Policy Analysis Case Study: How AI Validated My 10-Year-Old Graduate Analysis
Read for free on Medium. This is cross-posted on Substack for reach.
Policy Analysis Case Study: My 2015 Graduate Work on SNAP
This policy analysis case study documents my journey from graduate student to 2025, examining how policy analysis frameworks have evolved. This policy analysis case study compares my 2015 graduate policy analysis on SNAP senior enrollment with AI-powered analysis using the same root causes.
It was 2015 and I had enrolled in a public policy class for my graduate program. I had chosen to take an elective that was outside of my specialization (public health, physical activity and healthy lifestyles); I always went off script in my education program.
Our assignment for the year was to formulate public policy recommendations by analyzing a societal problem. She gave us a list to chose from. Having little exposure to political science or policy issues, I waffled between (1) low Supplemental Nutrition Assistance Program (SNAP) enrollment for low-income seniors and (2) farm-to-school practices; they aligned best with my degree program. We weren’t given much to go off of — just two or three sentences explaining what we were addressing. I leaned into my need to go off script and chose low SNAP enrollment for low-income seniors. I remember her saying, “You chose the most difficult one.” I didn’t quite understand why until I started my research.
Root Cause Analysis
I learned only one-third of eligible seniors were enrolled in SNAP. The barriers were:
- 76% reported stigma
- 67% reported embarrassment
- Misinformation about benefit amounts
- Complex application processes
- State-by-state implementation inconsistencies
We learned how to examine root causes through an economic lens. I argued that the underutilization of SNAP by low-income seniors is largely due to market failures — decentralized government and information asymmetries.
This policy analysis case study demonstrates how systematic root cause analysis identifies the core barriers affecting policy outcomes. Understanding these root causes in my policy analysis case study became the foundation for later recommendations.
My analysis examined four policy alternatives:
The status quo: I rejected this as an option because it did not address the problem
Data-matched outreach: Match SNAP data with other assistance programs to identify eligible non-enrolled seniors
- Evidence base: Philadelphia pilot using Better Data Trust technology achieved 15% enrollment increase within one year
Adjusting Funding Formula + Application Assistance: Fix underfunding and provide application help at food bank sites
- Evidence base: Michigan pilot showed 7% annual increase; New York model provided facilitated application assistance
Development of a simplified elderly application: Create a shorter, senior-specific application
- Evidence base: Alabama’s Elderly Simplified Application Project (AESAP) and similar pilots in Southern states
The four alternatives were assessed in terms of:
- Efficiency (are the net benefits more than the net costs?)
- Effectiveness (does the policy effectively enroll seniors?)
- Political feasibility (how likely will this policy be implemented?)
- Equity (does the policy reach seniors in the most need and reduce the burden on food banks?) and
- Ease of implementation (does the alternative require extra time, resources, or training to implement?)
My analysis concluded that the United States Department of Agriculture (USDA), in the short-term, should implement a simplified elderly application in order to reduce the perceived costs of applying. It ranked well on efficiency, had moderate effectiveness, and was politically feasible because it didn’t require major system changes. For the long-term, I recommended data-matched outreach because it was the most effective at actually reaching unenrolled seniors, even though it would be harder to implement.
The Present: AI Meets My 2015 Analysis
I decided to revisit my policy analysis case study from 2015 as a validation exercise. By inputting the same root causes into my AI framework, I could compare how this policy analysis case study evolved a decade later.
I’d developed an assessment framework for organizations and was searching for validation. I had an “aha” moment and decided to go back to my policy analysis work. I put the same root causes from my paper (sans the recommendations in the paper) into my AI framework. The goal was for it to analyze the SNAP senior enrollment problem, provide recommendations, and compare to my analysis and policies that were implemented around that time.
Time investment comparison:
- Graduate analysis: An entire semester (roughly 4 months of research, analysis, and writing)
- AI analysis: 20–30 minutes
Framework Recommendations
It generated 3 Quick Win strategies and 5 Long-Term strategies.
