How header bidding makes variable take rate auctions the new normal
Google’s “Project Bernanke” is an initiative designed to drive more ad spend through Google’s sell-side advertising technology systems through a wide range of potentially anti-competitive measures. A recently unsealed complaint filed by 17 State Attorneys General provided new details about the specific techniques Google employed, including the use of dynamic auction take rates.
Maybe this is an underhanded technique. Maybe it’s even anti-competitive — we’ll leave that to the lawyers. But dynamic take rates were an inevitability once publishers started implementing header bidding in 2016. In short:
Header bidding demoted exchanges from yield management to demand generation
Dynamic take rates make exchanges better at sourcing demand for publishers
But dynamic take rates also introduce risks for marketers and publishers
There is no technology platform that is optimizing for holistic publisher yield
1: Header bidding demoted exchanges from yield management to demand generation
The SSP (sell-side platform) moniker suggests publishers engage sell-side technology companies to optimize programmatic revenue. But that’s not what SSPs do, at least not anymore. We wrote about this back in 2020.
With the advent of header bidding, publishers demoted SSPs (we prefer to call them exchanges) from yield management to demand generation. A typical publisher ad serving configuration now looks like this:
Exchanges A and B are both integrated with this property through Prebid, Amazon Publisher Services, and Google Open Bidding. For each available impression, these exchanges issue three bid requests to DSPs. They then compete to deliver the highest net bid to the publisher.
2: Dynamic take rates make exchanges better at sourcing demand for publishers
The most obvious point of competition among exchanges is take rates, and there is a long running industry narrative that header bidding would trigger a price war among exchanges. All things equal, an exchange that lowers its take rate should achieve a win rate advantage vs. exchanges with high take rates. And so the exchange category should experience rapid margin compression. But that’s not quite how the industry evolved.
Consider a generic supply curve for programmatically-traded advertising:
Below the publisher’s price floor, win rates are exactly 0%. Win rates then grow in proportion to bid price and eventually plateau. This supply curve is at the core of Project Bernanke:
The exchange’s take rate is very important for DSP bids that are close to the price floor. If a DSP’s gross bid is 1% above the publisher’s price floor, the exchange can only fill the impression if its fee is 1% or less.
But the exchange’s take rate is inconsequential for very high DSP bids. If a DSP bids $100 CPM for an impression, the exchange can take a 1% fee or a 50% fee and still fill the impression.
And exchange take rates can also allow below-floor DSP bids to win the impression. If an exchange can justify taking a negative margin on some impressions, it will fill impressions that would otherwise go unsold.
Imagine a baseline scenario in which an exchange always applies a static 20% take rate to every bid. Also imagine the publisher operates a $1.00 CPM price floor, and there is no demand from competitive exchanges:
In this baseline scenario, impression #1 clears with an $8.00 payout to the publisher, and impression #2 fails to clear because the $0.72 net bid is below the publisher’s floor.
But what if this exchange operated a dynamic take rate?
By accruing an added fee on the high bid for impression #1, the exchange can subsidize the low bid for impression #2. Through this process, the DSP wins an impression it would have otherwise lost, and the publisher increases its revenue from $8.00 to $8.72. Critically, the exchange’s effective take rate is still 20% ($2.18 fee on $10.90 gross spend).
3. But dynamic take rates also introduce risks for marketers and publishers
Exchanges that implement dynamic take rates accrue extra margin on high bids to subsidize future low bids. That creates a tilted playing field where some marketers are funding bids for their competitors.
It is conceivable that Google or other dynamic fee exchanges could manage this inequity by accruing and applying fees on a per DSP seat basis. We are not aware of any exchange that takes this rigorous approach to fee management, and we think it would likely undermine the benefits — some marketers bid high and others bid low.
And so there are only a few ways for marketers to protect themselves from being a net loser in a dynamic fee auction. Marketers can disable participation in dynamic fee auctions. Marketers can negotiate a rate capping agreement with the exchange. Or marketers can always bid low.
Dynamic auction fees also create risks for publishers. Our example above demonstrates the business case in favor of dynamic take rates, but there are also three good arguments against dynamic take rates:
Pricing Conflict: Dynamic take rates allow the marketer to win impressions while bidding below the publisher’s floor price, potentially undercutting the publisher’s ability to sell directly to strategic buyers.
Revenue Assignment: Dynamic take rates can incorrectly assign revenue across properties that share an exchange account. Even if the exchange’s overall take rate with the publisher conforms to contractual terms, each property might experience an above-contract or below-contract effective take rate. (This is another scenario that could potentially be managed by a sophisticated implementation of dynamic take rates, though we’re skeptical any exchange is taking a rigorous approach to property-level revenue assignment.)
Demand Suppression: Dynamic take rates maximize the publisher’s demand from one exchange by suppressing demand from other exchanges.
To clarify that last point about demand suppression, let’s consider a more realistic scenario in which an available impression triggers bids from two exchanges:
In this example, dynamic take rates actually hurt the publisher’s revenue. If Exchange A and Exchange B both operated with a static 20% take rate, the publisher would be paid $9.00: $8.00 from Exchange A for impression #1 and $1.00 from Exchange B for impression #2. But Exchange A’s dynamic take rates allow it to fill both impressions while paying the publisher just $8.72.
This scenario of course also creates powerful incentives for Exchange B to implement dynamic take rates. And that is exactly what is happening. Google’s Project Bernanke (specifically the dynamic take rate capability) is now being copied by other major ad exchanges.
4. There is no technology platform that is optimizing for holistic publisher yield
So we’re left with two scenarios:
Dynamic take rates help publisher yield by subsidizing uncompetitive marketer bids
Dynamic take rates hurt publisher yield by depressing demand from fixed fee exchanges
Both scenarios likely happen. But which happens more often? On balance, is the yield upside of dynamic take rates bigger or smaller than the yield downside? And how would a publisher measure this tradeoff?
All of this begs the question: "Who is looking out for the publisher’s revenue?” Certainly advertisers and DSPs aren’t tasked with publisher revenue maximization. And while each exchange is incentivized to maximize the revenue it can bring to the publisher, that can actually create revenue downside for the publisher. You might reasonably think the publisher’s ad server is responsible for holistic yield management, but that’s not quite possible when Google Ad Manager is both an ad server and an exchange.
It’s worth highlighting that Google does enable publishers to disable dynamic take rates. But publishers are on their own to determine whether dynamic take rates are a net benefit to overall yield.
More generally, publishers are on their own, full stop.