State SNAP agencies are overloaded — what *actually* can be done?

Also: SNAP AI guidance, reading section 208 of the Intergovernmental Personnel Act of 1970

SNAP programs are slammed; what can be done?

You might have gathered by now, dear reader, that I spend a fair bit of time on the “food stamps” sub-reddit, /r/foodstamps.

(For what it’s worth, just helping people is a recharging experience relative to abstract work intervening in complex systems. I have knowledge on deductions for self-employed earned income that is otherwise completely useless. If you do too, I recommend answering some questions!)

There was a recent thread there complaining about one state agency “screw[ing] up” a user’s case.

This kind of post, venting, is not uncommon. But what I found more interesting was the replies from front-line eligibility workers (many of whom are on there, lending their program expertise in their free time in a personal capacity, bless them.)

One caseworker on the emotional toll:

There is also a lot of burnout in Human Services for the employees - local government employees are not paid well at all, but the main reason was the emotional strain. I lost track of how many times I cried during my breaks, in my car, at home, etc. during my time working in Human Services because we saw trauma all day every day - and it haunts us long after the interactions end. I had more people than I could count tell me some variation of "YOU are the reason I cant feed my children tonight" when they were denied benefits or didn't get the amount they anticipated - never mind that workers can't just hit "Approve" or put an amount in the system to pay out. The income limits are set by the federal government and all the workers can do is put your information in the system and tell you what the system has determined your benefit amount to be.

Another caseworker on the current work volume:

I've worked as a caseworker for 11 years, and the amount of work we have now is unmanageable. The workers in my office average 90-100 recertification a month plus applications and semi-annual reporting. The state keeps pushing new programs with new rules on workers without hiring more people to do the work. Add to that a system that wasn't built to handle the volume of work and stops working sometimes for hours at a time. Workers aren't intentionally messing with cases or ignoring renewals. They're simply drowning if anyone on here thinks they can do a better job, please apply for one of the many openings..

Another worker on staffing levels:

I am miserable. I am so burnt out and wanna sob everyday. My paychecks are actually less now than they were last year. I only bring home almost $1100 biweekly. I think people assume we make loads of money & in reality we are struggling just as much as everyone else. They keep us just above the income limit to not get help, but not enough to survive. We only have 37% of the staff needed to run normally. We cannot hire enough people. Or if we do hire them, they don’t make it through training. They throw more work on our already MASSIVE workload and don’t understand why no one is happy. It’s exhausting.

Texas, as I’ve written about before, is particularly bad right now:

I work for the information and referral side. It’s so sad and definitely taking a toll on my mental health. This is the worst I’ve seen in Texas in the 12 years I have been doing this.

One of the primary things that jumps out from that thread to me is the systems nature of the problem. In a lot of ways, caseworkers and the clients they serve are not in drastically dissimilar situations. The effect of the system being overloaded is different on each side, but this is not some zero-sum game where either clients get the benefit or the workers do: interests are more aligned than not, even if sometimes clients don't perceive it as so.

But the question I wrestle with in this is — given things are so overloaded, what actual interventions can be feasibly made here?

What are options on what could be done?

One answer, at an individual client level, would be: file an appeal (fair hearing.) That might work to escalate a long-overdue case to get it resolved. But I also think it’s reasonable to infer that an appeal is more triage; it does not fundamentally affect workload so much as reallocate limited bandwidth from one client pending to another.

(An argument could be made that mass appeals could lead to some second-order response, but I think the effective-advocacy question there would be: what change is it that one would want to see? So we get back to what is to be done?)

We also do have some answers from the pandemic. These included things like process flexibility (e.g. waiving interviews so a case could be approved without one), additional temporary funding (though how much can staffing be improved by temporary funding is a question), some more latitude on things like quality control evaluation metrics (recognizing that strict accuracy and timely benefits processing are to some extent a tradeoff in an emergency situation.)

You could answer that this is fundamentally about adding staff capacity. But that probably requires more funding at the state level? Sure, there was temporary federal funding, but that’s gone. And to my knowledge most if not all admin funding is a 50-50 match between state and feds. So in this model of change, the real target is the state budget? And that’s a long-lag cycle. Certainly not something that moves in weeks or months. Also, at least in my state, it's not like the budget situation is very rosy.

Another answer, as seen in the pandemic, is waiving the interview, which is time-intensive and creates coordination (scheduling) waste work. I recently learned that a group of advocates are petitioning USDA to remove the SNAP interview requirement.

(I was more surprised to learn that the SNAP interview is not statutorily required: I’d heard time and again it was! Reminds of of Robert Caro’s maxim: “turn every page.”)

That could certainly lead to faster throughput, but I’m also sure there’s much gnashing of teeth about what side effects that could have, for example on payment accuracy (the structurally highest-prioritized goal as I’ve written about before.)

And while I like the idea of making the interview optional/conditional on circumstances, there are also tradeoffs here:

• It adds burden to making sure the up front application or renewal is complete and accurate when the program is very complicated (and I’d say most clients should not need to understand all the details)

• With an interview, a client can be hand-held through questions and may not need to submit all verification documents up front, because some won’t need to be verified (in fast, most agencies should use Work Number for income in lieu of requiring paystubs when available, given new federal funding of that) and instead only the specific documents outstanding could be told to the client at the interview

Another answer might be automation or other technology interventions. At the end of the day, workload is an operational problem where technology does have leverage. If you can save 3 minutes of a worker’s time looking at paystubs by having OCR (and even, yes, AI) extract structured data automatically, that worker has more time. There’s lots of little sub-tasks in eligibility work where automation or technology is a lever. Though for a variety of reasons that people are aware of, the track record on technology is not superb in these particular programs necessarily.

