Link to home

News from the open internet


The door in the floors: Transparency, price discovery, and market efficiency in programmatic

Three hands place small blue megaphones within rooms in a dollhouse.

Illustration by Nick DeSantis / Shutterstock / The Current

There was considerable ink spilled about The Trade Desk’s decision to start sending bids below floors. There was a lot of hand wringing about the intent of this change rather than a substantive discussion about why this change was necessary and important. Our decision to bid below the floor was not to say floors were a problem but the ways in which floors were being broadcast was. We believe floors should exist. They’re an important part of how publishers manage their inventory. However, what we observed in the bid stream makes it abundantly clear that “floors” represented more of an ask than a hard data point.

Floors should be straightforward, but they have been so muddied by duplication, predatory practices, and overall black-box mechanics that at best we can only treat them as a suggestion. A principal attribute of any scaled market is accurate price discovery. Without it, there can be no confidence. Without confidence, investment recedes. To fix floors we need to bring as much data to bear as possible.

In a perfect world, floor prices sent to a DSP represent the minimum price necessary to purchase an impression. It’s a simple mechanism that allows publishers to establish a baseline price at which they will accept demand. In our perfect world, this floor is not arbitrary but based on careful study and analysis of demand where price and fill rate converge to maximize yield efficiency. This floor would be consistent and constant, save adjustments made by the publisher when yield curves move in one direction or another. Buyers would have a reliable signal, which they could use to price their bids at or above this floor price to maximize value, derive their own price discovery and look for equally calibrated ROI.

These floors would be the same regardless of path and play an important role in understanding value across the open internet. It’s so perfect I could wrap myself in it like a cashmere blanket.

But, my friends, brace yourselves. We don’t live in this perfect world. Not even close. Our world is incredibly complex with undulating waves of abstraction, specifically in regard to price floors. Let me explain. The average premium publisher works with approximately 26 different SSPs and exchanges. Our buyers now see every impression at least 26 times, and sometimes many more. More problematic is that each identical impression is represented differently to buyers specifically regarding floors. If floors represent a minimum price, why does an identical impression have 26 different floors? As buyers, we can only surmise but the conclusion based on this dissonance is not difficult. These floors are not floors at all. They are suggestions at best and manipulations at worst.

I don’t believe publishers are aware of the ways their inventory is being represented by intermediaries to buyers. Having full control and insight into how over three dozen partners federate bid requests is a herculean task.

Which brings me to my second point: I don’t believe these are publisher-set floors (most of the time). But it’s near impossible in real time for buyers to ascertain the true floor price because of all the noise being added to the supply chain.

And so to a third point: Some floors are dynamic, machine trained, and algorithm-derived. These techniques can range from benign to problematic. If a publisher is using machine learning to better set market price, that’s not a problem. If a publisher or SSP is using machine learning to prey on certain buyers to artificially increase their prices, that is a problem.

Price discovery is a two-way street. Back in the early days of programmatic I worked on the supply side. We actively discouraged publishers from sending floors in the bid request. That did not mean publishers did not have a floor — on the contrary. We just didn’t want publishers to broadcast them. Why? Because it would mean missing out on critical pricing information. How do you know your floor is perfectly calibrated to the nexus of demand and price? You need to see everything above and below the line. In this context, sending a floor might be one of the most expensive mistakes a publisher can make.

We decided, based on all available evidence that floors were too inconsistent, often ripe for manipulation and predatory practices, to be a reliable indicator for price discovery. That doesn’t mean floors don’t exist or that we can win impressions at prices lower than the publisher will allow. But it does remove a noisy signal in the oRTB protocol. We will see true floors simply by studying CPM and win rate. But perhaps more important, we firmly believe that publishers are missing critical data points when determining how to accurately price their inventory. This was the primary driver for releasing bids that were below the floor. I also think it’s important to acknowledge many of our top SSPs and publishers were champions of the change for these same reasons. Much of the hand wringing I am observing is coming from folks who either don’t understand the intent of this change or have something to lose when floors are not manipulated. After all, since when is more data a bad thing?

We are still in a rollout phase. This change has not been propagated system-wide. We notified our partners early so they would understand the change. Our buyers will benefit from understanding the true win rate relative to price, and publishers get full transparency into all our demand in order to set their prices. Ultimately, market efficiency is impossible without this level of price discovery.

I do want to directly address one specific criticism as it relates to this change because sustainability is important. It’s been said that by releasing all bids to publishers (regardless of floor price) the increase in total bids would be unnecessarily wasteful. I am a champion of sustainability but the math on this does not add up. Generous bid rates for the web fall somewhere in the low single digits. Even a 50 percent increase (which is on the high end of potential impact) would only represent a fraction of the network calls associated to overall request duplication. I am all for eliminating waste, but we need to have a sense of proportion. This is a rounding error. And it is minuscule in comparison to unnecessary supply duplication.

The Current is owned and operated by The Trade Desk, Inc.