A learning model is built on a combination of advertiser, publisher and user data. The learning model can be applied to all advertisers in an advertising system. The learning model provides predicted conversion rates for a given advertisement (“ad”) appearing on different publisher networks. A predicted conversion rate represents the probability that a click on a given ad appearing on a given publisher will lead to a conversion. The predicted conversion rates are used to generate a multiplier. The multiplier is used to automatically adjust the advertiser's bid (e.g., maximum cost-per-click (CPC)) for the given ad prior to an auction for the ad. Adjusting the advertiser's bid equalizes a cost-per-conversion among the publishers for the ad.