Reviews
Part of Liquidity first: the order marketplace model problems actually arrive in
Marketplace models metrics: definitions and traps
Marketplace models metrics show whether matching works, with definitions for fill rate, response time, segment coverage and repeat rate, plus measurement traps.
Gross volume is the number marketplaces report and the number that tells them least. A marketplace can grow volume for a year while getting worse at the only thing it does, which is matching a request to a supplier fast enough that the requester comes back. Here, “marketplace models metrics” means measures of that matching job, not just volume.
These measure that instead. Figures are invented placeholders.
What to take away
- Fill rate and time to first response are the product. Volume is a consequence.
- Measure coverage inside segments, not across the whole market, because a market that works in one city and one category is not a market that works.
- Repeat rate on the requesting side is the earliest honest signal that matching is working.
Fill rate, measured properly
Fill rate is the share of genuine requests that end in a completed transaction. Two definitional traps make published fill rates useless.
Fill Rate Definition Traps
Request Definition
- Trap
- No-intent searches inflate
- Flattered
- Only supplier-reached requests
- Real failure
- Requests reaching nobody
- Fix
- Count genuine requests
Timing Bound
- Trap
- Three weeks vs one hour
- Flattered
- Single bound hides gap
- Real failure
- Experience lives in gap
- Fix
- Report two time bounds
The first is what counts as a request. If searches with no intent are included, the denominator is inflated and the number is meaningless. If only requests that reached a supplier count, the number is flattered and the real failure, which is requests that never reached anybody, disappears.
The second is timing. A request filled after three weeks is not the same product as one filled in an hour. Report fill rate with a time bound attached, and report at least two bounds, because the gap between them is where the experience lives.
Time to first response
For most marketplaces this predicts everything else. It is also the number that degrades quietly as the supply side gets busier, because the suppliers who respond fastest are the ones who fill up first.
Time to First Response
- Average responsea few hours
- No response inside windowone third
Report the distribution, not the average. An average response time of a few hours in a book where a third never gets a response at all describes nobody.
Report the share with no response inside the window separately, because that group is the churn you have not seen yet.
The dynamics of a system where success on one side degrades service on the other are exactly the feedback structures treated in MIT's system dynamics course.
Coverage, by segment
A marketplace is not one market. It is a set of segments defined by geography, category, price band, and urgency, and liquidity is a property of a segment.
| What to compute | Why |
|---|---|
| Active suppliers per segment | Two suppliers is not a market, however large the total |
| Requests per active supplier per week | Both too low and too high are failures |
| Share of segments above your own liquidity threshold | The honest measure of how much of your market works |
| Fill rate by segment, sorted | The tail is usually far worse than the average implies |
The aggregate figure is a weighted average of working segments and dead ones, and it will mislead you for as long as you use it. What liquidity means and why it is the product at all is set out in marketplace models.
Repeat rate on the requesting side
The requester who comes back has told you the match worked, in a way no satisfaction score does. Measure it by cohort and by segment, and measure the second transaction separately from the third, because the drop from first to second is usually much steeper and has a different cause.
When repeat rate is high and volume is flat, you have a working product and an acquisition problem. When repeat rate is low and volume is growing, you are buying transactions, and growth will stop when spending does.
They need opposite responses, but gross volume reports them the same, and the cost side is the arithmetic in unit economics.
Concentration, on both sides
Ten suppliers at 30, 18, 12, 10, 8, 7, 5, 4, 3 and 3 percent of volume means the top three carry 60%. Losing the largest leaves 70% and requires 43% growth from the rest to return to level.
Measure it per side and per segment. A marketplace with thousands of suppliers where six carry a category has six suppliers there. Its power to set rules or fees there is much smaller than the headline count suggests.
Where the largest participants can negotiate terms others cannot get, differences in prices charged to competing business buyers may raise issues under the Robinson-Patman Act, which the Federal Trade Commission (FTC) enforces.
What these will not tell you
They will not tell you what the fee should be, which is a pricing architecture question, and they will not tell you whether the business shape is right, which belongs in business model types. They tell you whether the matching works, which is the prior question and the one most often skipped.
Common questions
What is a good fill rate?
No number here would be honest. What is useful is your own fill rate by segment, sorted, and the trend, because a marketplace competing against a segment's status quo has a different bar in every segment.
Should we count a match that settles off-platform?
Count it as a match and not as revenue, and track the two separately. Pretending it did not happen makes the matching look worse than it is and the leakage invisible.
How small can a segment be before the numbers are noise?
Small enough that you should report a count rather than a rate. A fill rate quoted on eleven requests is a decoration.
Our two sides have different repeat behavior. Which do we report?
Both, always labeled, and never blended. The supply side and the demand side churn for different reasons and on different clocks.
