Two Hierarchies for Consumer and Marketplace Startups: Focus First, Tip Second, Dominate Last

Sarah Tavel built her hierarchy of engagement out of frustration. After leading discovery at Pinterest, she began meeting consumer founders who all showed up with up-and-to-the-right graphs of sign-ups, downloads or monthly active users. In her view those numbers never got at whether a company was on the path to an enduring consumer product, and the point became, in her words, a pebble in her shoe.

The foundation of the hierarchy is what she calls the core action: the one behavior that shows a user understands the product's utility and is likely to return. For Facebook in the early days it was friending, for Pinterest it was pinning, and for Evernote it was writing a note. Users who never perform that action are not really users, which is why an MAU count tells you so little.

Tavel stresses that the choice matters because it decides what the whole team optimizes. In Pinterest's early days the team tracked follows, click-throughs, likes, pins and time on site, and experiments pushed different numbers in different directions. Picking one action gives the product roadmap and the new-user experience a single target.

Level two of the hierarchy is retention. Tavel's test at the individual-user level is that the product should get better the more you use it, and you should have more to lose by leaving. At Pinterest, her team shipped a picked-for-you feed that used every pin to personalize the home feed, possibly one of the first algorithmic feeds in a social product, while the boards themselves became a repository of a person's recipes, books and plans.

She is a lifetime Evernote user and uses it as her second example. Because she has thousands of documents there and knows search will find what she needs, she says she can never leave. Lenny Rachitsky points out the flip side: a great exporter would remove that friction, which makes switching costs an interesting growth strategy in themselves.

The top of the hierarchy is making the product self-perpetuating. Tavel imagines every click or tap as kinetic energy a user puts into the product, and a great product converts that energy back into a better experience. The strongest mechanism is a network effect, such as every pin adding an edge to Pinterest's graph, and around it sit growth and re-engagement loops like sharing, SEO metadata, collaboration and notifications triggered by other people's actions.

Evernote shows what happens without those loops: it had to pay to acquire users and tapped out, and its attempt at collaborative journals did not work. Houseparty and Clubhouse fell short for a subtler reason. Their realtime nature depended on push notifications, and as people followed more friends the notifications became overwhelming and users started ignoring them, so the flywheel broke down.

Tavel argues that this level matters because a consumer product aiming at hundreds of millions of users cannot afford to buy them. TikTok is the one example she knows of that spent more than a billion dollars on user acquisition, and only after it had worked out the first two levels. Rachitsky adds that a fast-spreading app can also decay fast when a few people stop showing up.

Tavel has seen anonymity trip up many talented founders, especially at level two. Pseudonymous identities on Twitter or Reddit build accruing benefits and mounting losses, because followers and reputation attach to the persona. Pure anonymity lacks both, invites bad behavior and means a user can leave and return months later to an identical experience.

To improve retention, her first piece of advice is to actually measure it with weekly cohorts, tracking both who comes back and who completes the core action. She wants to see the classic smile graph, where cohorts become more engaged over time, instead of a leaky bucket that never plateaus. Her second is focus, and she gives the example of a friend-making app growing through TikTok with no geographic concentration, which makes high retention very hard.

The mistake she sees most is that founders compare themselves with the full expression of Roblox, Instagram or TikTok. The result is a first-time experience with too much to take in, which dilutes attention away from the core action.

Tavel describes how Pinterest settled on weekly active pinners as its North Star in 2011, when it was unclear whether it was just another follow graph like Twitter or Instagram. The bottoms-up analysis ranked every action by how many users did it and how likely they were to return the next week. Pinning, repinning or clicking through produced a return rate above 90%.

The top-down half asked what Pinterest is for: if a user never saves something to a board, do they understand the product? The team concluded that an experiment which did not raise the share of people pinning, or the pins per cohort, was not a success. Rachitsky points to a guide he has written on the regression approach to finding an activation milestone.

Tavel also shares a correction. She had assumed YouTube's core action was watching a video, until YouTube's then chief product officer, Shishir Mehrotra, told her it was subscribing. Subscribing locks creators into the platform as their audience grows, and it gives viewers a reason to come back, so it serves both sides, which Tavel says is a sign you have found the right action. She adds that all social products are really marketplaces.

Her second framework, the hierarchy of marketplaces, grew from the same reaction. Founders fixated on a million dollars of annualized GMV as the signal that they were ready for a Series A, but Tavel says not all GMV is equal. A food delivery company could reach that figure in one city or by scattering small amounts across many, and the scattered version makes the job much harder.

She argues that racing toward GMV pushes founders toward large markets where they can skim the cream, while reaching the same number in a constrained market takes longer but is the path to something that endures. A marketplace is effectively two companies, one for each side, so the work is already very hard.

The key insight is to work backwards from the goal. Dominant marketplaces like Airbnb, Amazon, eBay and Google are very profitable, and a chart Tavel saw as a board observer at OLX showed profitability rising with the lead over the number-two player. To dominate you must tip the market, and to maximize the odds of tipping you must start with a very constrained opportunity: focus, tip, dominate.

Tavel likens founder ambition to the heat of the sun, which can warm an entire ocean a little or be focused like a laser on a thimble until it boils. Capital and founder attention are both scarce, so spreading across geographies and categories dilutes both. She contrasts Postmates, which chased big cities and many verticals, with DoorDash, which went after the suburbs, where uneconomic delivery times meant almost no competition.

