Today’s column is written by Eric Berry, CEO at TripleLift.
Header bidding increases yield. This dogma is unquestioned and largely unassailable as the revenue increases are quantitative and provable.
But despite the tangible gains, questions linger about why, beyond the superficial reason of increased competition, prices are increasing and whether it is sustainable at the current level for the long term.
An impression should clear for the second-highest price submitted in an auction. Between major supply-side platforms (SSPs), there is not much unique demand.
For generic supply, such as banner inventory, the main demand-side platforms (DSPs) – including DoubleClick Bid Manager, The Trade Desk, Tube Mogul and Criteo – and leading non-DSP platforms like the Google Display Network account for the lion’s share of all demand, and will do so regardless of which SSP is used. Ensuring that all DSPs can compete against Google’s AdX on fair terms means that at least one SSP needs to bid into Google’s DoubleClick for Publishers.
This begs the question: Why would multiple SSPs bidding against each other through a header materially increase yield if there isn’t much unique demand in an SSP?
The answer is that that DSPs do not have a “true” value of an impression. The value is dependent on the SSP source of the impression, which campaigns have recently served, and random numbers. This means an identical impression from three different SSPs could yield three different bids, for three different campaigns or even the same campaign.
DSPs spent years honing their algorithms to purchase inventory in a particular way. Header bidding is effectively arbitraging components of each DSP’s algorithm to find the maximum amount that would be bid for the same impression over a few slightly different representations.
A possible solution to the arbitrage could be the OpenRTB 2.5 standard, which introduced the concept of upstream transaction IDs. This means a header bidding impression could have a normalized impression across all the different sources. While the source transaction ID is not currently mandatory, DSPs can increasingly bias their spend toward sources that do provide it – and away from those that do not – to spur its adoption.
This data allows a DSP to choose a single source for its bids for a given impression – either biasing in favor of win rate or clearing price. While it may be complicated to develop overnight, it is relatively straightforward to implement such a system. It would equally reduce a DSP’s costs by allowing it to ignore duplicative header requests.
There are certainly several benefits to header bidding beyond algorithm arbitrage, including pushing SSPs for fee transparency and reduction and moving the ecosystem to a first-price auction. For these reasons, as well as ensuring fair competition in DFP, there will always be a place for header bidding.
But as DSPs learn to manage their algorithms and optimize appropriately, and as the data becomes available that helps them normalize inventory, the relative upside from incremental sources of header bidding demand will materially decline in the foreseeable future.