How AI-Driven Information Extraction Can Transform 3 Huge Industries

Published On November 13, 2019

Banking, insurance, and oil and gas exploration are three different industries, yet they share common traits that make them valuable candidates for information extraction using artificial intelligence (AI). Each industry has years-worth of documents that are locked in paper or other analog formats. New technologies like intelligent document recognition, natural language processing (NLP) and machine learning can unlock this value, according to a new report by Emerj Artificial Intelligence Research[i]. Emerj takes a deep dive into each industry to show how.

Banking

Because of the long-term nature of their many loan instruments, banks have years of records on customers and assets. Due to factors like industry consolidation, changing reporting requirements and the types of data gathered, finding and aggregating this information is laborious and expensive. But it is increasingly necessary. New regulations require banks to account for all the information they store about customers — data that may go back as far as 40 years.

Information extraction using optical character recognition (OCR) technology combined with AI can digitize records stored on paper and microfiche to extract useful information. A combination of natural language processing (NLP) and search engine technology then enables workers to query these unlocked data stores via speech or conversational language.

Machine learning algorithms can interpret and make decisions about data that previously required human intervention. For example, machine learning can deliver search results for a document’s contents that are similar to the string of search text without matching it precisely. Software can also scour unstructured sources such as text documents and emails to identify information like addresses or Social Security numbers without requiring that data to be labeled explicitly.

AI can be used to find information across multiple data silos, which is a useful feature in the heavily regulated banking industry. It can even compensate for errors like misspellings or transposition mistakes to retrieve information that is likely to match the user’s request.

Insurance

Like banking, the insurance industry has huge repositories of paper that can benefit from the information extraction features, as well as some applications that are unique to insurance.

For example, underwriting is a complex discipline in which decisions about coverage and premiums are made based upon demographics, historical patterns, claims history and other factors. Machine learning algorithms can identify patterns in digitized documents at a scale humans can’t approach. Intelligent optical character recognition can even decode information and documents written by hand.

These capabilities are also useful in customer acquisition by enabling insurers to better assess applicants’ risk to make more informed on-boarding decisions.

Claims processing is another area in which judgment calls can create inefficiencies due to “leakage,” or process inefficiencies that result in overpayments. Image recognition, which is a sub-discipline of machine learning, can analyze millions of photos taken by claims adjusters and classify them by the type and severity of damage. This enables adjusters to quickly call up past claims through image matching, and search claims histories to narrow down actual repair costs.

Oil and Gas

Energy exploration is a capital-intensive business with many assets housed in remote locations. Predictive maintenance is a form of data analytics that compares past performance metrics to live sensor data to predict when equipment might fail so that repairs can be made with minimal downtime. This not only reduces outages, but prevents breakdown-related injuries, yielding savings that can run into the millions of dollars.

AI can also deliver dramatic returns when exploring for new energy sources. Teams of geo-scientists have historically sifted through paper documents and images by hand, looking for seismic records, yield histories and geological maps to find promising new drilling locations. Digitizing this data for natural language query enables faster access and quicker decisions. Scientists can even point to a spot on a map and get a full accounting of everything that’s known about that location.

Machine learning can also find new drilling sites based on predictive analytics performed at scale. It can fold in data from public sources and even use machine vision to identify locations similar to those produced in the past.

[i] Source: Unlocking the Business Value of Unstructured Data with AI/ML, 451 Research and Iron Mountain, 2019, All Rights Reserved.