Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Measurement Planning

Measurement for Marketplaces

How a two-sided marketplace measures marketing: a funnel for buyers and one for sellers, the matched transaction, liquidity and channel credit per side.
Guideadvanced10 min readUpdated

Why one funnel is not enough

A shop has one customer and one funnel: visit, product view, basket, order. A marketplace that matches households with repair trades, clients with freelancers or guests with hosts has two funnels that meet in the middle. The buyer side runs from visit to search or request to booking and repeat booking. The seller side runs from sign-up through profile, verification and first listing to first match and continued activity. The journeys share almost nothing except the transaction where they meet.

Blending them breaks the basic numbers. A single sign-up count adds tradespeople to households. A site-wide conversion rate divides bookings by visits from both groups, so a recruitment drive for sellers lowers the reported conversion rate while improving the market. Channel reports mix a search campaign for buyers with a trade-press campaign for sellers, and a CAC averaged across both means nothing to either team.

The two sides also depend on each other in ways that one funnel cannot show. A buyer converts when there is supply in their area and category; a seller stays when there is enough demand. The growth loops guide describes the marketplace loop and its typical failure, one side arriving and finding nothing. Measurement has to show that failure where and when it happens, which means each side measured on its own terms and the connection between them measured explicitly. How to act on those measures, from finding the side that limits matches to expanding area by area, is the subject of the guide to growing a two-sided marketplace.

Recording which side a person is on

Everything below depends on knowing, for every person and every event, which side they belong to. Record it in three places.

  • As a user property in analytics. Google's documentation describes events as the unit of measurement, and the events and parameters guide explains that user properties describe the person rather than the action and attach to their later events. A side property with a short controlled list of values (buyer, seller, both) lets every report and exploration split by side. Set it when the person signs up or signs in, not on every page view.
  • As a parameter on the events that differ. The sign-up event carries the side, and events that only one side performs (a request, a listing, a quote sent) are separate events, because they mean different things to the business.
  • On the CRM and account record. The same field, the same values, so that analytics, the transaction database and the CRM split identically.

Two cases need rules. People on both sides, such as a tradesperson who also books other trades, should be recorded as both, with each event's own side carried as a parameter so the action is still attributed correctly. And visitors who have not signed up have no side yet. Use the pages they visit (the pages for becoming a provider, or a search for a service) to build a provisional Segmentation in analysis, labelled as inferred, and never treat it as the recorded property.

The matched transaction as the shared outcome

The outcome both sides share is the matched transaction: the booking, order or job where a buyer and a seller came together. Record it once, in the transaction database, with the internal identifiers of both parties, the request it answered, the area and category, its value, the marketplace's fee or commission, and its status (completed, cancelled, disputed, refunded). Analytics can record the request and the booking as events with their values, as signals for campaign reports; the transaction database is the record.

Two value measures sit on every matched transaction, and confusing them is the commonest reporting error in marketplaces.

MeasureWhat it isWhat it answersWhat it hides
Gross transaction valueEverything buyers paid through the marketplaceHow large the market the business runs isHow much of it the business keeps
RevenueThe marketplace's own share: fees, commissions, subscriptions, after refundsWhat the business earnsWhether the market is growing in activity
ContributionRevenue after payment costs, incentives, support and refundsWhat acquisition can afford to spendNothing it should; it is the figure for budget decisions

Judge acquisition on revenue or contribution, because a channel that brings large transactions at a low take can look strong on transaction value and lose money. Report transaction value beside it as the measure of market size.

One more outcome deserves a column: repeat transactions between the same buyer and seller. A pair that books once through the marketplace and then never again may have stopped needing each other, or may be dealing directly. Read repeat rates by pair and by category before deciding which.

Liquidity: measuring whether the market works

Liquidity is whether people on each side find what they came for. There is no standard figure for a healthy marketplace, and the measures below should be defined from the business's own data, with thresholds set from its own history rather than borrowed.

