The default playbook for marketplace launches has been canonized in a single book. Andrew Chen's The Cold Start Problem describes the formula in detail: identify the smallest "atomic network" that can grow on its own, drive enough density and breadth in that network to survive anti-network effects, then replicate the pattern in adjacent networks until the entire market tips[1]. Uber did it city by city. Airbnb did it city by city. Tinder did it campus by campus. The pattern is so widely understood that founders now reach for it as the starting assumption rather than as a strategic choice.
What is rarely said out loud is that the playbook has a hidden assumption embedded in it. Chen himself observes that being dominant in one city did not transfer to the next, because Uber's "network effects were localized primarily to each individual city."[2] The assumption is that local network density in one city builds a brand, an operational muscle, and a story strong enough to carry the next city. For some product categories that is true. For social products, it is not. The friend you make in Chicago does not stay in Chicago. The host you trust in San Juan moves to Miami for the winter. The connection you make on a Wednesday in CDMX matters when you fly to LA in March.
For HAOT, this is the design constraint that shapes the launch geometry.
The hidden assumption in city by city
The city-by-city playbook works cleanly for products whose unit of value is local and stays local. A rider in NYC does not benefit from a driver in San Diego. An Airbnb host listing in Lisbon does not gain from another host in Mexico City filling a date in March. Each city is a more or less self-contained network, and the company aggregates them.
Social products fail this test. The HAOT primitive is not a single matched ride or a single booked night. It is a friendship that forms when people show up in person to a shared plan. That friendship has a half-life of years, sometimes decades. It moves with the people involved. When my Chicago user spends the summer in Miami, she does not stop being a HAOT user; the question is whether the product can host her continuity. When my CDMX user lands at SFO for a conference, she does not stop being part of the bond graph; the question is whether activities in SFO are visible to her at all.
The single-city launch model answers no to both of those questions. It cannot, by construction. A product that exists in only one city is a product that punishes its earliest users every time they leave home.
Social marketplace density is redundant, not progressive. You need it in multiple places at once, not in sequence.
Why five, not three, and not twelve
The number is not arbitrary. Five is the smallest count we found that creates a geographically credible bond graph and stays operationally manageable. Three cities create gaps too wide for cross-pollination to feel coherent. Twelve require operational overhead we cannot underwrite at pre-seed. Five is a constellation.
What matters more than the count is the composition. The five seeded cities are not a top-five-by-population list. They are five distinct network roles, each calibrated to a different test.
Chicago. Temperate climate, dense urban core, mid-cost. A large young-professional cohort. Chicago is the canonical "could be any major US city" test. If the product works here, it generalizes to most of the secondary US metros that follow.
San Juan. My home city. The founder presence matters less for the network and more for the operational rhythm. San Juan is also the bilingual seed, a city that runs in both Spanish and English at the same time, which is the linguistic mode the rest of the product needs to operate in. Puerto Rico sits in mid-income territory by global classification, which is where the WHO Commission data reports the highest loneliness prevalence[3]. We want our seed users in the part of the income distribution the report identifies as most affected.
Miami. The Latin American to US connector. Young, dense, and bilingual in the same direction as San Juan but with a different cultural seam. Miami tests whether the same product runs cleanly across the diaspora connection from CDMX, Bogotá, Caracas, Buenos Aires, and back.
Mexico City. The largest Spanish-language metropolitan area in the world. CDMX is not a nice to have in the launch set. It is the city that determines whether HAOT works for the largest single linguistic segment of our natural addressable market. Building the product to function from week one in CDMX is what separates a global product from a US product with a Spanish localization tab.
Los Angeles and Orange County. Treated as one node for launch planning purposes, even though they are distinct geographies. Southern California concentrates two assets the launch needs: a culturally diverse creative-and-lifestyle population that maps cleanly to in-person activity products, and a large Latino demographic that ties this node back to the Spanish-language thesis seeded in San Juan, Miami, and CDMX. Media and investor density in LA also produces the writing about the category we belong to.
The constellation is not a wedge. It is not a beachhead-and-expand sequence. It is five distinct hypothesis tests running in parallel, each calibrated for a different network type. If three of the five work, we know the product has structural reach early enough to act on it. We learn faster in five than we would in one.
What this enables that single-city launches do not
Three things follow from this geometry that do not follow from the standard playbook.
The first is a translocal bond graph from day one. A user who lives in Chicago and visits Miami sees HAOT activities in Miami without switching markets, recreating a profile, or signaling she is new. The unit of value, the activity, is geographically anchored. The user is geographically mobile. The product reflects that asymmetry from the first session.
The second is cross-pollination of supply. A host in CDMX who runs a Sunday morning hike inspires a host in Miami to do the same. The platform's social proof is multilingual and multi-context from launch, not a single-language monoculture that has to be translated and re-platformed after the fact. The hard side of the network, the host economy, is seeded with examples from multiple cultural contexts at the same time, which produces a wider creative range in the activity catalog earlier.
The third is investor and category narrative coherence. When a VC asks for the GTM, "five cities, day one" answers a different question than "Chicago first." The first answer signals confidence in the geometry. The second answer signals caution about it. Sangeet Paul Choudary's work on platform economics, co-authored in Platform Revolution, has repeatedly argued that for two-sided networks, the early shape of the user base predicts the long-run category position[4]. We are choosing the shape deliberately.
The trade we are explicitly making
The honest counter is that none of the five cities gets the depth of density that a one-city launch would have produced. We are trading depth for breadth at the start. That trade is the bet.
The bet is that for social products specifically, breadth is the more defensible asset. Once minimum-viable density exists in five cities, growing each city deeper is mechanically simpler than convincing users in other cities that a product designed for someone else's geography is also for them. The reverse path, going deep in one city and then expanding outward, is the failure mode that every previous attempt in the friendship category has hit. We wrote about that pattern earlier this week, in the piece on why anti-dating is architecture. The Five Cities launch is the operational consequence of taking that earlier argument seriously.
Going deep in one city and then expanding outward is the failure mode that every previous attempt in friendship marketplaces has hit.
What follows from this
The geometry of activation is downstream of the thesis. If the thesis is that platonic connection has been failed by single-city, single-language, single-cohort products, then the only credible launch is one that contradicts each of those failure modes simultaneously.
Five cities, day one, is the smallest set that does that. It proves the geometry. It demonstrates the operational discipline required to run a multilingual, multi-context marketplace from week one. It gives the bond graph enough surface area to become defensible before any single city becomes the entire product. And it makes the strategy legible to the investors, partners, and journalists who will determine whether HAOT is read as a regional startup or as a category-defining one.
The default playbook has a place. For marketplaces whose value is purely local, it is correct. For social products that build networks of trust that travel with their users, it is the wrong starting geometry. We are starting differently because the product requires it.
Sources
- [1] Andrew Chen, The Cold Start Problem: How to Start and Scale Network Effects. Harper Business, 2021. Link
- [2] Andrew Chen, The Cold Start Problem, p. 323, on the localization of Uber's network effects across cities. 2021.
- [3] WHO Commission on Social Connection, From Loneliness to Social Connection: Charting a Path to Healthier Societies. June 30, 2025. Link
- [4] Geoffrey G. Parker, Marshall W. Van Alstyne, Sangeet Paul Choudary, Platform Revolution: How Networked Markets Are Transforming the Economy. W.W. Norton and Company, 2016. Link