A marketplace does not fail because the top of the funnel is empty. It fails because one side is
readable and the other is not, and the platform keeps buying traffic for the side that was never the constraint.
The unit problem
Signups are produced by one side alone and can be bought. Matched transactions need a buyer, a seller and an
agreement inside a window the business already treats as normal, so they cannot be inflated by promoting one
side. That is why liquidity — the matched rate — is the only number both sides share, and the only honest
readable unit for a two-sided platform.
The floor, per side
The sample arithmetic does not get easier because the platform is two-sided. One strict two-variant test at a
3% base rate and a +20% relative lift needs 13,914 contacts per arm — 27,828 in total at 80% power and
alpha 0.05.
Now put the platform's own volume into it. If 400 new sellers arrive a month, one arm sees 200 of them, and the
supply side alone needs 70 months to reach its floor. If the platform sees 600 matched transactions a
month, one arm sees half, and the matched rate needs 47 months. Neither number is a verdict on the market —
it is a calendar, and it is the one deliverable an agency deck cannot produce.
You can compute both floors from your own volumes with the free planner, and read a finished test before
anything is rebuilt on it:
- Two-sided verdict planner — https://axelfreeman.github.io/marketing-mindset/tools/two-sided-verdict-planner.html
- Channel verdict fit — https://axelfreeman.github.io/marketing-mindset/tools/channel-verdict-fit.html
- The same maths as a CLI —
npx marketing-mindset sample --base 0.03 --lift 0.2
Four decisions before the first contact
- Which side is the constraint. Written down by whoever answers for revenue. If supply is the constraint, a demand test is a promise you cannot keep.
- The matching rule inside a variant. How a transaction is attributed when a buyer saw one variant and a seller another. Without it no result is ever closable, because every result has a defensible exception.
- The window that counts as matched. 7 days or 30 days — chosen before, never after. A longer window makes any variant look better, which is exactly why it gets picked after the numbers arrive.
- Who owns the number when the month ends. One name per side, one for the matched rate. A number owned by "the platform" is reported by whoever has the most to gain from the framing, and both sides will frame it.
Two tests at once is usually one test
If both experiments draw on the scarce side, they split one arm's volume and turn both into documentation. Run
the constraint side first, with the read-out date fixed in advance; keep the other side as diagnostics until the
first verdict lands.
Related: marketing engineer for marketplaces
and the public packages — $900 Sprint, $1,900/month Engine, $2,900 Full Build, written up at
https://axelfreeman.com/marketing-engineer.html
Top comments (0)