Real-time bidding runs on averages, and averages lie about distributions. If genuine human attention and synthetic noise are pooled into one blended value, the optimizer learns a price that is wrong for both: too low for the rare, valuable impression and too high for the abundant, worthless one.
This is the "flaw of averages" applied to auctions. The bidder is not stupid; it is faithfully optimizing a number that has quietly averaged away the very thing it was supposed to find. Feed it a bimodal world and it will confidently aim for the empty middle.
- Gold (scarce, high-value human attention) gets underbid and lost to someone who segmented it out.
- Junk (abundant synthetic supply) gets overbid because it looks "average enough."
- The auction spirals toward the cheapest inventory: a voyage to the bottom of the price barrel.
The fix is not a smarter average. It is separating the distribution before you optimize (verify first, then bid) so the machine is aiming at a real target instead of a statistical mirage.
Futureproof TMT
Seeing something like this in your own data?
Our team advises enterprises, counsel, and families on the automated internet.