In Conversation with mRelief
Every year, nearly 8 million households in the U.S. qualify for SNAP benefits — also known as food stamps — but don’t receive those benefits due to barriers such as social stigma, lack of awareness, and inefficient application processes.
mRelief, a GitLab Foundation grantee, exists to change that. Since 2014, mRelief has unlocked $1.4 billion in SNAP benefits for more than 4 million individuals. Its tech-enabled SNAP eligibility screener takes just over two minutes to complete and its simplified online application takes about 13 minutes — compared to the standard application process which requires a 17-page application or 90-minute phone call and as many as 10 required documents. mRelief’s simplified applications are approved for benefits 70% of the time, compared to 58% without mRelief’s support.
With a GitLab Foundation grant through the AI for Economic Opportunity Fund, mRelief aims to increase application approval rates even further. Our CEO, Ellie Bertani, sat down with Porschia Davis, mRelief’s executive director, and Dize Hacioglu, its chief technology officer, to discuss the organization’s mission, how they overcome challenges, and their plans for the future.
Ellie Bertani: What does the process typically look like for someone applying for SNAP benefits?
Porschia Davis: I’ll share what we have found through research and my own personal experience being on food stamps in college, because my experience was very typical.
First, there’s usually no eligibility screener. You know you don’t have enough money to put food on the table, so you go to a state office and ask for the food stamps application.
Submitting that application is kind of like shooting a rock in the air to see where it lands. It’s not a transparent process. I got denied a couple of times, because I didn’t know exactly what was needed for my specific circumstances. It’s the same for Americans throughout this country who all have different needs.
For me, the process was cumbersome; it was exhausting; it was humiliating. And that's very similar to the clients that we see. They're tired. They've been through this process so many times and they have very little guidance.
At mRelief, our goal is to put ourselves out of business. We want to put food on people's tables. We're grateful to partner with the GitLab Foundation to put an end to unnecessary suffering. If you're hungry and you're eligible, you should be getting SNAP benefits.
Ellie Bertani: Absolutely. We should reduce as many barriers as possible and your personal experience seems to inform your leadership on these issues.
I imagine these challenges were part of the genesis for your GitLab Foundation grant. Let’s transition to talk about the specific problem you're seeking to solve and how you are thinking about applying AI.
Dize Hacioglu: One of the biggest pain points during the SNAP application process is the burden of submitting documents to verify the information that you provide. It can take a long time to collect and submit all the paperwork.
We facilitate that process thanks to our client support team, which does quality assurance on 6,000 to 10,000 documents per month. The team has two full-time staff members and one part-time, so they put in a ton of work to make sure that applications can be processed in a timely manner.
We think AI could be a great solution to this problem. Using AI to assess document quality not only allows our staff to dedicate time to clients in more meaningful ways, but also gives clients much quicker feedback. Ideally, AI will reduce the amount of time that our team — and understaffed agencies and case managers at the state level — need to spend on each application.
Ellie Bertani: How do you navigate challenges as you explore and implement AI?
Dize Hacioglu: This is our first AI project, so we did a lot of learning and ran into some challenges with identifying the right technology to use. We were really hoping to build our own machine learning model to help us identify which documents were going to pass our quality checks, but that proved not to be feasible. We finally settled on the right tool to use and now we're starting implementation.
Porschia Davis: We're committed to approach challenges by creating policies, guiding frameworks and principles for AI that focus on accessibility, equity and ethics. We plan to launch an ethics committee that will include local community members with lived SNAP experience and mRelief’s client advocates.
Ellie Bertani: That's great. A representative ethics committee that can guide this, and other AI projects over time, is the type of careful forward thinking that is so critical in the nonprofit space, where clients can be quite vulnerable.
Are there other ways that mRelief is currently using AI or thinking about using AI in the future? Are you optimistic about this technology’s potential impact on your work?
Dize Hacioglu: Definitely. We've been excited about AI since it first exploded because of all the ways we might use it at mRelief. In addition to the document verification process, we’re using AI to improve our existing chatbot so it can serve as a personalized application assistant. It will support clients through the entire process, from learning about their eligibility to receiving their SNAP card in hand. In the future, we plan to use AI to improve our code quality and architecture, and to build out simplified applications in more states.
Ellie Bertani: Who have you found to be strong partners in this work?
Dize Hacioglu: Tribe AI comes to mind. We’ve worked with them thanks to the GitLab Foundation grant. They were the ones to say, maybe you don't need to roll out your own machine learning model; you can use something that already exists.
We've also been lucky to work with Open AI within the GitLab Foundation grantee cohort, and we got assistance from Google through a grant from Google.org.
Ellie Bertani: Among the AI for Economic Opportunity Fund grantees, there are several doing work in the benefits access space. I'm curious if you found any benefit in hearing about their work or learning from how they approach problems.
Dize Hacioglu: We've gotten feedback from the cohort about our approach, particularly when we were trying to figure out what wasn't going well with our machine learning model. It’s helpful to hear about the similarities between our AI projects and the others in this cohort. We're trying to build a coalition with some of the organizations to do other, non-AI-related benefits work. We were able to get acquainted with them through this grant and our Google grant.
Ellie Bertani: Impact measurement is really important to us as a foundation. How do you think about impact measurement to validate and improve upon your work over time?
Porschia Davis: Our main metric is application approval rate, which is whether our clients successfully get approved for SNAP benefits. Those approval rates tie directly back to how many dollars a family receives to put food on their tables.
Dize Hacioglu: With the AI project specifically, we’re hoping to get an additional 100,000 individuals approved for SNAP within one year of launching the document screener, which would unlock $138 million in benefits for people in need. We also are trying to transform the system to make sure that it promotes the inherent dignity of every person. We want to do this work with the love and care that people deserve while delivering on our promises.
Ellie Bertani: This has been a real joy and a privilege — I appreciate your time. Thank you for doing the hard work to improve the financial stability of so many people in need.
Return on Investment
1,007x ROI
This project is estimated to increase annual earnings by $1,368 per person, for a total additional lifetime earnings of $1,368 per person
Total lifetime earnings increase across all participants: $136,285,632
$100,000 invested
