*This article originally appeared on Alex Chalmer's LinkedIn profile*
Index Exchange recently announced a significant advancement in bid resolution infrastructure. Their partnership with Bedrock Platform brings auction decision-making logic for DSPs to the point where the bid request originates, reducing the latency caused by 'the wire.'
This raises the question: will decision-making at the edge result in more revenue for publishers, or is it moving market fragmentation from one part of the stack to another? My take: both can be true.
Whilst this is a compelling solution to bid response resolution, three questions remain:
- If a DSP has many SSP containers, and/or manages hybrid infrastructure, how are these coordinated with each other and with central business decisioning?
- With historical bid request and response data logged on a container, will DSPs face additional egress costs? Is this the new QPS cost throttle?
- How will sampling bias affect DSP model training? In a hybrid world, a DSP observes all requests from one SSP and a shaped selection from another. How will DSPs evaluate the resulting model drift?
The original intent of header bidding was to provide a level playing field for inventory discovery. Latency gains may improve response quality for parts of the ecosystem, but evaluating this new paradigm means understanding the downstream impact on publishers when DSPs run hybrid solutions.
At Publift, we're monitoring these effects as we optimise our header bidding solution to deliver the best result for publishers.
Of the three questions above, the sampling bias one feels least examined in industry conversations so far. Would be interested to hear if others are weighing it the same way.