Interview transcript: 

Terry Gerton We’re going to talk about a very important program today, the SNAP program, Supplemental Nutrition Assistance Program. We got a glimpse of just how important this was during the government shutdown when those benefits were paused. You’ve worked with this program very closely with the state administration of this. Tell us about some of the biggest challenges states have in administering the SNAP program.

Andrew Joiner Well, certainly. Look, it’s the largest anti-hunger program in the U.S. I think approximately 42 million Americans rely on this benefit to help put food on the table. And look, the most recent requirements are that you submit for the eligibility of this program. And so the states administer that eligibility on behalf of the federal government. It’s about a $100 billion entitlement program that we fund through taxpayer assistance, so it’s quite a large program, quite a large amount of Americans, and it does put food on the table, and it’s administered and paid out on a monthly basis, and that eligibility is administered by the states who essentially have to go by applicant by applicant, household by household, to ensure the right amount of a benefit is paid to those individuals.

Terry Gerton The application process itself is pretty backward, maybe. It’s still in paper. And the One Big Beautiful Bill Act that passed last summer added additional application requirements. What do you see in terms of the strain that that puts on the state-administrating agencies?

Andrew Joiner Well, we like to call it the big beautiful bottleneck. I think everyone wants to get this assistance out to the families that need it. But at the heart of this application eligibility process are documents supporting your eligibility. Firstly, you have to submit your identity, that you’re a citizen and a resident of the state. You have to submit actually your, essentially, income and your income eligibility determines the amount of benefit you get. So those are things like pay stubs, bank statements. If you’re a head of household, you may have to submit utility bills. So it’s quite a lot of paperwork that has to get processed and adjudicated to get the benefits. We’ve typically tried to use caseworkers along with system integrators, we call them, which are large consulting companies that try to help administer this benefit. The reality is 44 states are failing, essentially, what we give as a payment error rate. So that is how many, what percent of errors can you make in your determination of the benefit? And then the second part is you need to pay the benefit within 30 days. And the average eligibility payment is about 26 days. So it’s taking quite a long time to get the payment. And 44 states are above the error rate, which is 6%. And the problem with that is the government can withhold, dollar for dollar, anything above 6% that you’re paying incorrectly. They can withhold that and payment to the states. So it’s quite a stressful environment in that we’ve got a time period to pay essentially this assistance and you need to pay it accurately. And both those are stressful to the states.

Terry Gerton You’ve said that the problem is not the portal, it’s the paper. The fact that this application process still happens in paper with all kinds of challenges that that poses, this seems like a tailor-made AI solution environment. How could we bring more technology into the process to streamline the evaluation of these applications?

Andrew Joiner Well, it’s a great question, and that’s what Hyperscience, the company that I run, is really focused on trying to do on behalf of the states. So there’s quite a huge human impact here. So the CIOs, the folks at the HHS organizations, essentially are stressed, because now twice the amount of the paperwork is now flowing in. And if you’re using caseworkers and system integrators to process that, you’ve essentially doubled your bill. That’s how you scale labor. So AI can help reduce the amount of burden, administrative burden that the states are going to have to take on to essentially handle twice the paperwork. But then you move to the caseworkers themselves who are trying to review this paperwork and calculate the accurate payment of these benefits. And they spend more than 80% of their time essentially addressing manual type efforts and paperwork, like correcting when the papers get scanned in, they’re upside down, they’ve got torn, they’re not clear, they have messy handwriting. The pay stubs themselves are complicated tables, and so calculating payroll deductions, all of that is complicated. Really, humans were not built to do this at scale, but AI, especially with something like Hyperscience, which was purpose-built to read human friendly information at scale, can really help reduce the amount of errors that the humans are making when they have to process 42 million of these on a monthly basis. And so we can read all of the documents, essentially we can make sure that everything’s in good order, that you’ve submitted the right identity documents, that you submitted the write pay support information or head of household information. So we can sure everything’s good order before you go off and go back to work and go on and do something. That’s typically a big delay in the process. Then we can accurately extract the information so the caseworkers aren’t having to sift through messy handwriting, multiple languages. The AI can handle all that. And then really what the caseworker can focus on is really helping the applicants understand if they’re missing documentation, if there’s gaps in their documentation, what they need to do to meet the policy to get the payment more quickly. They can focus on those human aspects. And I think that’s a win-win at the end of the day. At the end the day, the reason why there’s twice the paperwork, one side of the spectrum is trying to reduce fraud. They’re also trying to make sure the eligibility is going to the folks who are eligible for it. So that’s one side the spectrum. There’s another side of spectrum that wants to make sure that an accurate amount of benefit is paid and is broadly accessible. And so at the heart of this is, let’s just get through the paperwork as efficiently as possible. And this is an area where AI scales naturally, and AI can really help the caseworkers, it can help the CIOs who are stressed, and then it can also help the applicants who really want this benefit quickly.

