Billions of dollars in U.S. safety-net benefits go unclaimed each year.

These dollars fund programs that pull individuals from poverty, reduce food insecurity, and prevent emergency department (ED) visits, yet the people who need them most often cannot navigate the administrative red tape required to enroll. 

A non-English-speaking mother working multiple jobs and unable to afford daycare may not know she qualifies for WIC. A father of four receiving only $20 a month in SNAP benefits may be missing out on additional cash support. These are not fictionalized characters.

These are real people who face significant hardship. There is cause for optimism, however.

Closing this enrollment gap has attracted the attention of startups. Code for America has developed a benefits field guide to map online program access points nationwide, while Propel has developed a mobile banking app to check Electronic Benefits Transfer (EBT) card balances. ThriveLink centers its approach on AI social workers who help individuals navigate health, food, and prescription drug discount program enrollment.

Link Health’s Patient-Centered, Hybrid Enrollment Model

Link Health, a nonprofit operating within federally qualified health centers, takes on its own specific approach to this challenge, pairing AI and digital tools with human patient navigators. What distinguishes Link Health’s model is its commitment to a human-in-the-loop approach across all its enrollment initiatives.

What distinguishes Link Health’s model is its commitment to a human-in-the-loop approach across all its enrollment initiatives.

In the clinic, multilingual patient navigators conduct social needs screenings, complete applications with patients, and build trust necessary to navigate sensitive financial and citizenship information. 

Remote navigators connect with patients via text to answer outstanding questions and finalize applications. A four-question screening tool asking about insurance status and household composition flags which programs a patient is likely to qualify for, surfacing entitlements they may not have known to pursue.  

All navigators are certified through a Centers for Medicare & Medicaid Services navigation program, embedded within the Department of Health and Human Services. This program focuses on community health integration and social determinants of health assessments.

With the help of the AI for Impact Program, Link Health built a dashboard that streamlines the enrollment process and minimizes manual, error-prone data entry.

AI tools support Link Health in practical ways. Through the help of the AI for Impact Program at Northeastern University’s Burnes Center for Social Change, Link Health built an OCR-powered dashboard that extracts information from patients’ documents and auto-populates it across multiple applications, streamlining the enrollment process and minimizing error-prone, manual data entry.

Using a tablet, Link Health patient navigators access the organization’s online enrollment dashboard at a sign-up event in Boston.

A translation tool, Calliope AI, supports over 400 languages and dialects, enabling navigators to communicate with patients whose first language is not English, including approximately one-third of Link Health enrollees.

Meanwhile, a ChatBot named LEO allows patients to enroll independently by scanning a near-field communication panel or QR code in EDs and waiting rooms. LEO disables generative AI features to prevent misleading or excluding patients, while patient navigators regularly audit applications submitted with LEO for irregularities.

A Link Health patient navigator uses Calliope AI’s two-way conversation feature to communicate with a non-English-speaking patient. Image altered to preserve privacy.

Limitations

Benefit enrollment work is not without complications. Eligibility is complex.

While Link Health identifies patients likely to qualify, administering agencies may ultimately deny benefits due to nuances in household composition, prior receipt of the same benefit, or income calculations that differ from initial estimates. Patients in the clinic may feel rushed or concerned about privacy, even when experienced navigators attempt to address these concerns. 

Link Health’s current estimate of over $5 million in benefits returned to households assumes state median per-individual benefit receipt; actual disbursements vary, and some applicants will be denied benefits or receive more or less than projected.

There are broader limitations to acknowledge. Link Health does not yet compare its clients’ downstream outcomes, such as hospital visits, nutritional status, or ED visits, to control groups. Without these data, a causal tie between its benefit enrollment efforts and improved health remains an open question. 

An important evaluation would mirror studies like the Chelsea Eats initiative, where investigators found that a $400 cash benefit was associated with decreased ED visits and increased subspecialist appointments. Longitudinal investigations are needed to support sustained, positive health outcomes from benefit enrollment and inform effective resource allocation.

Policy Implications

Link Health’s experience demonstrates that AI technology may work best in benefit enrollment projects when tethered to trained navigators who deliver information at the user’s literacy level and in their native language. That framing may be a valuable tool for policymakers deciding where to invest next. 

AI technology may work best in benefit enrollment projects when tethered to trained navigators who deliver information at the user’s literacy level and in their native language.

This may include continued funding of patient navigator programs. Centers for Medicare & Medicaid Services targeted navigator programs for funding cuts in early 2025. Yet, Link Health’s work suggests that maintenance of humans trained to cut through bureaucratic complexity remains a key piece in realizing the potential of enrollment technologies. 

Consolidation of benefit programs into unified application portals remains a policy question. Some states offer single-application platforms spanning multiple programs; others require applicants to navigate separate websites with conflicting information. Link Health’s dashboard centralizes this process while ensuring patients receive accurate guidance from trained navigators. This model advocates for unified portals supplemented by human support. 

States should fund comparative research evaluating enrollment approaches, portal design, and navigator integration, and identify which iteration produces better outcomes for benefit uptake and utilization. 

States should fund comparative research evaluating enrollment approaches, portal design, and navigator integration, and identify which iteration produces better outcomes for benefit uptake and utilization. 

That evidence base is limited, and building it is a prerequisite to smarter public investment.