Myth: Turn on bid shading and stop worrying about clearing prices

Myth: Turn on bid shading and stop worrying about clearing prices

Most DSPs ship bid shading on by default — the engine lowering your bid toward the expected clearing price in first-price auctions — and buyers treat it as a sealed black box that handles overpayment forever.

The model has a self-corrupting feedback loop you have to watch.

1. Shading estimates the clearing price from your own win/loss data: the minimum you could have bid and still won. It shades your bid down toward that estimate plus a safety margin.

2. The problem: by shading down, you lose the marginal auctions you'd previously have won at the higher bid. Those losses remove data points from the top of your clearing-price distribution.

3. With the top of the distribution censored, the model now under-estimates the true clearing price, shades even harder, and your win rate quietly erodes — a censoring spiral. You see falling spend and assume the market got more expensive.

4. Robust shading engines counter this with deliberate exploration (occasionally bidding higher to re-sample the upper tail) and by modeling the full bid landscape, not just won auctions. Many default configs don't.

Why it matters: bid shading optimizes against the data it can see, and shading censors that very data. Without exploration, "set and forget" decays into systematic under-bidding that looks like a cooling market.