How Carousell Seeded a Marketplace With 70 Sellers and 700 Listings
Short answer: Launch a marketplace in one narrow category and location, recruit supply manually, and publish enough current listings that a buyer has at least five credible choices for the main search. Do not chase equal numbers of buyers and sellers. Seed the constrained side first, measure searches that return useful results and keep recruiting until transactions repeat without founder intervention.
Marketplaces do not fail merely because one side is small. They fail because the two sides cannot find a relevant match at the required time, price and place.
Seven hundred random listings can feel emptier than 70 current listings concentrated around one buyer need.
Carousell's manual starting inventory
Carousell's founders spent three months visiting flea markets at weekends and interviewing sellers and shoppers. They approached about 100 sellers; roughly 70 agreed to contribute around 700 initial listings.
The company's account of its early years says the founders also supplied their own items, including a camera, iPad and MacBook. A database failure around launch erased the recruited listings, showing how much the marketplace depended on manually assembled supply.
That evidence establishes the seeding method, not marketplace economics. Public sources do not provide buyer traffic, time to first sale, listing conversion, recruitment cost or the point at which organic sellers replaced field work.
My position is that a marketplace should launch dense and narrow, even when the vision is broad. Geography, category and time all reduce liquidity. Starting with all of them wide creates an impressive catalogue and a poor buying experience.
Use the Marketplace Liquidity Seeding Loop
The Marketplace Liquidity Seeding Loop connects five measures. Each pass should improve the next.
| Step | Measure | Initial working threshold | |---|---|---:| | Recruit sellers | Qualified sellers agreeing to list | 50% of approached sellers | | Create supply | Live, complete listings per seller | Five | | Produce choice | Relevant results for target searches | Five current options | | Create contact | Searches leading to buyer action | 20% | | Complete and repeat | Transactions and relisting | Three repeat sellers |
These thresholds are indicative, not market standards. A property marketplace may need fewer listings and longer cycles. A low-cost fashion market may need much more choice. Set your numbers from the buyer's decision.
Choose the constrained side first
Ask which side arrives and leaves faster. Buyers usually leave an empty marketplace immediately. Sellers may wait if listing is easy and expectations are clear, so supply often needs seeding first.
There are exceptions. A labour marketplace with many available workers but no credible jobs may need anchor demand. Measure existing alternatives, time sensitivity and switching effort rather than applying a universal supply-first rule.
Interview both sides about the last completed transaction. The guide on [how to get your first 10 customers](/sales-and-marketing/how-to-get-your-first-10-customers/) is useful for founder-led recruitment, but treat each side as a distinct acquisition problem.
Define one liquidity cell
A liquidity cell is one category, location and time window in which a match should occur. “Second-hand goods” is not a cell. “Used cameras available for collection in central Manchester this week” is.
List the ten searches or requests the first buyers are likely to make. Recruit inventory that answers those searches. A buyer should see enough credible options to compare price, condition and convenience.
Do not manufacture supply with stale or fictional listings. Founder-owned items are acceptable when genuine, described accurately and available on the stated terms.
Recruit and list alongside sellers
Early sellers may agree with the idea but never create a listing. Reduce work by photographing, writing or importing the first items with permission. Observe which fields affect buyer trust and transaction completion.
Track the full funnel: approached, agreed, first listing, complete listing, buyer contact, transaction and relisting. Carousell's reported result, 70 sellers from 100 approaches, implies a 70% agreement rate, but the later stages are not public.
Manual listing is a learning method, not a permanent operating model. Record minutes per listing and design onboarding to remove the repeated work.
Measure liquidity rather than registrations
The core measure is the percentage of high-intent searches or requests that produce a suitable, current match. Pair it with time to first buyer action and transaction completion.
Registrations can rise while liquidity falls if supply spreads into more categories or locations. Report each cell separately. Also remove sold, expired and unresponsive listings quickly, because apparent choice damages trust when it is not available.
Consumer, payments, data, product-safety and marketplace obligations vary by country and category. Clarify whether you are an intermediary or seller, and obtain qualified local advice on identity, prohibited goods, complaints, fees, tax and liability.
Worked example: ReList Market
ReList Market launches used music equipment in Leeds. These figures are illustrative.
The founders approach 80 sellers at rehearsal studios and weekend markets. Forty-eight agree, and 35 publish at least one complete listing. Agreement is 48 divided by 80 = 60%, while activation is 35 divided by 48 = 72.9%.
