Building an AI platform: Key principles and lessons learned
Moty Fania details Intels IT experience of implementing a sales AI platform. This platform is based on streaming, microservices architecture with a message bus backbone. It was designed for real-time data extraction and reasoning and handles the processing of millions of website pages and is capable of sifting through millions of tweets per day.
|Talk Title||Building an AI platform: Key principles and lessons learned|
|Speakers||Moty Fania (Intel)|
|Conference||Strata Data Conference|
|Conf Tag||Make Data Work|
|Location||New York, New York|
|Date||September 24-26, 2019|
In today’s sales and marketing landscape, knowing your customer is everything. Traditionally, this would be achieved by dedicated sales agents covering accounts they know very well. However, account coverage complexity grows as agents can’t transform the vast quantities of data about their accounts into trusted insights in a timely manner. Moreover, there will never be enough salespeople to cover hundreds of thousands of accounts. Moty Fania details the experience of the advanced analytics team at Intel IT as it implemented an internal sales AI platform to support the full cycle of sales. The sales AI platform continuously extracts and interprets massive amounts of internal and external public data and applies AI reasoning for taking the relevant actions. By imitating humans’ reasoning capabilities and decisions, AI technology helps by processing extensive, disparate data sources and converting them into actions or actionable insights for salespeople. This may allow effective coverage of a much larger number of accounts and gradually provides autonomous coverage by automating end-to-end sales services and actions. To enable all of this at scale, the platform is based on streaming, microservices architecture with a message bus backbone. It employs cutting-edge open source technologies such as Ray, Snorkel, TensorFlow, TensorFlow Serving, Python, Kafka Streams and was optimized to be easily deployed with Docker and Kubernetes. The platform supports different kinds of data and knowledge representations including knowledge graph, search, and more. In addition, it enables online deep learning inference at scale for natural language understanding and recommender engines. If you’re planning to implement a similar AI platform, you’ll learn from Intel’s experience, including how it identified the set of characteristics and needs that were required for sales AI scenarios and made them available in this platform, a thorough overview of the architecture Intel implemented with the related technologies, and how Intel uses this platform to address sales AI use cases that support end-to-end sales services to accelerate sales. The platform and related advanced analytic capabilities have increased Intel’s revenue by approximately USD 500 million in the past five years.