AI improvements to Medicaid must account for needs of eligibility workers, experts say

Artificial intelligence is increasingly being leveraged as a tool to combat snafus in states’ Medicaid programs, but officials must ensure such solutions are human- and user-centered in order to efficiently tackle common obstacles in their systems, experts say. 

The delivery of Medicaid services and programs can rely on a whole cohort of stakeholders, like state agencies, civic technologists, vendors and others. But when designing AI solutions to improve Medicaid, leaders must factor in the “often more obscured, but no less critical stakeholder,” which is the eligibility worker who will be using such services, said Daniel Mintz, director of safety net policy at Code for America, during a webinar hosted by the nonprofit Wednesday. 

Eligibility workers are the “front line of state Medicaid programs … who sit in the chairs between these sometimes quite heavily disconnected components of a Medicaid eligibility system,” he said. “AI solutions that are coming to streamline work, to help make Medicaid more accessible [and] to improve decision making … need to work concretely to meet the actual everyday needs of the workers who are at the intersection of these various components of the system.”

These staff are critical for helping residents apply for and access benefits, fielding their questions and assisting them though the enrollment process, but their work is often disrupted by clunky or disparate systems. 

As an example, Mintz pointed to the common challenge of duplicative or manual data entry that eligibility workers must complete in their daily work and that reduces their time to thoughtfully analyze the data or make meaningful connections with their clients. 

This challenge presents an opportunity where an AI-based solution could be applied, said Jennifer Thom, senior director of data science at Code for America. 

Agencies could deploy an AI-enabled data extraction tool that helps reduce redundant manual work, Thom said. Data extraction can be leveraged to, for example, transform unstructured and disparate data across beneficiaries’ forms, documents and other content into a standardized format, she explained. 

This kind of automation can also turn paper-based information into machine readable data to unlock more innovative processes for eligible workers, Thom said. Data extraction can, for example, enable functions like the pre-population of online forms to expedite the Medicaid enrollment and verification process for eligibility workers.

“The job would shift from typing to verifying and being able to check any field to see a source document,” Thom said. 

Another common challenge frontline workers face is navigating complex or siloed state data systems when working with clients to determine their household’s eligibility for Medicaid benefits, Thom said. 

State systems sometimes record the same data on a person in different ways, like documenting their phone number in different formats. Such data entries can be cumbersome for eligibility workers to track down if information is inconsistent or incorrectly recorded across systems, further complicating the user and beneficiary experience of Medicaid benefits, Thom said. 

Entity resolution is one technique agencies can explore as an AI solution, she said. It entails “the process of figuring out the different records that all link up to one person … Like a detective, it compares clues like names, addresses, phone numbers and dates of birth to merge all of these scattered pieces of information into one single, trusted profile,” Thom explained. 

This technique aims to prevent eligibility workers from missing critical information or documentation while assisting clients and therefore improves the timeliness and accuracy of service delivery, she said. 

More broadly, eligibility workers often struggle to translate Medicaid program rules and policies to clients, Mintz said. While workers must stay abreast of hoards of policy information, they must also absorb client data during the intake and application process, contributing to “information overload” for many staff, he explained. 

To streamline the documentation and sharing of Medicaid data, generative AI has emerged as a useful tool for summarizing elaborate documents and notes, Mintz said. 

“Generative AI can give workers and state systems as a whole a leg up in starting to move from program-specific jargon to more plain language,” he explained. 

A large language model can, for example, be leveraged to improve training materials for eligibility workers or generate more digestible resources for enrollees to help reduce staff and resident confusion about the Medicaid enrollment process. 

Ultimately, Mintz said, the goal of state agencies considering AI-based solutions to their Medicaid programs should be “to not only reduce the time that it takes for caseworkers to process the case, but … also to improve the experience of Medicaid enrollees and applicants of interacting with government often at times of critical need.”