Carlos Pita
Jul 18 · 2 min read

ABSTRACT

We propose a practical yet rigorous near-optimal bidding strategy for demand-side platforms (DSPs) that scales to thousands of advertising campaigns programmatically bidding in real-time (RTB) in billions of auctions per day.

The strategy is logically derived from two different — stylized but realistic — constrained global profit maximization problems, so there are no ad hoc rules for pacing, pricing, etc. The approach relies on a few assumptions that we identify and analyze, providing a basis for further extensions to other kinds of problem/contract. It is expected to find a near-optimal solution by solving a convex relaxation of the original hard combinatorial problem. It is based on Lagrange duality so it has a sound, well-known, theoretical foundation. Optimal bids for first/second-price auctions can be cheaply computed in real-time given the shadow prices of the problem constraints; on the other hand, shadow prices are daily updated by a simple subgradient descent algorithm that converges logarithmically. The algorithm should be robust in the face of noisy real-life environments and of market seasonal oscillations and structural breaks.

For a special case, we also offer an alternative derivation based on an intuitive continuous relaxation argument that reinforces our confidence in the general solution proposed here.

Read the full paper here.

jampp-engineering

Creating technology that helps mobile companies grow

Carlos Pita

Written by

jampp-engineering

Creating technology that helps mobile companies grow

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade