Optimization

The DSP's optimization technology gives you different approaches to maximizing tactic performance:

  • Algorithmic optimization adjusts the CPM bid price and status based on tactic goals that you set.
  • Machine learning optimization leverages machine learning and a unique algorithm to calculate the best price per impression based on the overall probability of achieving the tactic goals that you set.

These two types of optimization cannot be used at the same time.

Algorithmic Optimization

Algorithmic optimization automatically adjusts the bid price for your tactic's placements to an optimal value to achieve a KPI goal. This lets you recognize which inventory is performing better for your tactic objectives and concentrate spend on it, while investing less in inventory that isn't performing as well.

For information how to set up algorithmic optimization for your tactic, see Enabling Algorithmic Optimization.

When algorithmic optimization is enabled, the tactic bids on impressions, just as it would without optimization, and evaluates which domains and placements are helping the tactic meet the goal. After a learning period, the tactic creates rules that adjust bid prices depending on how a particular domain and placement is performing. Optimization rules can stop bidding altogether on under-performing domains and placements.

If your goal type is eCPC, eCPA, or eCPVC, you control the learning period by setting a learn budget, a dollar amount set aside for open bidding before any rules are created.

If your goal type is CTR or VCR, you control the learning period by setting an impression threshold, the number of impressions that will be won and analyzed before creating rules.

Machine Learning Optimization

Machine learning optimization leverages machine learning and a unique algorithm to maximize a tactic's performance. See Enabling Machine Learning Optimization for more information.

Machine learning operates in two distinct modes:

  • Learning mode: In this stage, the technology captures information to create a model. In learning mode, the tactic uses the tactic's Default Bid values to bid on all impressions. A model is created after a set number of actions have been acquired by the brand: 150 clicks if the goal is eCPC, 150 viewable impressions if the goal is viewable CPM, 40 conversions if the goal is CPA, and so on.

    Bid multipliers are taken into effect during learning mode.

  • Optimized mode: In this stage, a model is available and is used to decide if the DSP should bid on an impression or not, as well as how much to bid.

    In optimized mode, bid multipliers are ignored.

You can see a tactic's optimization mode in its line item details in DSP Optimize view:

In optimized mode, the campaign uses the KPI Goal value and the Maximum Bid to calculate the price to bid for the impression based on the overall historical probability of achieving the event that affects the KPI (e.g. a click, a view, an acquisition). Bid prices are unique for every impression and depend on many factors, including the inventory available, the timing of the impression, and the device that will display the impression.

A minimum spend of $5,000, per brand, over the course of 2 weeks is recommended.

Note: The optimization technology constantly refines its model to try to achieve your desired KPI. However, you should still check your tactics regularly to be sure that they achieve the results you're looking for.