Intelligent document processing drives faster, smarter government

This interview of Maximus digital government solutions expert Michael Bostian by Nina Jacobs was featured in Government Technology Insider.

Many interactions between state government programs and the people they serve now occur over digital channels, but physical and scanned documents remain a critical part of program workflows. And whether information comes in through a mailroom or an online portal, the data needs to end up in the same place, in the same format, to facilitate a seamless process.

AI-powered Intelligent Document Processing (IDP) is transforming the way state program data is handled, bringing speed and precision to agency workflows and bridging the gap between physical and digital information.

AI streamlines data entry for faster service delivery

Data entry has undergone several transformations over the last 50 years, evolving from a strictly manual process to digitization via photocopiers or optical character recognition (OCR).

OCR is well-suited to extract clean, structured, typewritten data from a document, but its ability to manage multiple formats and structures simultaneously is limited. AI-powered IDP can work with different types of forms, various data formats, and even unstructured data without having to train on every possible variation. 

Michael Bostian, Associate Managing Director of Digital Government Solutions at Maximus, gives the example of a driver’s license. “Legacy processing solutions, even early IDP solutions, can extract information from a driver’s license, but how many variations of a driver’s license are there? Even within a single state, there may be several variations, and the automation engine needs to recognize all those formats,” he said. “We’re at a point that AI-based IDP tools don’t require all that upfront training.”

Considerations for IDP integration

The myriad document types that government agencies handle require different technologies to process them. Unstructured or handwritten data may require more sophisticated processing with a large language model, while basic OCR may be sufficient for a standardized form. Agencies can work more efficiently and cost-effectively by maintaining the processes that already work, then using AI only when and where it delivers additional value.

Having the flexibility to apply the right technology to the right business process gives states the ability to onboard new document types without needing to train every tool on each new template. Rigid infrastructure can inhibit states’ ability to adapt as the document types they receive change. This is especially important as the approach to document processing evolves, with the next wave combining the powers of agentic AI with IDP.

Driving efficiency with AI without sacrificing quality

As states introduce more sophisticated AI tools, they must establish ground rules for where, when, and how AI can be used effectively. Human oversight of AI activities is critical to maintaining quality, and Bostian emphasizes that it should also be applied appropriately. “If a human has to double-check everything produced by AI, the time and effort saved will be minimal,” he said. 

Deciding where and how to keep humans in the loop will look different for every business process. To start, Bostian recommends considering confidence thresholds for straightforward, unambiguous activities and human verification for analyses that influence decision-making. Through experimentation, states can arrive at a distribution of responsibilities that puts AI’s information processing capabilities to good use while letting humans do what they do best: reasoning, logic, and deduction.

Delivering value with IDP

By integrating with and building on existing infrastructure and workflows, AI-powered IDP enables states to address challenges across multiple agencies and departments with a single solution. “When a state introduces a new technology into its workflows, what they’re looking for is value,” says Bostian. “Any time a state can take one robust solution and make it work for many agencies and programs, they’re reducing costs, complexity, and technical debt.”