Why Adgent
Product All features It reads Account memory Daily verdict Ground truth Trust Gate Creative intelligence Change Ledger It builds Campaign build Media plan Live dashboards Agents & reports
Solutions By team For agencies For in-house By industry Ecommerce & DTC Lead generation Travel & hospitality Marketplaces Local & multi-location Mobile apps & games
Pricing About Blog Request demo

For marketplaces

The platform counts the transaction. You bank the take rate.

A marketplace reporting 6× ROAS at a 13% take rate is running at 0.8× on the revenue it actually keeps. That is not a rounding error, it is an order of magnitude — and it is the default state of every marketplace ad account where the conversion value is set to transaction value. Which is most of them.

what the pixel counts as revenue gross what you actually bank 13% take rate
Reported return against real revenue
The gap, from public filings

Every marketplace knows this number. Almost no ad account uses it.

These are not estimates — they are gross volume against net revenue, straight out of FY2025 reporting. The last column is how far a return figure built on gross volume sits from reality.

MarketplaceGross volumeNet revenueTake rateReturn overstated by
Airbnb$91.3B$12.24B13.4%7.5×
DoorDash$97.0B$13.7B13.5%7.4×
eBay [Q4]$21.2B$3.0B14.2%7.1×
Etsy$11.92B$2.88B24.2%4.1×

FY2025 investor reporting. Gross volume is named differently by each — GMV, GOV, GBV, GMS — which is itself part of the problem.

DoorDash states it plainly in its own filing: sales and marketing was 2.5% of gross order value and 18.1% of net revenue. Same spend. Two numbers seven times apart, and only one of them is the business.

A setting worth checking today
We are at 6× ROAS. That is healthy, right?
Check what the conversion value is set to first. On this account it is passing gross transaction value, not your take.

At a 13% take rate, a reported 6× is about 0.8× on the revenue you actually keep. The algorithm has been bidding against the bigger number all quarter, exactly as instructed.
How much has that cost us?
I can show you the spend that sat above break-even under the corrected value — it is the majority of the last two months. Fixing the field takes a minute; unlearning it takes the algorithm a fortnight.
Corrected valuesame spend
what the algorithm bids against gross transaction value what you actually keep 0.8× at a 13% take rate same spend, one field apart a minute to fix the field; a fortnight for the algorithm to unlearn

Nothing in the interface flags it.

Two numbers, seven times apart, for the same spend.

DoorDash states it plainly in its own filing: sales and marketing was 2.5% of gross order value and 18.1% of net revenue. Your bidding is optimising against whichever one the conversion value happens to carry.

  • the ROAS reported off gross transaction value
  • 0.8×the same campaign on the revenue you keep
  • 13%take rate is the whole distance between them
The second thing the pixel cannot see

You have two costs of acquisition. It can only count one.

A marketplace has to buy buyers and sellers, and the two are not remotely comparable. Take a marketplace clearing $85 an order at a 13% take rate: allowable buyer acquisition cost lands near $11, while an active seller — who keeps producing orders for years — can be worth many times that.

An order-of-magnitude difference in what a click is worth, inside one ad account, invisible to every platform reporting it.

Setting conversion value on a marketplace

Never set a target return off gross volume.

It is a single field, it is wrong by default in most implementations, and it instructs the bidding algorithm to overbid by the inverse of your take rate — permanently, and with total confidence. At a 13% take rate, a reported 6× is an actual 0.8×. Nothing in the interface flags it, because from the platform's point of view nothing is wrong.

And the third mistake: bidding against your own supply

Marketplaces routinely buy branded and category terms that their own sellers, or their own organic listings, were already winning. eBay's controlled shutdown of brand-keyword advertising found the traffic simply rerouted to its own organic results; measured return on that spend was negative[Blake, Nosko & Tadelis, Econometrica, 2015].The reported figure was real. The incremental one was not.

None of which corrects itself while you decide. A conversion value left on gross volume overbids by the inverse of your take rate every day it stays there — at 13%, that is a reported 6× sitting at roughly 0.8× on the revenue you keep, for another ninety days of spend.

Fixing the field takes a minute. Fixing what the algorithm learned in the meantime takes a fortnight on top, so a quarter of waiting costs a quarter of mispriced bids plus the unlearning.

Where Adgent stands today

The take rate lives in your systems, and we should say so.

Net revenue, take rate by category, and cost of acquisition per side sit in your own backend or warehouse — there is no standard connector for that the way there is for a storefront. Adgent does not read it today. What it does is everything upstream of that number.

01

It reads every account daily, geo by geo

Breakdowns by geography are part of the read, not a report you have to go and build. A campaign performing on average while failing in half its cities is the normal case, not the edge case.

02

It pushes back on targets your history cannot support

The Trust Gate refuses a target the account has never reached, and flags campaigns optimizing toward an event that carries no real value — which is precisely what a conversion valued at gross volume is.

03

Every change is logged and reversible

In an account where one conversion-value setting can misprice every bid, seeing exactly what changed, when and by whom — and putting it back on your approval — is not a compliance feature. It is how you find the setting.

Next

This is the ground-truth problem in its most extreme form.

Every vertical has a gap between what the pixel counts and what the business banks. In marketplaces that gap is a multiple rather than a margin — which makes it the clearest case for judging spend on a number the platform cannot see.

Running these accounts for clients?

The same read on every client you run — including the ones nobody opened this week. A seller client who learns their real return from you, rather than from their own finance team, is a client who renews.

How this works across a book of accounts
Free conversion-value audit

Find out whether you are bidding on gross volume.

It is one field, it is wrong by default in most implementations, and it instructs the algorithm to overbid by the inverse of your take rate. We will check it in an afternoon.

  • Conversion value and target settings, read directly
  • Geography broken out, because liquidity is local
  • Take-rate reconciliation needs your warehouse — not yet
What comes back

What your accounts are actually optimizing toward, what that means at your take rate, and which cities are being bought into with no supply to fill them.

We usually reply within a couple of working days.

Thanks — we’ve got it.

We’ll be in touch within a couple of working days.