A broken SNAP call center gets comeuppance from the courts

Also: I have a new job; and some musings on agency in agencies

Programming note: Dave has a new job

I recently joined Propel as Director of Emerging Technology, with a focus on applying AI to helping users navigate management of their SNAP case. I’ve known the folks at Propel for a long time — we all started messing applying technology leverage in SNAP around the same time, about 10 years ago; I like to think of us as kicking off multiple novel experiments in models in the space — and it’s been a pleasure so far.

Propel runs the Propel App (previously called Providers), which has more than 5 million active monthly (SNAP-enrolled) users. As you might imagine, the opportunity to help people at that scale (perhaps the single most concentrated surface area beyond large national retailers?) is quite exciting.

As part of my remit, I do plan to build and learn in public (and what a nice thing it is to be in an org supportive of that!) but I must reiterate to you, my dear reader, that all thoughts and opinions expressed here remain solely my own.

Missouri’s SNAP agency gets comeuppance in the courts over its broken call center

A federal court recently ruled against the State of Missouri in a case brought by SNAP advocates claiming broadly that:

1. Given SNAP requires an interview

2. And given Missouri uses an “on-demand interview waiver” which means that rather than scheduling an interview, people must call into the call center to be interviewed

3. And given that it is extremely difficult for many (most?) clients to get through to anyone on the call center, given factors like staffing, workload, etc.

4. Then this situation violates federal legal requirements related to SNAP (and ADA)

A reader might take issue with my characterizing the call center as “broken,” but I do think it’s fair given the facts here. I recommend reading the whole ruling — it’s a real barnburner, if a bit depressing.

One line that I think justifies “broken”:

Defendant’s system continues to automatically deny all applications when an interview has not been completed within 30 days and the record reflects the number of denials due to a missing an interview is around 50%.

50% of denials are due to not being able to get an interview! And no other reason!

You can read lots more documents on the excellent CourtListener service (which saves and publishes court documents that cost money using a nifty browser extension called RECAP): https://www.courtlistener.com/docket/63105919/holmes-v-knodell/

To be quite honest, I was surprised when I read this! Not because this is clearly bad but because it’s been an issue for so long but I’d yet to see a court find it illegal.

In the transcript of a motion hearing (email me if you want a copy), there are a number of interesting bits, for example:

in July 2023 there were more than 64,000 calls deflected just from the interview line alone

The court, asking a reasonable question:

At that point am I micromanaging the agency, telling them how to allocate their staff?

And this bit, which I find reads more nicely in its original court document aesthetic:

A somewhat delightful moment where a plaintiff’s attorney accidentally encapsulates the long arc of these programs’ evolutions:

I'm going to be honest with Your Honor. I was still in college when we switched from food stamps to SNAP.

And a bit on the tough moral choice the federal oversight entity (FNS at USDA) faces in dealing with less-than-awesome state agencies:

But on the other hand the state has some words for FNS!

Later, the state refers to this as a “frankly asinine requirement.”

(Detail-oriented readers will remember that advocates are in fact petitioning FNS to remove the interview requirement. FNS is also studying the impact, per the Federal Register. This is a thread to pull another time, though I’ll confess that at the moment it’s hard to make the case to remove the interview barrier when FNS has data on payment errors but FNS does not require collection of data on benefits not paid due to interview denials. Interviews likely help make payments more accurate, but they also make some people not get paid anything. But we have an asymmetric data fidelity problem.)

The state ends up arguing — not unreasonably! — that they have tried to increase staffing, but they simply cannot hire and retain sufficient staff to do more interviews at what they currently pay.

The judge — a former state legislator himself, funny enough! — gets rather cranky at that, noting that the state legislature is implementing tax cuts.

(This is in fact an unspoken reality of much of federally-overseen, state-adminstered benefits programs: in many cases, the state agency leadership cannot take the actions needed to fix a problem. That decision authority may rest with the legislature [as in budgeting], the Governor, etc. This is a principal-agent problem, and at times this leads to the funny reality that at times a state civil servant wouldn’t mind being sued — because absent such pressure, they do not have the mandate to fix a thing.)

