Food Access in Rural Michigan

This project began with a broad question from our partner, the NTCA Rural Broadband Association: How can AI be deployed in rural spaces?

Working collaboratively as a team of five, we narrowed our focus to food insecurity among older adults in rural Michigan, a population facing barriers of limited transportation, mobility challenges, and fragmented access to food assistance programs.

Through 8 stakeholder interviews and external data analysis, we uncovered that the real challenge wasn't a lack of resources but a breakdown in communication between organizations. Our solution, an AI-powered resource platform paired with broadband partnership incentives, empowers food assistance networks to collaborate and scale their reach.

Nearly one-sixth of food-insecure households in the U.S. are in rural areas. In Michigan, 19.6% of the population is 65 or older, and 10.8% of seniors live in poverty. Food insecurity correlates strongly with chronic diseases like diabetes and hypertension, which disproportionately affect older adults in high-poverty counties.

While Michigan has robust food assistance programs, many organizations operate in silos with limited funding and minimal staff capacity. Communication gaps prevent resources from reaching those who need them most.

Our Research Question

How might we leverage AI to improve organizational efficiency of food assistance organizations in rural Michigan so that services and programs can best serve older adults in need?

This question evolved through research. We initially considered individual-facing AI tools but pivoted after interviews revealed that organizational coordination was the real bottleneck.

Literature Review

We analyzed 35 sources covering Michigan demographics, food insecurity data, existing assistance programs, and AI applications in food systems. Key findings included the correlation between poverty and diabetes rates across counties and the fragmented nature of food distribution networks.

External Data Analysis

Our team analyzed datasets from Data.gov, the U.S. Census, and USDA, including Income-to-Poverty Ratios and Senior Project FRESH program evaluations. Approximately 13% of surveyed individuals faced food insecurity, and 9% were unaware of available food resources.

Food security status and income-to-poverty ratios across Michigan, revealing the scope of need among older adults.

Stakeholder Interviews

We conducted 8 interviews with food distribution managers, community educators, a health equity researcher, an AI expert in agriculture, an aging services representative, and a farmer. Interviews lasted 30 minutes to 2 hours and were transcribed, then analyzed using affinity mapping in Figma.

We organized 455 affinity notes from our interviews into meta-clusters, identifying three core themes in our research.

Our partner, Joshua Seidemann from NTCA, confirmed that shifting focus from individual-facing AI to organizational tools was the right move. His feedback gave us confidence to prioritize administrative efficiency and cross-organizational collaboration over consumer-facing apps.

Based on research, we established four design recommendations:

An AI Resource Platform

We proposed a Retrieval-Augmented Generation (RAG) AI platform that consolidates information from food programs across Michigan into a searchable, conversational interface. Unlike static databases, RAG models retrieve real-time, context-specific information from organizational knowledge bases, ensuring accurate and up-to-date responses.

For organizations, it helps quickly locate partner resources, grant opportunities, and program eligibility criteria. For staff, it provides access to policy documents and service directories in one interface without administrative searching. For coordination, it identifies overlapping service areas, gaps in coverage, and opportunities for resource sharing.

Research shows RAG models in supply chain and social work contexts improve decision-making and reduce administrative burden, critical benefits for rural food systems with limited staff.

Broadband Partnership Incentives

To encourage adoption, we proposed partnering with broadband companies to offer discounts to organizations that contribute information to the platform. Organizations expand their outreach and share resources, while broadband companies grow their customer base in underserved rural areas.

The platform would be designed for desktop use, as most food assistance staff access professional tools via full-screen interfaces. It assumes stable internet access and basic digital literacy, which limits reach in areas with broadband gaps.

We acknowledged several constraints that shaped our solution.

Our work was selected by peers and instructors to present in a combined-section showcase in front of stakeholders and clients from other projects. This recognition validated our research rigor and the strategic value of our solution.

Partner feedback emphasized the importance of organizational collaboration and AI for back-office efficiency, both central to our final design. Peer feedback encouraged us to clarify our problem statement and incorporate recent policy changes like SNAP cuts, which we integrated into our final report.

Our solution emerged directly from interview insights, not assumptions. The pivot from individual tools to organizational infrastructure only made sense because we listened to stakeholders who told us where the real pain points were.

In addition, the limited time for this research project and access to older adults to interview forced us to prioritize organizational efficiency over end-user apps. This constraint strengthened our solution by addressing systemic barriers rather than surface-level symptoms.

Furthermore, using comparative nested matrices and usability scales helped us move beyond gut feelings to data-informed decisions about which solution would provide the most value with the least friction.

Next Steps

If we had more time and resources, I would conduct usability testing with food assistance staff to validate the platform's interface. I would also interview older adults directly to understand their preferences for receiving information and prototype the RAG model with a pilot group of organizations to test real-world effectiveness.

Overall, this project taught me that good UX research isn't about having all the answers. It's about asking the right questions, listening deeply, and designing solutions that acknowledge complexity rather than oversimplifying it. >