Delivery Optimization
Delivery optimization is the process of using machine learning (ML) to decide whether to send a request to a DSP. Instead of sending every possible request, the system analyzes historical bid response data and learns patterns that predict the likelihood of a successful bid.
By filtering out low-value requests and prioritizing high-value ones, delivery optimization helps you:
- Increase bid rate
- Reduce unnecessary traffic
- Improve campaign efficiency
To start using delivery optimization activate the Smart Supplier module first.
Activating ML
You can activate ML for each DSP per datacenter (EU, US, Asia). To have ML enabled for a DSP, please contact your account manager. They will activate the feature and ensure the DSP is configured for data collection and model training.
Once ML is enabled, the system automatically progresses through the lifecycle: Collecting Data → Feature Selection → Model Training → Predict ON.
- Collecting Data – gathering enough traffic to start.
- Feature Selection – preparing the training set.
- Model Training – model is being built.
- Predict ON – predictions are live and requests are optimized.
- Predict OFF – ML is disabled or has not been started.
You can view the machine learning status for each DSP in the DSPs screen:
- Go to Demand → DSPs.
- Look at the ML Status columns for the datacenters (EU, US, Asia).
Each status is represented by an icon in the EU / US / Asia datacenter columns, with tooltips available on hover for quick identification.
When the ML state changes for a specific DSP (for example, from Collecting Data to Model Training, or from Model Training to Predict ON), the system automatically generates a notification. These notifications are sent to the assigned BizDev and AdOps users responsible for that DSP.