  • Match rate: the share of requests or searches that end in a match within a time the business sets.
  • Time to match: how long a request waits for an accepted response.
  • Time to first match for a new seller: how long after verification a new seller waits for the first job.
  • Active seller share: the share of sellers with at least one match in a period.
  • Searches with no results: buyer searches that showed no available supply, by area and category.

Read every one of these locally. A marketplace-wide match rate averages full markets with empty ones, and the empty ones are where buyers leave and new sellers give up. Areas multiplied by categories quickly become a large grid, so start with the combinations that carry most requests and most spend, and add others when a decision depends on them. Report liquidity beside acquisition, because liquidity decides what the next pound of acquisition is worth. The academy exercise on measuring liquidity and the matched walk for a dog walking marketplace practises recording which side each person is on, building a funnel for each side and setting these measures before deciding where to expand next.

Channel credit on each side

Each side has its own channels and its own cost per acquired customer, and each should be judged on the outcome that side contributes.

Buyer channels are judged on matched transactions and the revenue they produce, not on requests. A request that finds no seller cost the same to acquire as one that became a booking. Where liquidity differs by area, buyer campaigns aimed at areas with spare supply are worth more than the same campaigns aimed at areas without it, and the reports should show that.

Seller channels are judged on sellers who reach a first match, not on sign-ups, and on the demand those sellers serve. A new tradesperson in an area where requests already go unmatched adds transactions; the same tradesperson in an area with idle supply adds nothing but competition. The fairest measure of a seller campaign is the matched transactions its sellers completed in the places where demand was waiting.

The cross-side effect is what ordinary attribution cannot see. A seller campaign raises buyer conversion in the areas where it adds supply, and no click path connects a tradesperson's sign-up to a household's booking. Acquisition cost and lifetime value per side, worked out as the LTV and CAC guide describes for a single customer, are useful within a side; the balance between sides is a question for liquidity measures and tests.

Testing across both sides

Experiments in a marketplace leak. If a test gives some buyers a discount and they book more, the sellers they book are less available for the control group, so the control looks worse than it would have without the test, and the measured difference overstates the effect. The same happens in reverse with seller incentives. Tests that randomise people on one side should be read with that interference in mind, and are safest for changes that do not consume shared supply.

For marketing questions, regions are the natural unit. The incrementality testing guide describes geo tests that treat some regions and hold out others, and in a marketplace they have an extra advantage: they measure the whole local market, both sides at once. Add seller recruitment in some areas and not in comparable others, and compare matched transactions, match rate and buyer conversion in each. Watch spillover where sellers serve neighbouring areas, match the areas carefully because they are few, and run the test long enough for new sellers to reach their first jobs. The result is Incrementality for the market, which is the only measure of a seller campaign's full effect.

Fee and price changes are the hardest tests of all, because they change both sides' behaviour at once. Run them by region where possible, and read them on contribution and liquidity together.

The numbers also need defending. Marketplaces invite self-dealing and fraud: sellers booking themselves to climb rankings, buyers and sellers colluding to claim referral credits, duplicate accounts collecting first-order incentives, fake requests from competitors. Flag suspected cases in the transaction data, exclude them from acquisition and liquidity reports, and report how many were excluded, because a channel that brings fraudulent accounts looks excellent until someone checks.

A two-sided KPI tree

A KPI tree for a marketplace has two branches that meet at the matched transaction. Each level is a stated function of the one below, so a fall at the top can be traced to a side, then to an area and category.

LevelMeasureHow it is builtSideSource
OutcomeContributionRevenue minus payment costs, incentives, support and refundsBothTransaction database, finance
OutcomeRevenueCompleted transactions multiplied by average transaction value multiplied by the takeBothTransaction database
OutputCompleted transactionsRequests multiplied by match rate multiplied by completion rateBothTransaction database
DriverRequestsActive buyers multiplied by requests per buyerBuyerAnalytics, transaction database
DriverActive buyersNew buyers from each channel plus returning buyersBuyerAnalytics, CRM
DriverMatch rate, by area and categoryRequests matched within the set time divided by requestsBothTransaction database
DriverActive sellers, by area and categoryNew sellers reaching a first match plus retained sellersSellerTransaction database, CRM
DriverSeller responsivenessShare of requests answered, and time to answerSellerTransaction database
GuardrailCancellations, disputes and excluded fraudCounts and shares, by side and channelBothTransaction database

Place the north star at the top if the business has chosen one; the north star and OKRs framework covers the choice; whatever is chosen, a measure built on completed transactions keeps both sides in view. The KPI owners follow the sides: buyer acquisition owns active buyers, seller acquisition owns active sellers, and liquidity is owned jointly, because neither team can move it alone.

Illustrative example — not a real result

Stallwink Market finds its constraint

Stallwink Market is an invented marketplace matching households with local repair trades across six areas. A request counts as matched when a tradesperson accepts it within 48 hours. In the second quarter the team raises buyer spend, and requests rise, but completed jobs rise much less.

Buyer sideFirst quarterSecond quarter
Buyer spend£18,000£30,000
Requests3,2005,100
Matched within 48 hours2,270 (71%)2,960 (58%)
Completed jobs2,0402,620
Buyer spend per completed job£8.82£11.45

Split by area, the match rate in four areas moved only from 73% to 70%, while in two growth areas it fell from 66% (594 of 900 requests) to 39% (780 of 2,000). The seller funnel in those two areas explains it: 140 tradesperson sign-ups, 96 verified, and 61 with a first job within 30 days, too few for the demand the buyer campaigns created. The team moves £8,000 a quarter from buyer ads in those areas to tradesperson recruitment there, and holds back recruitment in one comparable area as a check. In the third quarter the match rate in the two areas recovers to 63% while the held-back area stays flat.

Common mistakes

  • One sign-up count for both sides. Buyers and sellers added together describe nobody.
  • Calling transaction value revenue. The business keeps its take, not the buyer's payment.
  • Judging buyer channels on requests. An unmatched request cost as much as a booking and earned nothing.
  • Judging seller channels on sign-ups. A seller with no first match has not joined the market.
  • Reading liquidity as one average. Full and empty markets cancel out; read by area and category.
  • Scaling the side that is easiest to buy. More buyers in a market without supply means more disappointed buyers.
  • User-level tests that share supply. The control group is starved by the test group; prefer regions.
  • Leaving fraud in the numbers. Self-dealing and incentive abuse flatter the channels that bring it.

The measurement planning hub places this guide beside the KPI tree and measurement plan it adapts for two sides.

Frequently asked questions

Should a marketplace report gross transaction value or revenue?

Both, labelled, and never one under the other's name. Gross transaction value is everything buyers paid through the marketplace and shows the size of the market it runs; revenue is the marketplace's own share after fees, commissions and refunds, and pays its costs. Judge acquisition on revenue or contribution, because a channel that brings large transactions at a low take can look strong on transaction value and still lose money.

How do we know whether to spend on buyers or on sellers?

Look at liquidity by area and category. Where a high share of buyer requests or searches end in a match and sellers have spare capacity, buyers are the constraint and buyer acquisition pays. Where requests go unmatched and sellers are fully booked, more buyers only produce more disappointed buyers, and supply is the constraint. The answer usually differs by place and category, so read the measures locally rather than as a marketplace-wide average.

Can ordinary attribution credit a campaign aimed at sellers?

Only partly. Attribution can credit a seller campaign with the sellers who signed up, but most of its value appears on the other side, as buyer requests that now find a match. Judge seller acquisition on sellers who reach a first match and on the matched transactions they serve, and use regional tests to measure the effect on the whole market, because no click path connects a tradesperson's sign-up to a household's booking.

Terms used on this page

Frameworks

Where this page is used

Part of these learning paths

Also related

Sources

  1. [GA4] About events — Google Analytics Help
  2. Google Analytics Help — Google Analytics Help

Related