If you run a marketplace, your retention rate is a weighted average of two different populations that leave for different reasons. Hosts do not churn like guests. Drivers do not churn like riders. Sellers do not churn like buyers. One number cannot tell you which side is leaking, and until you split it, every fix you ship is a guess.
I read the pages currently ranking for this. Both of the strongest are published by companies selling software into the problem, and neither separates the two sides. One says the sides have "different priorities" and then proposes unified data collection for both. The other covers buyers only and names no marketplace at all. Neither explains how to measure the two sides apart, which is the part that decides what you do next.
Section 01Supply and demand churn for different reasons
The two sides are not two segments of one audience. They are two products sharing a database. A seller's job is to earn; a buyer's job is to find. Almost nothing that keeps one keeps the other.
| Supply side (hosts, drivers, sellers) | Demand side (guests, riders, buyers) | |
|---|---|---|
| Why they leave | Not enough earnings to justify the effort, or the effort rose | Could not find what they wanted, or the experience was unreliable |
| What churn looks like | Silent. The listing stays up, the seller stops restocking | Obvious. They stop opening the app |
| The leading signal | Time to first sale, then earnings per active period | Search-to-booking rate, then repeat interval |
| What fixes it | Demand routed to them, faster payouts, less admin | Better matching, reliability, fewer dead ends |
| How fast it bites | Slowly, then all at once when the best supply leaves | Immediately, and it is recoverable |
| Cost to replace | High. You recruited, verified and onboarded them | Lower, and often just reactivation |
The row that matters most is the last one. Supply is expensive to acquire and slow to replace, and it degrades the product for the other side on its way out. Fewer listings makes search worse, worse search loses buyers, and fewer buyers pushes the next tier of supply out. That is the growth loop this site has written about running backwards, which is why two-sided growth loops are worth understanding before you touch retention at all. More of the retention writing here works through the same loop from the other direction.
Section 02How to measure each side separately
This is the part the ranking pages skip, and it is not complicated. You are running the same calculation twice against two cohorts, with two different definitions of active.
- Define active separately, and honestly. A seller is active if they could transact: listing live, stock available, responding. A buyer is active if they transacted. Using "logged in" for either will flatter you.
- Run two cohorts, never one. Compute your retention rate once for supply and once for demand, on the period that matches each side's natural cadence. They are rarely the same period.
- Match the period to the behaviour. A food marketplace's demand cadence is weekly and its supply cadence is daily. A property marketplace's demand cadence may be years. Monthly cohorts on both sides will make one of them look broken when it is not.
- Weight supply by contribution. Losing the top decile of sellers is not the same event as losing the bottom decile, but a headcount retention rate scores them identically. Track retained supply by volume as well as by count.
- Only then compare the two curves. Put them on one chart. Whichever falls first is your actual problem, and the other one is probably a symptom.
Section 03What actually holds each side
Once you know which curve is falling, the plays stop being interchangeable.
- Supply: get them to a first transaction fast. A seller who has earned once behaves differently from one who has only listed. Time to first sale is the single number worth engineering against, and routing early demand to new supply is usually cheaper than recruiting more of it.
- Supply: remove effort, not just add earnings. Payout speed, listing admin and dispute handling are retention features. They read as operations, so they lose budget arguments to marketing, and then the best sellers leave quietly.
- Demand: fix dead ends before you fix messaging. A buyer who searched and found nothing has been told your marketplace is empty. No lifecycle email undoes that. Measure the zero-result rate and treat it as a retention metric.
- Demand: make the second transaction easy, not the first. The gap between first and second purchase is where marketplaces lose people, because the first was solving a problem and the second requires a habit.
Yelp is the long version of the supply lesson. It launched into a market that already had Citysearch, and its early growth was slow because a review site with few reviews helps nobody. Yelp's slow start is worth reading next to your own supply curve rather than in the abstract.
Section 04App user retention in a marketplace
Most marketplaces are app-first on at least one side, and mobile app user retention behaves differently enough to deserve its own read. The curve decays hardest in the first week, and an uninstall is a harder stop than a lapsed web session: you lose the notification channel at the same moment you lose the user.
Two things follow. Measure app retention by day rather than by month for the first fortnight, because a monthly cohort hides the whole shape of the drop. And treat notification permission as an activation step, not a settings detail, since on the demand side it is often the only route back after a lapse. Both belong next to the rest of your customer retention metrics rather than in a mobile silo.
Before your next retention meeting, split last quarter's cohort in two and plot supply retention against demand retention on the same axes. Most marketplace teams have never seen those two lines together, and the one that falls first is the only one worth arguing about.
Section 05FAQ: customer retention in marketplaces
How do you increase customer retention in a marketplace?
Start by splitting the number. A marketplace retention rate blends supply and demand, two populations that leave for different reasons, so the blended figure cannot tell you which side is leaking. Measure each side on its own cohort and its own natural period, then fix the side whose curve falls first. Supply is usually held by getting sellers to a first transaction quickly and by removing operational effort, while demand is usually held by eliminating dead-end searches and making the second purchase easy.
Why is marketplace retention different from normal customer retention?
Because you have two products sharing a database. A seller's job is to earn and a buyer's job is to find, so almost nothing that retains one retains the other. The asymmetry matters too: demand churn costs you a customer, but supply churn shrinks the product for everyone still using it, since fewer listings make search worse, worse search loses buyers, and that pushes the next tier of supply out.
Should you measure supply and demand retention separately?
Yes, and with different definitions of active. A seller counts as active if they could transact, meaning a live listing, available stock and responsiveness, while a buyer counts as active only if they actually transacted. Use each side's natural cadence rather than forcing both into monthly cohorts, and weight supply by contribution as well as headcount, because losing your top decile of sellers is not the same event as losing the bottom decile even though a simple retention rate scores them the same.
How should a marketplace measure app retention?
Not against an industry figure, because the right number depends on the transaction cadence your category supports: a food delivery marketplace and a property marketplace have completely different natural repeat intervals, and comparing them tells you nothing. Measure by day rather than by month for the first fortnight, since monthly cohorts hide the shape of the early drop, and compare each cohort against your own previous cohorts rather than an industry average.
Rishikesh Ranjan
I reverse-engineer how the fastest-growing companies actually grow, then run the same plays for a living. First growth hire at ngram, growth at StreamAlive, Spaceonova, CloutFlow, and Airbook. Teardowns, playbooks, and field notes for people already in the trenches.
