Predictive Bidding Model

Hypothesis

Ad platforms are designed to be black boxes to marketers. But we know a couple of things. They cannot escape the laws of diminishing marginal returns and supply and demand.

My core question was simple:
Are we bidding past the point of commercial efficiency without knowing it?

I hypothesised that paid channels exhibit measurable incremental and diminishing return curves, and that hidden inefficiencies in bidding were compressing net margin across commercial dimensions.

Approach

This project genuinely excited me because it sat at the intersection of economics, data, and real commercial impact.

I treated each platform as a market.

Using my economics background, I structured both:

• Supply-side signals: CPC, CPM, spend, cost efficiency metrics

• Demand-side signals: volume, revenue, net margin, conversion elasticity

The goal was to map where each platform sits along its effective demand curve, identifying:

• Elasticity of returns

• Marginal efficiency thresholds

• Structural supply constraints

• Points where incremental spend begins eroding commercial gains

I tested multiple regression specifications to model non-linear curvature in spend vs outcome relationships. After extensive validation and out-of-sample testing, I selected a LOESS model for its ability to capture real-world non-linearity while maintaining strong predictive performance.