“In your first weeks of work you may have been trained to recognise insurance slips and documents that brokers might present to you as an underwriter,” said Mike Hobday, AntWorks’ senior vice president for Europe.
“You would be trained to understand the data you need to extract, to think about the standing of the broker and the nature of the risks. You might even make some calculations and underwrite the offered risk by entering data into the core systems of the business.
“As a very senior and experienced underwriter, there might still be a large part of your day spent doing the administration you were taught in those earlier days. So much data coming into the industry is unstructured, as much as 80 percent.
“The reality is that document classification and the extraction of information is as labour-intensive and fraught with the inefficiencies of errors in translation—often due to boredom, stress or poor training—as it was in the last century.
“Our solution is not about taking ‘the robot out of the human’, but taking the ‘junior underwriter out of the human’,” he explained.
AntWorks’ ANTstein SQUARE is an AI-based integrated automation platform (IAP), designed to recognise and understand structured as well as semi-unstructured types of data, such as PDFs, images, cursive handwriting and a range of document types. It offers a new way to automate complex processes, with a cognitive machine reading component that combines machine learning, natural language processing and natural language understanding.
“ALL THE RULES AND LEARNING ARE COMPLETELY DISCOVERABLE, ADDRESSING CONCERNS OFTEN HELD ABOUT AI AND RPA TECHNOLOGIES.”“We can help insurance companies to transition to digital enterprises—leveraging the huge opportunity to crack the data ingestion challenge,” said Hobday.
“AntWorks is not a first-generation robotic process automation (RPA) business—it is an ‘Automation 2.0’ business, able to emulate human resources when receiving data and categorising images and documents.
“ANTstein SQUARE extracts data from multi-page documents having enhanced its readability with its AI-infused processing capability that takes low resolution documents and enhances readability beyond the human eye,” Hobday explained.
“We can extract all the business-critical data necessary, enrich understanding from institutional knowledge from other data sources and execute business rules via application programme interfaces (APIs) or RPA.”
Proof of use
An example of this in practice comes from AntWorks’ work with one of the world’s leading brokers. AntWorks is helping it address a huge revenue opportunity. Insurance slips, once processed, have for years been filed in a document management system.
“They are unstructured documents and vary between carriers. If you wanted to do some analytics on the wealth of historic data on clients, their risks, assets and so on, you would need an army of knowledgeable colleagues to read and collate the data from many thousands of multi-page documents,” said Hobday.
“That’s the problem. As much as the business would like to access the data, to go back to these past clients with new insight and offers based on this deep institutional knowledge would not be cost-effective. That’s where ANTstein SQUARE, AntWorks’ integrated automation platform comes in.”
The client gave AntWorks 43 insurance slips from the aviation sector. In just five weeks, AntWorks had trained ANTstein SQUARE to extract 80 critical data points and create a file output.
“We could equally have input the data in structured form into analytics platforms through an API, or our own or your chosen RPA solution,” he said.
Accuracy on the 43 documents was running at 93 percent after only five weeks from the start of the project. The client then carried out a blind test with 200 documents—ANTstein SQUARE achieved 87 percent without machine learning starting to kick in.
“In the relatively small number where there were issues, a data point might have been unclear or not present,” said Hobday, “but crucially there was no need to refer to a human in the loop.
“What excites the client is that we can train our platform so quickly, and they can build internal capability alongside their data and RPA centre of excellence (CoE) to build future models,” he added.
“Most important is that all the rules and learning are completely discoverable, addressing concerns often held about AI and RPA technologies.”
According to Hobday, ANTstein SQUARE has the potential to transform many other processes within the re/insurance industry.
“This is not just about underwriting,” he said. “We are now working with insurance clients on claims processing and corporate life programmes.”