How to diagnose whether bid shading is helping or hurting you — step by step

How to diagnose whether bid shading is helping or hurting you — step by step

Bid shading (the algorithm that lowers your first-price bid toward the predicted clearing price) is a black box on most DSPs. Here's how to interrogate it.

1. Export bid-level logs with three fields per auction: your submitted bid, the shaded bid actually sent, and the clearing price when you won.

2. Compute the shave: (submitted − shaded) / submitted. Bucket auctions into deciles by this value.

3. In each decile measure two things — win rate and surplus, where surplus = (shaded bid − clearing price). Surplus is the money the shader left on the table that you didn't need to spend.

4. Look for the inversion point: the decile where pushing the shave deeper starts dropping win rate faster than it grows surplus. That's the shader's efficient frontier.

5. If high-shave deciles show near-zero surplus AND collapsing win rate, the model is over-shading you out of inventory. If low-shave deciles show large surplus, it's under-shading and overpaying.

Most shaders are tuned for the median auction and mishandle your tails — branded direct deals get over-shaded, remnant open exchange gets under-shaded.

Why it matters: You cannot fix what you cannot see. Logging the submitted-vs-shaded delta turns an opaque margin into a controllable lever, and the inversion-point analysis is the only honest way to know if your DSP's shading is working for you or for the exchange.