Take Rate Is a Ceiling, Not a Strategy
Founders model on-demand marketplaces top-down: transactions times take rate equals revenue, and the take rate is set by copying whatever the category leader reportedly charges. This inverts the analysis. Your take rate is not a lever you freely set; it is a ceiling that competition, provider earnings expectations and regulatory attention press down on from above. The useful question is not 'what percentage can we charge?' but 'what minimum percentage covers our costs at realistic volume, and is that number defensible when a competitor undercuts it?' Framed that way, the take-rate discussion becomes concrete instead of aspirational.
Work the number from the bottom instead. Take your full cost to serve one transaction, support allocation, payment processing, guarantees or insurance if you offer them, refunds and promotion amortization, and divide by the gross value of the transaction. That floor is your true minimum take rate. In ride-hailing and delivery, payment processing alone typically consumes a meaningful slice of gross booking value before you have paid a driver or a courier a single unit of currency, which is why thin-margin categories punish small players first. Exact figures vary by market and processor, so measure yours.
Then remember the take rate funds the platform, it does not enrich it. Everything the marketplace does, matching, routing, trust and safety, support, is paid from that percentage, and competitors can run at a lower rate only while someone subsidizes the gap. If you launch on a ride-hailing platform with mature dispatch and wallet infrastructure already built, more of your take rate is gross margin rather than technology tax. The same logic applies to delivery: infrastructure you do not have to rebuild is margin you get to keep. Build versus buy is a margin decision before it is an engineering one.
CAC Has to Fight Purchase Frequency
Customer acquisition cost is meaningless in isolation; the number that matters is CAC against purchase frequency. An acquisition spend that sounds alarming for a user who orders once a year is trivial for one who transacts weekly. On-demand's core promise is frequency, that is the entire reason the category attracts capital, so your real unit economics question is: does my market, my vertical and my execution produce repeat transactions fast enough to pay back acquisition before the user churns away for good? That framing alone kills most naive launch plans before they can do damage.
Frequency is largely a property of the use case, and it is worth being unromantic about it. Daily-need categories, food, mobility, groceries, can support aggressive acquisition spending because the payback window is short and the habit is natural. Occasion-driven categories, events, travel, local services, cannot, so their CAC must come from channels that compound, referrals, search, partnerships, rather than paid ads that reset to zero each month. Founders routinely apply daily-category acquisition playbooks to occasion-category businesses, and then wonder why payback never arrives on schedule. Match the acquisition playbook to the category's rhythm, not to someone else's.
The practical discipline: instrument payback period from the very first week, cohort by cohort, and treat rising repeat-purchase behavior as your most important early metric. If your first hundred users do not transact again within the natural rhythm of the category, more acquisition spending is just paying to fill a leaking bucket. Fix the repeat behavior first; scale the spend second, and only against evidence that the loop now holds without subsidies. This sequencing feels slow for exactly one quarter, and then it feels obvious to everyone watching the dashboards.
Supply Density Beats Advertising
Marketplace failures rarely look like failures; they look like slow pickups and long ETAs. Users do not leave because the app crashed, they leave because the nearest courier was eleven minutes away and the food arrived cold. Liquidity, the probability that a user's request is matched quickly with a willing provider, is the product. And liquidity is primarily a function of supply density: how close the nearest available driver, restaurant or sitter is to the next request that arrives. Everything else, including your marketing, operates downstream of that one probability.
This is why launch geography is a strategy decision rather than a logistical one. Spreading a thin supply layer across a whole city produces mediocre liquidity everywhere; concentrating it in a few dense neighborhoods produces excellent liquidity where you can defend it. Users in covered zones get an experience worth repeating, providers in covered zones get enough orders to stay online, and both sides have a reason to return tomorrow. Density is also cheaper to defend: a competitor can outspend you city-wide and still lose the neighborhoods where you are meaningfully better.
Operationally, this reshapes your first months: recruit supply within tight radius boundaries, gate demand marketing to those zones, and expand only when liquidity metrics, match rate, ETA, utilization, hold in the current footprint without heroics. Every on-demand platform we ship, mobility, delivery, home services, carries the same warning in its onboarding: the map is not the market. Launch the smallest geography where the loop closes, and let the evidence draw the expansion map for you. Expansion earned this way costs less and holds better than expansion bought with ad spend.
Launch City Selection Is a Strategy Decision
Given everything above, choose the launch city the way an investor picks a position. Look for demand density you can reach without national advertising, supply you can recruit in weeks rather than quarters, a competitive landscape with real demand but weak local execution, and a regulatory environment you can actually staff for. Be equally honest about your own network effects: a city where you have relationships, language coverage and operational presence will beat a bigger city where you have none of those things. Your unfair advantages should count for more than raw market size in this decision.
Hedge accordingly: conditions vary enormously by country, city and category, and there is no universal recipe for picking a winner. What does generalize is the sequence, density first, frequency second, take rate defended by superior infrastructure and service, plus the discipline of measuring match rate and payback from week one. Teams that respect the sequence tend to find the unit economics eventually cooperate; teams that skip it tend to spend heavily to buy what they were supposed to earn. The sequence is unfashionable precisely because it keeps working, quarter after quarter.
One last note on the technology layer, because careful unit math is wasted on a platform that leaks. Payment failures, tracking gaps and dispatch latency all depress frequency in ways that look exactly like marketing problems from the outside. Our on-demand solutions are built so the infrastructure is not the bottleneck: proven dispatch, wallet and order flows that let you spend your energy on density and retention, which is where marketplaces are actually won. Every point of engineering attention that stays on the market rather than the plumbing compounds in your favor.
Reading the Four Numbers Together
The four metrics are not independent, and the interactions are where strategy lives. High take rates suppress frequency, because users feel the price every time they transact; thin take rates invite subsidized competition you cannot outlast; weak density inflates ETAs, which suppresses frequency, which stretches payback, which makes every acquisition dollar worse. When founders ask which number to fix first, the honest answer is the earliest one in the causal chain: density, then frequency, then take rate, in that order. Fixing downstream numbers while upstream ones fail is how launch budgets quietly disappear.
This is also why marketplace dashboards disappoint people in the first quarter. Revenue looks small because density is still forming; CAC looks high because frequency has not yet matured; the take rate looks fine because nothing is being tested against it yet. Judge the machine by its leading indicators, match rate, ETA, repeat behavior, and not by the trailing ones, and give the leading indicators enough time in one geography to actually mean something before you act on them. A quarter of patience in one market beats a year of noise across three.
Finally, resist the temptation to buy your way out of a structural problem. Subsidies can mask weak density for a while, and paid acquisition can mask weak frequency, but both only postpone the reckoning and usually deepen it. The marketplaces that endure are the ones whose operators understood the causal chain, invested in the earliest links, and used technology that did not tax every transaction along the way. That is the whole discipline, and it fits on one page. Most failed marketplaces ignored a page they could have read in five minutes.