Bid shading on vs off: when the DSP's discount is working against you

Bid shading on vs off: when the DSP's discount is working against you

Bid shading is the DSP lowering your submitted bid toward the predicted clearing price in a first-price auction, so you do not overpay. It mimics the old second-price benefit. But it is a model, and models fail at the edges.

1. With shading ON: you bid $4, the algorithm predicts the impression clears at $2.10, it submits ~$2.30. You win and save. This is correct on liquid, high-volume inventory where the prediction has data.

2. With shading ON in thin auctions: on rare, high-value, low-frequency inventory the model has few comparable auctions. It shades on a bad prediction, submits too low, and you lose the impression you most wanted.

3. With shading OFF: you pay full bid every win, but you never lose a must-win impression to a bad shade. Sometimes correct for scarce retargeting pools.

4. The lever most DSPs hide: shading aggressiveness is often a single global setting. The right answer differs by line item — aggressive on broad prospecting, conservative or off on tight retargeting.

Why it matters: a blended win-rate looks fine while your most valuable segment quietly loses auctions to over-shading. Split win-rate by line item and bid-to-clear gap in log-level data. If your highest-value segment has falling win-rate and a wide shade margin, turn shading down there specifically.