot.
Tipping needs growth loops and happiness loops working together. Growth loops add participants: Hipcamp lets a booker invite friends at checkout, Etsy gave sellers business cards linking to their store, and Uber paid driver referral bonuses. Happiness loops use search ranking and reputation to reward good suppliers and churn out bad ones.
Early Uber Eats ranked restaurants by prep speed, so a great restaurant taking 40 minutes ranked low. She calls these loops the kidneys of the marketplace.
Rekki showed her what tipping feels like. Rekki first sent a sales force door to door in London to corner chefs in kitchens. Suppliers then preferred the clean order forms to voicemails and sent Rekki CSVs of their customers to onboard.
Founder Ronan did not predict this. Close relationships with both sides let him notice it and build momentum. "All of a sudden something goes from being really freaking hard to so much easier."
She checks whether a market can tip before backing it. Tipping needs saturation, which comes far faster in Des Moines than in New York. Concentrated supply, entrenched rivals like Grubhub in New York, and homogeneous supply like Mechanical Turk block it.
Incumbent suppliers are fat and happy and won't lean in, so a flywheel needs hungry suppliers. Her post lists six conditions.
After tipping, Tavel says put more wood behind fewer arrows. Grow in three steps: strengthen the tipped market, add use cases there, then add more markets. Uber went from black cars to UberX to UberPool in cities it already dominated.
Spread $1M over 10 cities and a rival who puts $1M into one city wins it, then raises more. "You always want to put more wood behind fewer arrows."
Tavel looks for a current in the market, not a big market. She thinks of a market as a current, not a body of water, so even a plank of wood gets pulled forward. A market with no momentum forces you to build something big and fancy to make progress.
Benchmark asks about the dynamics of change, the why now question. She sees stablecoins as having a strong current because of inflation and many local currencies, and DeFi as having lost its own.
She leans toward markets that look small from the outside. Airbnb and Etsy were underestimated for years, and small markets have no competition early. Her pitch to founders is that if others say the market is too small, she wants to meet them.
The market must not be a cul-de-sac. Hipcamp began with hardcore campers, then hosts built bathrooms, tree houses and yurts, which opened it to glampers like Tavel.
Three signals make a marketplace pitch exciting to her. An earned secret about an overlooked opportunity with something changing now, focus like a laser beam, and intellectual honesty about what works instead of vanity metrics.
She says honest metrics are how you build long-term, enduring value, and that about 95% of marketplace hurdles are ordinary startup problems.
She saw Etsy lose trust by chasing GMV before its IPO. Etsy let mass-produced goods onto a handmade marketplace, and handmade sellers and buyers reacted as trust eroded. Tavel saw one seller spamming Pinterest with pins of clearly mass-produced scarves.
She thinks it may have precipitated the CEO change, though she isn't sure.
She says marketplaces get disrupted if they stop improving. Grubhub was a public company that couldn't access capital like private markets and was too slow to change its supply to delivery it fulfilled itself. DoorDash and Uber Eats overtook it.
Consumer expectations constantly rise, as Bezos and Amazon argue. "You can never rest on your laurels when you're building a marketplace."
She hunts for new marketplaces by changing the atomic unit of supply. Three routes: find the low-NPS pocket of a horizontal player (sneakers on eBay led to GOAT), find a niche no one watches, or change the unit of supply. DoorDash, Uber Eats and Postmates took any restaurant and supplied the delivery.
She thinks LLMs may help most with the third route by automating outreach and onboarding of long-tail supply, but hasn't seen an example yet and is on the hunt.