BidTheatre DSP can automatically adjust bid levels to optimize against a given campaign KPI. The optimization models are continually updated using Machine Learning algorithms fueled by previous buying activity.
How does BidTheatre ML work?
Machine learning models are used to predict the effect of bid opportunities on some common campaign KPI's. The models are trained from known effect of previously bought ad impressions. The machine learning algorithms used to train the models consider attributes such as when (weekday, time of day), where (site, category, ...), what (ad size, category, ...) and who (device type, previously clicked ads, ...)
Machine Learning (ML) is a feature that is free to use and activated on bid strategy level. Once activated, the system will use ML models to score how well a given bid opportunity will fit the campaign KPI. The score between 0 and 1 will be multiplied with the set max CPM bid, to get the final bid.
Advertisers with more than 1 million impressions per week will get models trained specifically for their own impressions.
How do I activate ML on my campaign?
Use a Filter with an ML option selected.
Bid Optimization - Don't Sacrifice Pace: Will optimize without sacrificing campaign delivery goals
Bid Optimization - Strict: Will optimize fully even if it risks sacrificing campaign delivery goals.
As always, Filters can be used on specific bid strategies, or as campaign wide default.
In addition to using a Filter with activated ML, the campaign will need to have a set KPI. The set KPI will rule which machine learning model is used to score bid opportunities.
|Cost Per Click||Adjusts bid proportional to likelihood of impression being clicked|
|Cost Per Viewable Impression||Adjusts bid proportional to likelihood of impression being viewable|
|Click Through Rate||Adjusts bid proportional to likelihood of impression being clicked. Only bids on high likelihoods.|
|Quality Click Through Rate||Adjusts bid proportional to likelihood of impression being clicked and user browsing the landing page. Only bids on high likelihoods.|
Things to think about when using Bid Optimization
Using bid optimization will on average lower you eCPM with 30%. This means that you can and should set a higher max CPM, to enable buying the most valuable impressions.
If you're targeting deals, keep in mind that the ML optimization may place many bids under the deal floor.
Also, keep in mind that the optimization rarely excludes poor performance completely, but rather places low bids.
When using Click Through Rate or Quality Click Through Rate KPI's, use Strict optimization for best performance.
Each bid strategy can have an assigned optimization strategy that governs how bids are optimized. To set the optimization strategy, click the ID of an existing bid strategy. Optimization evaluates the performance of a specific ad on a specific RTB Site, and adjusts individual ad/site bids according to the optimization strategy selected.
A few standard optimization strategies are provided for use by BidTheatre. You can also create your own optimization strategies in Assets / Optimization Strategies.
Currently optimization can be made on two functions
|Min CTR||CTR (0-1)||Optimize on this minimum click-through-rate CTR||Suitable for optimization of branding campaigns|
|Max eCPC||eCPC||Optimize towards this maxmium effective cost per click||Suitable for optimization of performance campaigns|
|Min Inscreen Rate||Inscreen Rate (0 -1)||Optimize towards a minimum inscreen rate||Suitable both for branding and performance campaigns|
Optimization Type - Initial Learning
Currently there exists one type of optimization strategies - Initial Learning.
In Initial Learning, a fixed part of the the budget is devoted to a learning phase where Optimization Type Value number of impressions will be bought from each site. After the learning phase, sites that performed worse than the Optimization Function Value of the the Optimization Function will be placed on the campaign exclusion list.
Pros: No risk of underdelivery unless target function value is set too tight. Conserves best performers onto a new sitelist and bid strategy.
Cons: No optimization is performed after the learning phase.