State and Local Agencies Deploy Artificial Intelligence for Document Processing

Powered by AI, on-premises IDP solutions, as well as cloud-based tools such as Google’s Document AI, help make sense of information. They can automatically classify documents and extract relevant data, a powerful capability that promises to help agencies cut through clutter.

From tax filings to permit applications and service requests, AI-driven IDP “can automatically extract data from all of this, validate the data and process it a lot quicker,” says Alan Shark, executive director of the Public Technology Institute. “The biggest complaint we always hear is that it takes government so long to do all of these things. This is going to speed it up.”

Automation Helps Agencies to Gain Time

In King County, Preyapongpisan leverages Amazon Textract in several pilot projects.

“We have incredibly skilled subject-matter experts,” she says. “Our objective through these pilots is to give them more time to apply their specific expertise in their domains, to help them save time on some of these rote tasks.”

“We’ve also built a prototype tool to help us categorize the inquiries and public comments we’re receiving via our Contact Us form on county web pages. We’re able to identify emerging topics and areas for service enhancement by evaluating customer service data,” Preyapongpisan says.

These initial efforts show considerable promise.

“We see tremendous potential for time savings, an incredible efficiency gain,” she says. “The pilot for the redaction service could scan and filter thousands of documents in moments. In that pilot, the AI pipeline reduced the time it takes to redact each application from 30 minutes to less than five seconds.”

This is still an emerging field, and with that in mind, “the human in the loop” is absolutely required when using AI in IDP, Preyapongpisan says. “We are not releasing these tools or moving any of these pilots into a production phase without a human reviewer there to validate everything.”

For those in state and local government looking to go a similar route, Preyapongpisan advises beginning by focusing internally.

“Our approach has been to focus first on employee productivity versus any kind of AI system that would provide an interface to the direct public,” she says. “We know these systems are not 100% reliable, and to mitigate some of the risk, we’re really focusing on internal productivity and not an external interface.”

Automation Speeds Access to Citizen Services

At Covered California, the state’s healthcare benefit exchange, IDP is helping residents get timely access to medical insurance.

“We pride ourselves on not being your typical state agency,” says CIO Kevin Cornish. “We spend an incredible amount of time, effort, energy and money building a brand that people trust. And it’s because we think access to healthcare is a basic human right. Everyone should have it, and it’s our job to make sure everyone in California does.”

In this context, “my personal job is to envision the technology and design , build and deliver the platform that supports that promise,” he says. To that end, Covered California has been leveraging Google’s Document AI tool.

“We intake a ton of documents from consumers: Prove your needs, tell us what your income was, tell us what your expenses are — all of these things,” he says. Based on those documents, the agency makes eligibility decisions. But manual processes slow that down, and consumers may wait days or weeks to know whether they have coverage.

Following a proof-of-concept effort, the agency implemented Document AI as a replacement for its legacy solution. With AI-informed automation for document processing, “we’ve completely turned around that experience for our consumers,” Cornish says.

Users can take a photo of a document and upload it — the same way mobile check deposit works in banking apps. “In four or five seconds, we’ll tell you that we understand what that document is,” he says. With IDP, “we’re able to extract the data we need out of it to put into our system.”

Consumers get decisions faster, and IDP also has taken a massive burden off the employees, who may field 30,000 to 40,000 calls a day during open enrollment season. In the past, many such calls involved humans validating documents, a time-consuming and tedious task. IDP eliminates the need for that, getting those representatives “back on the phone, talking to consumers who want to talk to a real person,” Cornish says.

“In our first year, we freed up about 10,000 people-hours of time, time that used to be spent manually validating consumer-uploaded documents,” he says. With IDP enablement, “those folks are now doing the job they really want, which is talking to consumers and helping them solve problems.”

Document Processing Creates a Digital Mail Room

The California Department of Motor Vehicles is using intelligent document processing tools from digital intelligence company ABBYY.

“We have this concept of a digital mail room, where people send information via paper mail, and then our automated machines open the paper and scan it. Then, with the AI-informed piece of IDP, the work is to classify, extract and then assess the information,” says Ajay Gupta, the DMV’s chief digital transformation officer.

SUBSCRIBE: Click the banner to sign up for the StateTech newsletter for weekly updates.

The IDP system classifies incoming documents. When people are asked to submit copies of their driver’s licenses, vehicle titles or birth certificates, for example, “AI is able to classify the documents into those categories: This is an identity document, this is an address verification document, this is a vehicle title,” he says.

“It’s able to read printed information that a customer has handwritten, or typed information that’s available on the form, to get the information out,” he says. From there, the AI can assess documents for completeness.

“For example, if it’s a car title and it doesn’t have odometer reading, we’re able to say that document is fine, but it’s not complete. Extracted information is then passed on by AI to other systems and robotics for further processing.”

The department has achieved 65% to 70% automation of document handling, and the success has had a profound operational impact. Customers are able to upload forms instead of mailing them in, “and they’re able to get near real-time responses to deficiencies in the documentation,” Gupta says.