SNAP Application Simplification Initiative
- Standardized senior-friendly application materials
- Larger fonts, clearer language
- Visual guides, telephone options
- Staff training on senior-appropriate service
Senior SNAP Information Campaign
- Target misinformation about benefits
- Partner with trusted community organizations
- Multiple communication channels for seniors
Healthcare-SNAP Referral Pilot
- Screen for food insecurity in healthcare settings
- Create referral pathways to SNAP enrollment
- Train healthcare staff on SNAP processes
Long-Term Strategies:
Comprehensive SNAP Modernization
- Automatic enrollment for seniors on other federal benefits
- Simplified recertification across programs
Integrated Data Systems
- Cross-program eligibility verification
- Automated data sharing between programs
State Performance Incentives
- Reward states for improving senior participation
- Technical assistance for implementation
Community-Based Outreach Network
- Leverage senior centers, faith groups, community organizations
- Systematic training and coordination
Sustainable Funding Framework
- Long-term capacity planning for demographic changes
Finding Alignment
Below you’ll see a table with the root causes identified, what I recommended in 2015, what AI recommended in 2025, and what USDA and Colorado implemented. Note: My analysis did not correlated with implementation of these policies; this is intended to compare recommendations. Additionally, in the “actually implemented” column where it states “not implemented systematically,” I was not able to find these programs — if you are able to or know them, feel free to leave a comment.
Core agreement: Both my 2015 analysis and AI identified simplified applications and data-matching/enrollment as the critical interventions.
Key difference: AI developed additional strategies around healthcare integration and community partnerships that I didn’t prioritize in 2015.
This policy analysis case study table reveals the alignment between my 2015 recommendations and what was actually implemented. The policy analysis case study shows that both approaches identified simplified applications and data-matching as critical interventions.
The Differences
My 2015 analysis looked at pilots and interventions that other States had conducted with accompanying data that showed their effectiveness in elderly SNAP enrollment. I had limited knowledge of how States handled enrollment due to decentralization of how SNAP programs are administered. My policy recommendations were framed with this in mind, i.e., general recommendations with the ability for different States to adopt how they see fit.
Because USDA is a large federal body, individual implementation from state to state cannot be directed. This is often why we see changes such as funding adjustments, recertification period guidelines, or allowances for states to implement their own simplified elderly application.
What Does This Validate?
AI is not better in policy analysis. In fact, when I put in data from my graduate policy paper, it arrived at similar conclusions. However, it does show that contextual understanding of the core problem matters.
This policy analysis case study validates that contextual understanding of root causes matters more than the tool used for analysis. Whether conducted over 4 months or 30 minutes, this policy analysis case study demonstrates that evidence-based frameworks produce similar core recommendations.
When Data and Root Cause Analysis Meet
Two different analysis identified core interventions could be and implemented and scaled (simplified application and cross-program integration) by utilizing evidence-based frameworks that are context specific.
- When you have a core understanding of the problem and how it impacts your stakeholders, AI can be an effective thought partner. My 2015 analysis took about 4 months of research how seniors were impacted, how SNAP was funded and distributed, how states implemented and determined eligibility. My Professor was correct — I did chose the most difficult problem. I had to learn about a whole governmental ecosystem AND THEN propose a problem (little did I know). However, after I had this understanding and put it into my framework 10 years later, the analysis took 20–30 minutes and provided additional strategies to address other root causes.
- Humans determine the quality of outputs through contextual understand and nuance — either through thorough our own analysis or through AI. There were multiple strategies I did not recommend because I understood some of the programs were difficult to implement or did not have political feasibility. AI’s output largely depended on how information and what constraints I gave it. When I initially ran the report, I did not add add constraints about focusing on policies or structures — AI provided suggestions like “Develop a Senior Navigator Program” for outreach and enrollment. Sounds like an amazing program, however if I didn’t constrain what the output, then suggestions do not fit my goals or needs. Without understanding your community or constraints, program or policy proposal will likely to fail.
What I Learned
When I first started this, I wanted to see what AI would generate but what I learned from the outputs was how to better design for human input and constraints. I discovered that I need to reframe questions to accommodate the different type of work organizations do.
Reflecting on this policy analysis case study, I realized the value isn’t in the speed of analysis, but in the quality of input and constraints. This policy analysis case study taught me how to better design for human understanding and organizational capacity.
Limitations: With access to the internet, it’s possible that AI analyzed the problem and found strategies that were already developed. The next step is to conduct this with an organization that is currently working to understand it’s challenges. However, I’ve done similar analysis with individuals who are currently working on problems and AI found strategies to fit their needs.
This is really about:
Good data → Systematic root cause analysis → Identifying high-impact interventions
The difference in output is in the nuance, contextual understanding of your problem and how it impacts your stakeholders, and understanding what you have capacity for. This policy analysis case study is really about demonstrating that good data,
systematic root cause analysis, and understanding your stakeholders lead to
high-impact interventions. Every policy analysis case study is unique, and
context determines success.
Have a policy analysis case study from your own work? The most valuable lessons come from reviewing past analysis against current outcomes. Your policy analysis case study could reveal insights that inform better future decisions.
To think, I was just a graduate student doing an assignment because she felt like going off-script one day.