Seeding a Marketplace Is Not a Growth Loop

Every cold-start guide lands on the same checklist. Work out which side is harder to acquire and put your money there. Seed the thin side by hand so nobody hits an empty result. Geo-fence the launch so density arrives before demand does. That checklist is correct, and I have run it.

It also is not a growth engine, and mistaking it for one is why marketplaces that look alive at launch go quiet two quarters later.

Seeding is a subsidy. A subsidy buys the transaction you paid for and nothing after it. The thing that buys the next transaction without you is a loop, and most teams never write theirs down. They draw a circle on a slide, call it a flywheel, and go back to funding the subsidy.

Funnels need refilling. Loops do not.

Funnels are linear. Money goes in the top, customers come out the bottom, and when you stop putting money in, growth stops. Loops are different in one specific way: the output of one cycle becomes the input of the next. Customers produce something (a referral, a listing, a review, a page, a data point) that goes back in and brings the next customer.

That is a mechanical difference, not a philosophical one, and it is testable. If you cannot name what the output is and how it re-enters as input, you have a funnel with a nice diagram over it.

Name which of the four you are actually building

There are four types worth separating, because they need different investment and they fail differently.

Viral loops. A customer brings another customer directly. Referral programs, invites, sharing.

Content loops. Using the product creates an asset that acquires the next user. User reviews, listings, public profiles, anything indexable that exists because somebody transacted.

Data loops. Usage produces data that makes the product better, which brings and holds more users. Netflix and Waze are the clean examples. Every trip makes the routing better, which is why the next driver shows up.

Economic loops. Each customer improves your unit economics, which lets you buy the next one at a price competitors cannot match. Density lowers your cost to serve, scale earns you better rates, retention lifts LTV, and the extra margin goes straight back into acquisition.

Nearly every team I talk to says viral and means economic. They fund a referral program, which is a viral loop, while the thing actually compounding underneath them is a cost-to-serve advantage that nobody is measuring, defending, or reinvesting. If you are only going to run one loop properly, run the one that is already turning. Working out which one that is, before you fund anything, is a good use of a 20-minute conversation.

The sentence you have to be able to finish

Before I let a loop into a plan I make somebody finish this out loud, with the real nouns from the business:

A [customer] does [action], which produces [output], and that output brings in the next [customer] because [reason].

You are not allowed "and then it goes viral." You are not allowed "brand awareness." If the sentence needs an "and then marketing does something," the marketing spend is the input and you are back to a funnel. Half the loops I see die at the last clause, because the honest answer is that the output sits in a database and nobody ever routes it back to the front.

That sentence is the whole Loop Canvas in miniature. Write it for each candidate loop, then throw out the ones you cannot finish.

Then pressure-test it with K

For a viral loop, the sentence is not enough, because a loop that turns slowly is not an engine no matter how good the story is. The formula:

K = customers × shares per customer × click rate × conversion rate

In practice you read it per customer. Multiply the three rates and you have the number of new customers each existing one brings back. Five percent of buyers share, half the recipients click, ten percent of those convert, and each customer is returning 0.0025 of a customer. That is not a rounding error you scale your way out of. That is a program you should either fix or stop describing as growth.

My rule: below about 0.5, it is an assist, not an engine. Above 1.0 it is self-sustaining and you have a very different company. Between 0.5 and 1.0 it is a real accelerant on paid acquisition and it deserves headcount.

The reason I hold that line is that both of the best loops I have built sat under it.

On The RealReal's seller side, the loop was clean. A seller consigns, shares with friends, and the friends become sellers, with a $50 credit as the reward. The offline version of it, consignment parties, worked better than anything we ran online. The K-factor got to 0.4, and 40% of new sellers came in through it.

By my own threshold, that is an assist. It also brought in nearly half the sellers on a supply-constrained business and cut CAC materially. Both statements are true, and holding them at once is the point. The failure mode is not the 0.4, it is the board deck that calls a 0.4 loop the acquisition strategy and quietly defunds the paid engine that is doing the other 60%.

At Cubii we built a two-sided referral, $25 credit for the referrer and $25 off for the friend, triggered 30 days after purchase rather than at checkout, because on a physical product nobody recommends a thing they have not used yet. We optimized it through five versions and got K to 0.18, which was 18% of new customers. A good assist. Never once the plan.

Sizing the reward is arithmetic, not taste

Two numbers save a lot of arguing.

Set the total reward at 10% to 25% of your CAC. Below that it does not move behavior, above it you are buying customers at a premium through a channel you told the board was cheap.

On a two-sided split, go 50/50 or 60/40 favoring the referrer. The person doing the work of recommending you is spending social capital, and that is the scarcer input.

The loop and the subsidy belong on different sides

This is where cold-start advice quietly misleads people. The side you subsidize is usually not the side your loop belongs on.

At Shiftgig, businesses were harder to acquire than workers. So the investment went disproportionately into B2B: targeted outbound, industry events, partnerships with staffing associations, while the worker side came through cheaper scalable channels like social, SEO, and job board syndication. More than 20,000 businesses came onto the platform.

The subsidy went to the constrained side because that side would not fill itself. The loop, when it works, belongs there too, and that is the harder build by definition. On the other side, cheap acquisition is already doing the job, and a referral program there will take a quarter to build and move very little.

What I would do first

Write the sentence for every loop you believe you have. Most teams find they have one, and it is rarely the one on the slide.

Then compute K on the viral one, honestly, using shares that actually happened rather than shares you enabled. If it comes back under 0.5, keep the program, cut its line in the plan to what an assist deserves, and put the argument back where it belongs, on the loop that is already turning without a coupon attached.

Seeding gets you a marketplace that works. A loop is what keeps it working after you stop paying for it, and those are two different projects with two different budgets. If you want to find out which parts of your growth are still on subsidy, the Pain Ladder Diagnostic is a reasonable place to start.