What is MRE?
Andrew Eifler, VP Product at AppNexus points out in his blog post discussing Market Rate Estimation that "bidding a fixed price throughout the entire day actually underpays for inventory when supply is low (at night) and overpays for inventory during the day (when supply is high)." In order to solve this problem, he suggests that advertisers need to find a way to determine the most efficient bid price point to win in the RTB market. One solution is to "guess and check" the bid price until it matches the equilibrium point between supply and demand. However, this approach is manual and can require spending more budget than needed during the "guess and check" phase. We need a more automated way to solve this problem.
MRE (Market Rate Estimation) is an AppNexus-derived prediction of bid price for a given piece of inventory at a given time. MRE breaks down online ad impressions into buckets comprised of unique combinations of Tag, Hour, Geo, Size, Day/ Hour of Week, and makes predictions by looking back at historical prices plus a component for adjusting the bid price up or down based on the data collected from the current day. AppNexus MRE produces two values: Estimated Average Price (EAP), an estimate of a bid that is likely to win about half of impressions, and Estimated Clearing Price (ECP), an estimate of a bid that is likely to win approximately 80% of the impressions from a given publisher. Unlike the manual "guess and check" method, this solution automatically adjusts the bid price depending on the changing variables in each auction and provides a calculated bid price.
How do we use MRE (EAP or ECP) in Bonsai?
APB (AppNexus Programmable Platform) uses a domain specific programming language called Bonsai (ref: Introduction to Bonsai Decision Tree Language), which allows advertisers to model decisions in a tree-like structure. Each tree has multiple leaves where the final bid price for the auction is determined. There are two types of leaves: regular leaves and smart leaves. While regular leaves contain a simple numeric value, Smart Leaves allow advertisers dynamically compute the bid price, and thus, lets them control more aspects of the ad delivery process. Advertisers can simply add EAP or ECP in the computation to determine the bid price in each leaf. Let's compare using Bonsai without MRE to using Bonsai and MRE together.
Tree with Regular Leaves
Using regular leaves, advertisers can value impressions at a certain dollar value (CPM) depending on various scenarios. Below example describes a simple campaign which values US impressions at $5 CPM, otherwise bid $1 CPM.
if country = "US": 5 else: 1
As noted above, advertisers have to "guess and check" until arriving at the most ideal price they have confidence in for a single tree. When a new tree is built, the same process needs to repeat. Unless advertisers already have high confidence in their bid values, this method can turn out to be an inefficient strategy potentially spending more advertising budget than needed during the manual learn phase.
Tree with Smart Leaves
Using smart leaves, advertisers can value impressions dynamically by applying EAP or ECP modified in the compute method.
Here is the syntax for compute:
compute(input_field, multiplier, offset, min_value, max_value)
The example below describes a tree that uses ECP as the input field in an effort to win 80% of impressions matching the criteria (in this case, country = United States.) There is no offset, and a minimum price of $1 CPM and a max of $10 CPM.
if country = "US": value: compute(estimated_clearing_price, 1, 0, 1, 10) else: value: no_bid
If advertisers are testing new trees and are unsure how much each impression is worth, EAP or ECP calculated by MRE can help them meet the supply and demand curve.
Please see Smart Leaves for the complete documentation
Steve Kwak is a Technical Solutions Consultant (TSC), member of Technical Services team under the AppNexus Advertiser Technology Group. His previous work experience includes digital strategy consulting and web application development.