Rules
Part of The five tolls that decide whether platform models actually work
Platform models metrics worth tracking, and the ones to ignore
Platform models metrics that separate the posted fee from the fee actually kept, plus the funnel leak and the side concentration that decide whether it holds.
A platform's headline fee is the number everyone quotes and almost nobody collects. Between the posted rate and the money that stays with the platform sit four deductions, and each one is usually owned by a different team.
These are the measures that make the difference visible. Every figure is invented to show the arithmetic.
What to take away
- Compute the realized rate, meaning what the platform keeps divided by what flowed through it. A posted rate is a claim about a price list.
- The interesting number in a transaction funnel is the step with the worst conversion, not the end-to-end figure.
- Concentration on either side is the risk that revenue growth will never show you.
Posted rate against realized rate
Take a month with 2,000,000 of transaction value and a posted rate of 15%, which implies 300,000. Then subtract what is actually given back:
Posted vs realized rate
- 15%Posted rate
- 1.8Promotional credits
- 1.2Seller rebates
- 2.1Payment processing
- 0.9Refunds and chargebacks
- 9.0%Realized rate
The realized rate is 9.0%, or 180,000, which is 60% of the posted number. Every one of those deductions was a deliberate decision by somebody. None of them appear in the fee schedule, and in most reporting they are booked away from the revenue line, which is why the posted rate survives in every internal conversation.
Compute this monthly. The first time is usually a surprise, and the second time it becomes the number that pricing decisions are actually made on. The structural question of what the fee is charging for in the first place is in platform models.
The funnel, step by step
End-to-end conversion tells you almost nothing you can act on. Break it into steps and the actionable step appears. On an invented 10,000 searches:
Funnel from search to payment
- 10,000 searches
- 72% usable7,200
- 55% contact3,960
- 61% price agreed2,416
- 83% paid2,005
That is 20% end to end, and lifting only the last step from 83% to 92% adds 217 transactions, an 11% increase. That is a different problem from the other three.
The first step is supply coverage, the second is discovery and trust, and the third is price and terms. The fourth is whether people settle off the platform, a rules and payments question, not a demand question.
Report the four separately every month. A single conversion figure moving up while two of the four are falling is a report that hides its own news. Feedback loops of exactly this kind are the subject of MIT's system dynamics course, and it is worth seeing why a stock and its flows behave differently from a ratio.
Concentration on each side
Take ten sellers at 30, 18, 12, 10, 8, 7, 5, 4, 3 and 3 percent of volume. The top three are 60%. Losing the largest leaves 70% and requires 43% growth from the rest to get back to level.
Measure concentration on both sides separately, and by volume rather than by count. A platform with ten thousand sellers where six carry the volume is a platform with six suppliers and a long tail. That fact changes what your rules can require, what your fee can be, and how much of the value you can keep.
Where a platform's share and conduct become a legal question rather than only a commercial one, the relevant reading starts with antitrust rather than with a pricing model.
Two measures that mislead
Registered accounts. A count of sign-ups is a marketing number. The platform measure is accounts that transacted in the period, and the two diverge quickly.
Average transaction value. An average across a two-sided book with a long tail describes no transaction anyone made. Report the distribution or report the median with the top decile beside it.
What these numbers imply about the shape of the business belongs in business model types, the cost side is worked out in unit economics, and the fee structure that produces the deductions above is a pricing architecture question.
Common questions
Should payment processing count as a deduction?
If you collect the money and pass it on, yes, because the cash never stays with you. Presenting the gross flow as revenue is the most common overstatement in platform reporting.
Our sides overlap. Some accounts buy and sell.
Then measure the roles, not the accounts. An account that does both is two rows in the analysis, and treating it as one hides which side is thin.
How do we measure transactions that leave the platform?
Imperfectly. You can watch the step where a matched pair stops progressing, survey a sample, and compare the paid rate across categories where leaving is easy against ones where it is not. None of that is exact and all of it beats assuming the number is zero.
Is a falling realized rate always a problem?
No. It falls on purpose when you subsidize one side to build supply. It becomes a problem when nobody decided it and it is discovered a year later in a margin review.
