Dynamically optimize the customer experience with real-time predictions
In Treasure Boxes for Next Best Action (NBA), we have demonstrated an end-to-end templatized solution for effectively identifying Next Best Product (What?), Channel (Where?), and Time (When?):
The NBA solution is based on probabilistic segments derived from our out-of-the-box predictive analytics functionality, Predictive Scoring:
Here, since the predictive models are completely white-box, it is possible to run the NBA models in a real-time system to dynamically optimize web contents as follows:
- Train predictive models on the customer data stored in Treasure Data.
- In addition to Predictive Scoring UI, you could also use Hivemall and Python Custom Scripts for advanced configurations.
- Export the models and store them into the following external storages:
- Treasure Data's Active Data Layer is a custom API endpoint for real-time data I/O.
- If you have your own databases or streaming systems, you could also use them as Treasure Boxes for Click-Through Rate Prediction demonstrates MySQL-based dynamic predictions in an ad-server use case.
- When an application observed an event, the app fetches the models from the storages and make predictions to find out the best scored action for delivering an optimal experience in a timely manner.
- We could have multiple models and compare the predictive scores to rank the potential actions.
Use Cases
- Audience Segmentation
- Campaign Orchestration
- Dynamic Marketing Budget Allocation and Targeting
- KPI (ROAS/ ROMS) Optimization