Intelligence Driven User Communications at Scale with Gaurav Agrawal and Mani Parkhe
Intelligence driven automation of user communications is aimed at determining what kind of communications should be sent to whom, when, and how. In addition to better user engagement and cost analysis, this framework unlocks several critical use cases for Uber. These include user targeting for marketing and other campaigns, appropriate content and channel routing, offensive content discovery, delaying or blocking marketing communications in events of disasters, relevance sorting of communications based on user interests, etc. We will talk about how we are using data science analysis to understand our communications data and build classification models to categorize our content at scale and using those predictions for segmentation, automation, and tuning communication delivery. We are also exploring deep learning techniques to achieve this objective. This talk will include some of the architecture details around model training, deployment and predictions as well as longer term plans on how we envision this platform.
Recommended
More Related Content
Similar to Intelligence Driven User Communications at Scale with Gaurav Agrawal and Mani Parkhe
Similar to Intelligence Driven User Communications at Scale with Gaurav Agrawal and Mani Parkhe (20)
More from Databricks
More from Databricks (20)
Recently uploaded
Recently uploaded (20)
Intelligence Driven User Communications at Scale with Gaurav Agrawal and Mani Parkhe
- 1. INTELLIGENCE-DRIVEN USER COMMUNICATIONS AT SCALE October 3, 2018 Gaurav Agrawal, Mani Parkhe, Manisha Mundhe
- 2. Team Gaurav Agarwal Engineer, Uber Prior: Microsoft Mani Parkhe Engineer, Databricks Prior: Uber, LinkedIn, CAD Manisha Mundhe Engineering Manager, Uber Prior: eBay, PayPal
- 3. 01 Mission 02 Communication Terminology 03 Vision for Intelligent Communications 04 Architecture 05 Scale 06 Use Cases 07 Q&A
- 4. Drive billions of individual interactions intelligently and efficiently, across all communication channels that customers use, and adapt in real time to their behaviors. WHY | MISSION
- 6. Tools Authorization and Personalization of Content Create IFTTT campaigns Delivery and Engagement stats
- 8. Suggested channels: ● Based on metadata ● Engagement metrics ● Cost ROI ● ... Campaign planning: ● Previous campaigns ● Engagements ● Experimentation ● Schedules Personalization: ● Channels/Medium ● Formatting ● ... In-flight rerouting: ● Channel Affinity ● Real-time feedback ● Marketplaces ● ... Vision: AI-driven Marketing and CRM campaigns High level objective: (human) ● Type of campaign ● Objective function ● Target audience ● Budget
- 9. Tools API Clients Gateway Comms Data Flow Intelligence Recommendations: Channel, Content, Target, Schedule Suppression: Unsubscribes, Preferences Comms Providers Delivery Channel specific delivery Post Delivery Delivery Metrics, Data Aggregation Pre Delivery Author, Schedule, Target DataLayer Feature store Content stores Raw Data storage
- 10. 3.5BMessages Sent Daily 500M Events Processed Daily Communication Scale ● 500M messages sent and 3.5B message related events processed through the system ● Transactional vs. Marketing SLAs ● Global: Supports 168 locales Scale
- 11. Suppression Framework City Supp. User Supp. Hard Bounce Control Groups Freq. Rate Limit Personalization … Filters: Suppression Framework Blocking communications in-flight ... SYNC: should send? Gateway holdout list … to Providers Batch processing Uber services Uber services Users C* Experiments ASYNC handles async actions
- 12. ● Estimate most likely city for a user ● Disperse features to Cassandra via Spark app periodically ● Suppressor framework stops marketing messages in real-time Example: City-based message suppression Scheduled Spark JobHive Cassandra Suppressor shouldSend() User Features Everyday Feature Dispersal
- 13. Thank you!
- 14. Proprietary and confidential © 2018 Uber Technologies, Inc. All rights reserved. No part of this document may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval systems, without permission in writing from Uber. This document is intended only for the use of the individual or entity to whom it is addressed and contains information that is privileged, confidential or otherwise exempt from disclosure under applicable law. All recipients of this document are notified that the information contained herein includes proprietary and confidential information of Uber, and recipient may not make use of, disseminate, or in any way disclose this document or any of the enclosed information to any person other than employees of addressee to the extent necessary for consultations with authorized personnel of Uber.