Terry Gerton I’m speaking with Andrew Joiner. He’s the CEO of Hyperscience. You mentioned earlier that 44 states are failing to meet the performance requirements embedded in the SNAP program. Why aren’t more of the states picking up on some of these technology solutions to help make the processing easier, reduce their backlog?

Andrew Joiner Well, there hasn’t been a technology like Hyperscience that works across such a broad spectrum of documents and because we have to administer this to these applicants, the best way to do it is through caseworkers. It just happens to be a high turnover job because it’s quite stressful. There’s time barriers, about 30 to 40% attrition rates of the caseworkers nationwide. So it’s quite a high stress. And that has really only historically been the best way. There haven’t been a good set of technologies that allowed states to administer this quickly and at scale. AI, the advancements have come so quickly that now, no matter what the types of documents that are being submitted, whether it’s the form that’s being submitted, whether it’s the identity documents that validate what was on the form, it’s the income requirements that are also specified on the form, we can do now this kind of cross document comparisons at scale and with high accuracy because of the power of AI. And so I think most of the states within the next 18 months have no choice but to adopt a technology like AI to assist with the administration of these programs. We’ve already done this at the Social Security Administration in one of the largest document programs issued by the government. We do it for over 250 million Americans to make sure that that assistance is paid out when you submit your social security claims. And we also do it for the Veteran Affairs Association. There’s over 11 million veterans who are submitting complex documents to get their claims reimbursed for all their medical assistance, a very important constituency. They used to wait over three to six months to get their claims adjudicated. But with the help of AI, we’ve now gotten it down to less than three days. So these are the types of things that I think most states will start adopting, because the results are measurable and there’s quite a big human impact that we can produce on the other side of it.

Terry Gerton You mentioned some other examples of casework in programs across the federal government, VA, Social Security, other sorts of benefit programs. If AI is deployed to really improve this caseworker customer interaction and streamline the process, what needs to come next? Are there new policies, new oversight functions, new governance mechanisms to make sure that we keep private information private and that information flows smoothly and that customers receive the benefits they’re entitled to?

Andrew Joiner It’s a great question. The federal government is really leading the way in the ethical use of private information and forms of delivering efficacy with the government. So what we’re able to provide the states is their own state instance where the citizens and residents of that state that benefit from AI, all of their information never leaves the jurisdiction of the state, it runs locally. And we have safeguards in place where, if information is extracted from the program, we can redact, we can use synthetic information to help train the models. And so there’s actually quite a leadership position that the U.S. government and the states are able to take in the handling of information to help adjudicate some of these government processes. We’ve used it for logistics, for warfare, for a number of different reasons all throughout the U.S. government. The U.S. government has some of the strictest safeguards in terms of security and governance of information with the FedRAMP and now StateRAMP programs. And so we’ve gone through that process with Hypercience as an example, to ensure that leaders and information professionals who lead these states know that the information and the handling and the use of AI will meet the highest, stringent safeguards that they’ve put in place.

The post The SNAP program is under pressure, and states are drowning in paper as new mandates kick in first appeared on Federal News Network.