Advanced Computing in the Age of AI | Thursday, June 8, 2023

‘Natural Language Understanding’ Appliance Targets the Email Deluge 

“Eighty percent of customer service should go to the other 20 percent.”     - Yogi Berra

Like many things he said, Yogi never said that. But the statement gets at a serious point about the 80/20 rule in customer service: four out of five customer phone calls, emails and chat requests are routine, repetitive and increasingly within the skills reach of intelligent machines.

The goal of AI in customer service is to automatically respond to the high percentage of relatively simple communications (“How do I reset my password?”) that pull customer support away from the lesser number of complex requests that require specialized knowledge.

Because of the Coronavirus pandemic, automated customer service is more of an imperative. Customer support centers are closed but customers still need support – in some service sectors, more support than usual. today launched a “message intelligence” capability that utilizes natural language understanding (NLU), a subset of natural language processing, designed to help companies handle the deluge of daily enterprise email.

The product, called Message Intelligence, processes and routs messages in real time using NLU to analyze the semantic content—the meaning and intent of messages – at massive scale. The product is delivered as an appliance incorporating a Xilinx FPGA-based Alveo U50 Data Center accelerator card and an AMD 2nd generation Epyc processor integrated by UK-based Boston Limited in a Supermicro 1U rackmount server.

In’s benchmark tests on the Enron email dataset (literally, 500,000 emails from employees of defunct Enron Corporation obtained by the Federal Energy Regulatory Commission in its investigation of Enron's collapse nearly 20 years ago), the company claimed the appliance registers between 1,000 and 10,000 times faster than BERT, the Natural Language Processing technique developed by Google, while achieving the same level of accuracy. said the product’s Semantic Folding method enables users to create custom semantic classifiers and filters “and can be quickly trained without supervision in the specialized vocabulary of any business domain and can function across multiple languages.” It also provides a dashboard to monitor the performance of the appliance.

“We are starting with the most crucial means of enterprise communications, email.” said COO Thomas Reinemer. “But ultimately our solution can be applied to a broad range of messages, including instant messages, tweets and blog posts. Organizations can start by applying it to more efficiently deal with emails, but the value in this technology extends to streamlining and automating business processes, increasing efficiency, responsiveness and productivity in all areas of the organization – thereby reducing costs, improving bottom lines and, most importantly, improving customer satisfaction.”

The company said an annual introductory subscription is priced at $48,000.