Intel IT Advanced Analytics developed an AI system that mines millions of public business web pages and extracts an actionable segmentation for both current and potential customers.
Intel is no different than other large enterprises when it comes to the need to identify new customer opportunities across many geographies and languages. However, it has its own specific customer segmentations, based on domains and modes of operation. Additionally, in the era of globalized business, existing customers are often expanding into new domains, requiring sales and marketing staff to constantly keep current with changes in a wide variety of industries. That’s why our team within Intel IT Advanced Analytics developed an AI system that mines millions of public business web pages and extracts an actionable segmentation for both current and potential customers.
The system focuses on two key classification aspects. The first is industry segment, ranging from broad verticals such as “healthcare,” to more specific fields such as “video analytics.” The second is functional roles, like “manufacturer” or “retailer”, that further distinguish potential sales and marketing opportunities.
Our customer segmentation system is comprised of two major components. The first component is tasked with large-scale data acquisition from web pages and other public sources. It then consolidates that information with internal corporate data in a custom knowledge graph (Fig. 1). The system acquires a constant streaming of textual data from millions of sites, updating the multi-million node knowledge graph with gigabytes of data every hour.
Figure 1: Illustration of Customer Knowledge Graph.
The second component is a suite of machine learning and natural language processing (NLP) models for segmenting potential customers. Web pages are fed into a multi-label convolutional neural network (CNN) text classification model that was developed by Yoon Kim. We further boost it by utilizing a pre-trained multi-lingual BERT language model developed by a team at Google to help scale across languages and classes with scarce training data. The data we use to train the model is enriched by crawling tens of thousands of company sites with labeled industry information found on Wikipedia. For companies without labels, we take advantage of the vast labeled Wikipedia corpus by employing semi-supervised learning.
Sales and Marketing Results
Intel sales and marketing staff have traditionally used manual search and vendor tools in order to identify potential leads, however, these methods lack the ability to align with the internal language used by Intel staff to properly segment and tailor their outreach plans. With our new AI customer segmentation system, which started as part of a proof-of-concept experiment, we were able to discover new leads in specific industries much faster and more accurately than using traditional methods.
Our customer segmentation system is only one of thousands of AI applications that will improve enterprises in the coming years. As Intel continues to expand its portfolio for AI workloads, including 2nd Generation Intel® Xeon® Scalable processors with built-in AI technologies, optimized frameworks and software libraries, we will see businesses find new and exciting ways to harness cutting-edge technology to move, store, and process data wherever it is best suited.
To see further details about our AI customer segmentation system, read the longer paper “Learning a faceted customer segmentation for discovering new business opportunities at Intel.” If you would like to learn more about how your enterprise can excel with AI, read our practical guide on the subject. For the latest updates from the Intel AI team, follow us on Twitter.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
Mary is the Community Manager for this site. She likes to bike, and do college and career coaching for high school students in her spare time.