They help create 245 listings, seven per active seller. Manual recruitment and listing take 126 hours at £22, costing 126 x £22 = £2,772.
During the first month, 600 target searches occur. Four hundred and twenty return at least five relevant current results, so search liquidity is 420 divided by 600 = 70%. Ninety searches produce a message, 38 produce a reported sale and 12 sellers relist.
| Measure | Result | |---|---:| | Cost per activated seller | £2,772 divided by 35 = £79.20 | | Search-to-message rate | 90 divided by 600 = 15% | | Message-to-sale rate | 38 divided by 90 = 42.2% | | Seller repeat rate | 12 divided by 35 = 34.3% |
The marketplace misses its 20% search-to-action guide. Support records show that buyers distrust listings without collection times. ReList makes availability mandatory and asks sellers to confirm weekly. In the next cohort, search-to-message rises to 23% without adding categories.
ReList does not expand nationally. It first requires 80% search liquidity, 20% action and 15 monthly sales for two consecutive months in the original cell.
Build recovery before launch
Carousell's lost listings demonstrate that seeded supply is an asset worth protecting. Back up the database, test restoration, retain source images and exports with permission, and know who can restore service.
Do not assume a cloud service provides the backup and recovery you need. Test a restore before public launch. Protect personal data and delete it according to your lawful policy.
Have a manual contact plan for the first sellers. If a failure occurs, explain what was lost, what you are doing and whether they need to act. Trust can survive an early incident when communication is direct and accurate.
Run a four-week cell launch
In week one, define the cell and ten target searches. During weeks one and two, approach 50 to 100 relevant suppliers and help activated sellers create five to ten genuine listings each.
In week three, invite a small buyer group and record search liquidity, contacts and completion. Fix freshness, trust and matching problems before acquiring more traffic. In week four, require repeat listings or transactions. Expand one dimension only after the original cell meets its thresholds for two cycles.
Related guides
Frequently asked questions
How many listings do I need before launch?
Enough to give the target buyer credible choice in the searches you promise to serve. Start with the ten most likely searches and aim for at least five current, comparable results for each as a working guide. One listing can be sufficient for a rare, high-value request, while fashion may require dozens by size and style. Count current, responsive listings, not database rows. Test with real buyers before public promotion. If they cannot find a match, narrow the cell or recruit specific missing supply.
Should I pay sellers to list?
Only when the subsidy purchases genuine, relevant and available supply, and you can prevent low-quality listings created solely for payment. Reimburse specific work or guarantee a limited early outcome rather than paying for account creation. Calculate the cost per activated seller and completed transaction. A subsidy can reveal marketplace behaviour, but it can also disappear when payments stop. Tell users about material commercial relationships where required. Incentive, advertising and tax rules vary by jurisdiction, so obtain qualified advice for the arrangement.
Can I list sellers' products for them?
Yes, with clear permission and accurate information. Assisted listing is useful because it exposes the fields, photographs and trust signals that matter. Confirm price, condition, availability and who handles buyer enquiries. Give the seller control to correct or remove the listing. Do not scrape or copy listings without a lawful basis. Intellectual-property, data and platform terms vary, so seek qualified local advice when importing material. Track the founder time and reduce it through better onboarding once the repeated workflow is understood.
Which side should pay first?
Charge the side receiving the clearest, measurable value and for whom payment least damages early liquidity. That may be sellers through listing or transaction fees, buyers through service fees, or neither during a capped pilot. Test the fee on completed value rather than access to an empty market. Include payment, dispute, support and refund costs. Keep terms clear. Consumer, tax and payment rules differ by country. A free launch can support learning, but it should have a dated pricing test.
When is a marketplace ready to expand geographically?
Expand after one cell produces repeat matches without continuing founder seeding. Require target-search liquidity, buyer action, completed transactions and repeat supply for at least two normal operating cycles. Then add one adjacent location while holding the category stable. Expansion can reduce overall marketplace liquidity if staff and sellers are spread too thin. Compare the new cell separately and actively protect the original service. A national waiting list is not proof that any local market will be dense enough to work.
What should I do after a marketplace data loss?
Stop writes if necessary, preserve evidence, assess personal-data impact, restore from a tested clean backup and communicate with affected users. Do not recreate listings from memory or conceal what disappeared. Follow applicable breach-notification and consumer requirements, which vary by country, and obtain qualified technical and legal advice. After recovery, document the cause, backup frequency, recovery objective and named owner. Run another restore test. Early users may forgive an incident, but repeated untested recovery turns a technical error into an operating failure.
Comments (0)