Price Floor Optimization: From Static Rules to AI Floors

A price floor is the minimum CPM you'll accept for an impression. Set it too low and buyers win premium inventory for pennies. Set it too high and fill rate collapses. Most publishers set a handful of static floors and revisit them quarterly — which is roughly like pricing airline seats once per season.

Why static floors leak money

The market value of an impression swings constantly with user geography and device, time of day and day of week, advertiser budgets (quarter-end spikes are real), seasonal demand (Q4 vs the post-holiday "Q5" crash), and the individual user's value to specific buyers. A single floor for "newsdaily.com — 300x250" averages all of that. Averages are precisely where money leaks: half your impressions are floored too low, half too high.

What bid-landscape modeling does

Modern floor optimization builds a bid landscape: a statistical model of how each buyer bids for each inventory segment. If the model knows that buyers for US mobile traffic on your sports section bid $2.40 at the 75th percentile, it can place the floor just beneath the demand curve's sweet spot — capturing buyer willingness to pay without choking fill.

First-price auctions made this critical

In the second-price era, floors mattered less. In today's first-price world, buyers shade their bids — they bid less than their true value, expecting to pay what they bid. A well-placed floor pushes bids back up: it is the only counterweight publishers have against bid shading. This is why sophisticated floor strategies routinely produce 15–30% CPM uplift.

What an AI floor engine actually does

• Ingests auction logs, bids, wins and losses across GAM, AdX and Prebid.
• Trains per-segment pricing models, retrained every few hours.
• Pushes floors to Ad Manager as Unified Pricing Rules via API and to Prebid dynamically.
• Holds out a small slice of traffic as a control group to prove uplift honestly.
• Protects fill with guardrails so floors never overshoot demand.

That control-group discipline matters: any vendor can claim uplift; few measure it against a true holdout. Ask for the holdout methodology before you believe a number.