Getting to this point of my writing, I actually went to Google to look up something about automation in SNAP, and stumbled upon… new guidance on “advanced automation” from FNS as of January 10, 2024!

It is here: https://www.fns.usda.gov/snap/advanced-automation

I will confess upon first reading... juxtaposed with the caseworker quotes at the beginning of this post, it’s… not guidance that makes me excited that new tooling is being encouraged to be deployed to address the dire state of things?

My quick read (and please reply to me with corrections!) is that it effectively makes most automation that states would like to use subject to either pre-notification or actually require permission from USDA/FNS. (This is through the “major change” rule, waivers, etc.)

I’m not a state person, but in my experience, these are not low-burden activities for states. I also get very curious about what kind of technical assistance or support states get in return for engaging USDA on these, or if it is more a one-way street of waiting, reporting up, then getting the green light to go ahead.

I know there’s a lot of fear of the downsides of technology and automation interventions… and also it makes me return to a theme I see time and again:

We collectively (and government specifically) seem to be much more attuned to the risks of change than the (ongoing) risks of continuing a status quo path causing harm.

I mean, if clients in Texas and Florida are waiting 8 weeks, 12 weeks, longer without benefits (when the federal standard is 30 days) we have to ask ourselves — if we can’t meet all of the competing policy goals in reality today, which do we give more weight? This is the uncomfortable question every public administrator faces, and many faced in extremes during the pandemic.

A friend of mine once called this “the incoherence of the compromise”: it is not a logical necessity that policies that accrete by negotiation and compromises across competing aims and priorities are themselves logically consistent or clear in what the hierarchy of goals is. (It’s not easy being a public servant given this! Someone is always yelling at you about the tradeoffs necessary in implementing anything.)

I’m also surprised because this guidance does touch specifically on AI and broadly makes it “by permission” in my read. This seems to me in a bit of contradiction with the Administration’s AI Executive Order, which tasked USDA with issuing guidance on benefit programs’ use of AI where the first-listed goal is:

“(A) maximize program access for eligible recipients”

Now if one’s mindset is that technology is primarily a risk to program access, highly permissioning its use makes sense!

But… if we have technology that could help ease workload and get people benefits and we’re slowing that down… well, I don’t know.

I confess I find the situation in states like Texas (where I see reports of 120-day waits) more intolerable than abstract risks in use of tech. But, again, I’m maybe here because… it’s not like I see some other levers readily available to be pulled to fix that intolerable situation. Given that, I get concerned about hampering the use of what levers do exist.

Also, it’s worth noting if you use more automation and free up human time, you can potentially reallocate that human bandwidth to handling appeals where clients disagree with the decision.

To some extent, this is a better use of time, because it is applying the human attention and judgment specifically to where we think computers have done something wrong.

Compare that to putting human attention and judgment towards tasks where policy actually allows very little judgment to be applied, and it is mostly a math calculation that computers are good at. (Again, see the caseworker quote to the effect of all I can do for my clients is put their information in and see if the system approves them or not.)

I’m not calling for some radical shift in the opposite direction: rather, I would say I’m surprised to see such caution/administrative overhead on automation at a time when I’m hearing about 3-month delays in eligible people getting money for food.

I look forward to seeing how USDA’s agency-level AI guidance called for by the Executive Order approaches this. This FNS guidance is not that which the AI EO calls for, but rather program-level guidance. USDA itself is committed in the EO to releasing its own as it relates to benefit programs it runs.

On merit staffing: a weird tangent wherein I end up reading implementing guidance of section 208 of the Intergovernmental Personnel Act of 1970

One of the big things mentioned in the context of automation and AI (all around but specifically in SNAP) is on work that must be done by “merit system personnel.”

A lot of the talk tends to round to: “robots can’t do X, a merit staff member must.” And I understand much of the rationale for this!

But, like, weird, non-lawyer question for the lawyers and wonks out there:

In 5 CFR 900.603, part (b) states that one of the principles of a merit-based system is “Providing equitable and adequate compensation.”

In the states where eligibility staff are (partly due to inflation) being paid less than fast food restaurants down the street¹ (which is not all staters but some!)… wouldn’t that be… against the merit staffing rules? And wouldn’t remediation of that require increasing that pay, a bit of a legal requirement for the state to allocate more dollars to retaining staff?

I don’t know! But I’m trying to turn every page.

It certainly strikes me that improving the compensation of staff to be able to recruit and retain more effectively and giving them the tools to be more productive in how much workload they can tackle (tech and, yes, even AI) seems a more fruitful path.

Anyway, I’m puzzling through this crisis myself. I use this as much to think out loud and develop my thinking as anything else. If you have some ideas on what can be done here, please really do shoot me a reply/email me!

See you on Reddit!

1. This is not made up! I have heard a caseworker in a state say they are losing staff to fast food restaurants because the pay has become relatively higher due to inflation!