Anyway, the whole thing’s interesting. Most interesting to me is that the SNAP interview barrier might be found to be illegal in some situations in a way that it has honestly to this point been largely a story of “tough cookies.”

I do think there’s also a kinder situational awareness explanation of some of this. When I was running a SNAP application assistance service at scale in California, one of the consistent findings we had in (high quality data) client outcomes was that a whole lot (20-40%) of application denials were for a missed interview.

While many county directors knew quite well the other most common denial reasons (missing verifications, over income) very few guessed that the interview denial rate was that significant.

I have to wonder if the SNAP ecosystem overall suffers a similar unknowing cognitive bias. We don’t have data on denial reasons across states in the way we do for payment accuracy. And to some extent these are silent failures that can be written off (“oh, that person must not have cared enough to call back and reschedule.”)

I am interested to see how FNS studies this. It seems to me that a study focused mostly on the payment accuracy impacts of interviewing — interviews definitely make things more accurate! so much opportunity to clarify details!) — misses out on dollars unpaid to eligible folks. That’s a bit harder to study, but it’s analytically the more symmetrical tradeoff dimension to look at.

Also I should emphasize that I have empathy for virtually every individual involved here. While systems are not people, systems express themselves through people. More often than not those individuals are just in a run-of-the-mill pursuit of meeting Maslow’s hierarchy of needs. None of this is a story of heroes nor villains, but rather some mess of a third thing.

A related tangent: on government agency (or the lack thereof) in government agencies

Something that’s been on my mind recently is how so many people seem to have an implicit mental model of a path to effecting some change being convincing some decisionmaker in some government entity to do something.

An observation: overwhelmingly my experience with the actual people occupying seats inside government is that they feel like they can do very little.

Having worked across levels of government reinforced this. I say only with minor hyperbole that in some time at the federal level I heard frequently the rejoinder that we have very little authority over the states while at the state level I’ve heard many, many times comments to the effect of there’s no way the feds would let us get away with this.

I do think a part of this is a flavor of

Dan Davies

’s “accountability sink” paradigm — it can be convenient (politically, logistically, emotionally) to view the constraint on oneself as truly external. But also outcomes can also flail out into Very, Very Bad territory when no actor sees itself as having the authority to fix it.

And there’s also a flip to this, which is that sometimes actors that feel extraordinarily constrained very much welcome an outside who can act in alignment with the constrained actor’s goals, seemingly without the same constraints.

This kind of quiet and productive uncoordinated coordination is for me both (a) an underutilized lever in heavily constrained systems, and (b) a more common under-the-official-schema reality in such systems than we think when looking at the surface level.

Miscellaneous thoughts on my mind, which I may or may not write about in the future (psst — you can affect this!)

• There’s this interesting phenomenon — often expressed as best practices — where one takes things that have been done as somehow implicitly an indicator of these things being good. I’ve been thinking a lot about how what is lacking in such naive analysis is an understanding of whether the feedback loops X work has been subject to has in fact incentivized it to improve, or if instead it is merely Something That Happened. (I advise against following in the mold of the latter!)

• I have medium-to-high confidence at this point that the Paperwork Reduction Act is both a true legal constraint on things we should want government to do and that it has a chilling effect even through its very existence that wholesale repeal is most likely the most sensible move.

• People who complain about government chatbots miss the whole point: the biggest reason they get adopted is because they are usually a thing agency decisionmakers can do. If one’s proposed alternative is “wholesale change your hiring and organizational structure [with budget changes beyond your control]” or “make changes to that legacy system [which you have a very poor track record of doing successfully]” then one is likely just taking yelling from a moral high ground detached from reality.

• Means and ends in government in some ways start to blend together. Is the goal anything other than how this specific law is written? (Yes! Obviously! But how many people are so wedded to specific means that they strangle any space to effect changes on the ends?)

• Most disagreement about public policy tends to boil down (I think) to disagreement about what’s changeable and what’s fixed. Getting more explicit about assumptions there may be more fruitful.