Because customers do not care how big you are, she proposes tracking happy GMV, the transactions that make a buyer or seller want to return. Rachitsky relates it to minimum viable happiness: do the unscalable work until a meaningful percentage of people retain after a transaction. The right constraint is often a city, a niche, or both, as Etsy started with craft-fair goods.

Thumbtack is a counterexample: it launched nationally across many categories, and Tavel notes that low repeat use is a lasting difficulty because people searching for a plumber start at Google. Marketplaces need enough touch points that buyers think of them first. She says you know level one is working when people retain and tell you they are happy, and she prefers Sean Ellis's question about how disappointed users would be, with 40% saying very disappointed as the bar, over NPS.

Level two is about replacing things that do not scale with things that do, in service of a tipping point. Tavel's story is REKKI, a marketplace connecting restaurants and their suppliers. At first REKKI's own sales team had to corner chefs in London kitchens, but each new chef sent suppliers a tidy order form instead of voicemails, and eventually suppliers sent REKKI lists of their customers to onboard. The hard sales work turned into leads on a silver platter.

She says great founders keep their ear to the ground so they can lean into the signals when they appear. She describes two kinds of tipping loops that work together. Growth loops include Hipcamp inviting friends to a booking, Uber's driver referral bonuses, Etsy's business cards that linked to sellers' stores, and DoorDash restaurants telling their customers to order through the app. Rachitsky adds Faire, where retailers and brands invite their existing partners.

The second type is happiness loops, which Tavel calls the kidneys of a marketplace. Unlike a social product, a marketplace wants healthy churn on the supply side, because some sellers will never give buyers a good experience. Search ranking and reputation systems reward the sellers who make buyers happy. Early Uber Eats ranked restaurants partly on how fast they prepared food, because speed was its presumed advantage.

Tavel offers a reminder and a caution. The reminder is that you must reach a saturation point to tip a market, which is why a food delivery launch in Des Moines is easier than one in New York or LA, and why Postmates in San Francisco and DoorDash in the suburbs both chose small opportunities.

The caution is that some markets are not susceptible to tipping. One condition is supplier concentration: Bill Gurley says great marketplaces create new incumbents, but incumbents are fat and happy and will not lean in. Rachitsky and Tavel debate supply versus demand and agree both matter: fragmented supply opens the door, but lasting value comes from cornering the demand side.

B2B marketplaces face a further problem, which Tavel says is homogeneous supply. Airbnb benefits from endless variety in its listings, so each addition improves the experience. For Mechanical Turk, a hundred thousand workers or five million may feel the same to a customer, which invited clones. Competition also matters, since tipping is far harder against entrenched rivals like Grubhub in New York, and Tavel notes that you can see tipping in better cohorts and organic growth.

Levels one and two produce a playbook, and Tavel says founders at level three are chomping at the bit to grow. The playbook may not repeat everywhere: REKKI's approach depended on supplier fragmentation in London, which Berlin lacked. Still, there is usually something that rhymes.

She describes three growth vectors: deepening the existing market, expanding the use cases within it as Uber went from black cars to UberX to UberPool, and then running as many markets as you can handle. Contribution profit from won markets funds new ones. Do not plant a thousand flags, she warns, because a competitor putting the same capital into one city will win it and gain capital to keep going.

On whether DoorDash dominates, Tavel is unsure, but says it has clearly built an incredible company in a still crowded market. Rachitsky adds that you can still build a great business without dominance, you just earn less cash flow.

Tavel says Benchmark gets excited by opportunities that look small from the outside. Airbnb and Etsy were underestimated for years, and Hipcamp began as a market for hardcore campers. Such markets have little competition early on, and she tells founders that if others say the market is too small, Benchmark would like to meet them. The market must not be a cul-de-sac, though, because it needs adjacent markets to expand into.

Hipcamp shows how that works: hosts used their earnings to build fire pits, then tree houses and yurts, which opened the market to glampers. Etsy is the warning. As it approached its IPO it chased GMV by letting in mass-produced goods, eroding trust among handmade sellers and buyers, and Tavel thinks that precipitated a CEO change.

When founders pitch her, she looks for an earned secret, meaning an overlooked opportunity that has become possible now, together with real focus and intellectual honesty about what works. This ties to the current idea she opened the episode with: markets are not bodies of water, and the question is what momentum would pull a plank of wood forward. She thinks stablecoins are one example today, while a trough in some DeFi areas shows the opposite.

Tavel's coda is that you can never rest on your laurels, because the history of marketplaces is one of disruption and consumer expectations keep rising. Grubhub, she says, was too slow to change the atomic unit of its supply and lacked private-market capital, while DoorDash and Uber Eats took over. Rachitsky adds that even strong network effects do not protect a complacent leader.

Asked about companies that are not marketplaces, she says open-source products must solve one developer's need extremely well before expanding, and that new companies built on large language models should also win by focusing on a constrained market first. She calls Parker Conrad's compound startup approach at Rippling a rare exception, with ambition that is both a sun and a laser. Rachitsky sums it up as getting happy developers or enterprise users who keep coming back.

For finding whitespace, she lists three paths from her post: take the low-NPS segment of a horizontal marketplace, as GOAT did with sneakers; find a niche no one watches; or change the atomic unit of supply. She sees LLMs as a real opportunity for the third path, by automating outreach and onboarding to long-tail supply. She has not seen a company yet but is on the hunt.