Recoding America: Why Government Is Failing in the Digital Age and How We Can Do Better

by Jennifer Pahlka

A fifth of low-income families eligible for the earned income tax credit fail to claim it, while the wealthy have armies of lawyers and accountants looking for every opportunity to use the enormously complex tax code to their advantage.10

The Infrastructure Investment and Jobs Act, signed into law by President Biden in November 2021, provides over a trillion dollars, most of it in grants to state and local governments, but for local officials to access this capital they have to sort through almost four hundred separate grant application processes. Many are concerned that the communities that need these funds most will be the least likely to receive them. About $80 billion of the total is supposed to go to upgrading water systems, including drinking-water systems in places like Flint, Michigan, and Jackson, Mississippi, but a study of $25 billion granted for these purposes in the past decade found that small communities and communities with larger minority populations are less likely to receive assistance.11

The first principle read “Start with needs.” But there was an asterisk after “needs,” so I scrolled down to see what it referred to. “User needs, not government needs,” it said.

Uploading documents was the step that clients most needed to work, but it was the one that worked worst. So when Jake, Dave, and Alan built their site, they made it easy to upload the pay stubs and other documents near the beginning of the process.

Dave Guarino, one of the three creators of GetCalFresh, says that technocrats’ desire for a comprehensive screener is a misunderstanding of client needs. The way he recommends screening clients instead is very simple: “Made less than $X last month? You may be eligible.”5 As he explains, “Oftentimes people have an acute need that triggers them to go get that specific need met—like food. People just want to know if it’s worth their time to apply.” Given that the burden of applying for even one safety-net program, like SNAP, is often quite high, it defeats the purpose to have clients start with a process that will ask them even more questions in the hope of referring them to other programs they probably wouldn’t have the bandwidth to apply for anyway.

It’s hard to convince the welfare agencies that administer these programs to screen clients Dave’s way. You will inevitably end up with some (small) number of people who’ve been told they’re likely eligible but aren’t. The mere notion that the screening process might not be completely accurate can create a chilling effect. As Natalie Kates saw at CMS, being 100 percent thorough and technically accurate is always safer, even if it’s worse for the people the service is supposed to help. But to Dave the tradeoff is worth it. The upside is a better experience for the vast majority of clients, which ultimately results in higher enrollment. A detailed, heavyweight screener may meet an agency’s requirements, but it doesn’t meet the users’ needs.

Applications for government benefits might be completed under similarly chaotic circumstances. Users who have a lot else going on in their lives need to be able to apply for the service without an undue burden of time, technology, and cognitive overhead. If they’re asked for documentation, the documents need to be ones they have access to. If they need to correspond with the program, there has to be a way for them to do so even if they lack a stable mailing address. If they have family who are undocumented immigrants, they may need reassurance that applying for a program won’t get them or their relatives in trouble.

When services respect our time, minimize intrusion, use language we understand, and don’t make us feel stupid, that deeply broken relationship with government can start to mend.7 There’s a lot of mending needed—low trust in government erodes our ability to fight climate change, to respond to public health threats, and to maintain our national security and our democracy.

Dave Guarino and his colleagues had a creative service-design answer to the problem of screening benefit applicants for eligibility. They programmed a lightweight screener into the GetCalFresh application itself. (A similar approach can be used for multibenefit applications, such as those that let people apply simultaneously for food benefits, housing support, and childcare assistance.)

the team looked at which questions were the most likely to disqualify clients and put those at the front. Now, clients start filling out the application form and, if after the first few questions it seems like they won’t qualify, they get a message letting them know they might not want to continue.2 And for those who do seem likely to qualify, they’ve now answered those questions in the actual application rather than in some screener that will direct them to start a separate application from scratch.

Project management is the art of getting things done. Product management is deciding what to do in the first place—and

I met one IT project manager in state government who was acting quite a bit like a product manager, fighting the same kinds of fights that Natalie had. She had gone through the PMP training but hadn’t bothered to take the certification test. She didn’t find the coursework very useful. “It’s all about the planning, mostly at the beginning of the project,” she said. “And then tracking the requirements.”

projects could cost 10 percent of what they do and still provide 85 percent of the functionality.5 I hereby dub this “Byrne’s Law.” It’s not that that final 15 percent of the functionality shouldn’t ever be built—the software can and should eventually support edge cases. It’s just that trying to have it all done by launch, before you’ve had the chance to work out the kinks with the core workings of the project, will often tank the operation of the other 85 percent.

Gall wrote. “A complex system designed from scratch never works and cannot be patched up to make it work. You have to start over with a working simple system.”

(Equity policies can also make it hard to choose a phased-rollout approach. Government services are not supposed to treat different groups differently,

Teams building new products usually worry about the quality of their design and engineering, the size of the marketing budget, the strength of the sales force, and so on. They forget to worry about something much more important: whether anyone wants the product at all.

After months of work to launch a product no one had the slightest interest in using—a social network built around instant messaging—he realized he could have just bought a few hundred dollars’ worth of Google ads promoting the service and discovered that no one clicked on them. Or he could have put together a barebones mock-up of the product in a couple of weeks and tested it with a few potential users—they would have told him the same thing.

It starts from the recognition that most of the people who need to be heard won’t show up—their relationship to government went sour long ago, sometimes generations ago, as Genevieve Gaudet observed. It is government’s job to figure out who they are and go knock on their doors, literally and figuratively.

Just as it doesn’t take orders from above literally, product management doesn’t take user feedback literally either. You don’t gather a list of what your users say they want and check off that whole list, any more than you should gather a list of internal requirements and fulfill all of those. The art of product management is finding elegant ways to give users what they really need,

She was surprised to find that it worked—they could make do with just the one test environment. Over time, she began to prefer this streamlined process and the speed it offered.

Sometimes the tech teams adopt some of the trappings of agile, like having daily stand-up meetings or dividing the work into two-week sprints, and then wait for their next orders. Everyone still has requirements to check off, and it is still no one’s job to figure out the right thing to build.

Sometimes they just need to do the thing that will get the agreed-upon outcome. Success shifts from the number of orders you faithfully executed to the results you delivered.

you need to include the implementers much earlier in the process.

marijuana offenses aren’t like theft; there’s no victim who could show up and make the case that the conviction should stand. It was a given that no one would oppose, which made Cristine wonder: Why make people petition at all?

“Public defenders have to work client by client,” she told me. “The reason I became a prosecutor is because I realized in a situation like marijuana legalization, I could make a whole file cabinet disappear just by making the right policy decision.”

Her office had access to the database that held the RAP sheets, as well as to the county’s court management system. It basically had the information needed to build a petition right there in its computer systems. Why couldn’t the DA just file petitions on behalf of those who were eligible?

Nationally, as many as 1.9 million workers are estimated to be shut out of the labor market because of criminal histories, costing the US economy up to $87 billion each year in lost GDP.1

the new software program, called Clear My Record Automatic, identified 8,132 San Franciscans whose records were eligible to be cleared. The DA’s office filed a single petition for expungement with an attachment listing those 8,132 names as the petitioners, and the court agreed to render one judgment for all 8,132 based on the logic in the software. Instead of decades of work, the software program had taken about thirty seconds to run.

“Nothing should be petition-based!”

The only way to establish that any given conviction meets the criteria for expungement under Prop 47 is through laborious research performed by human beings. The way the law was written makes automated expungement impossible.

Tom, who now works with the global transformation consultancy Public Digital, has now seen that realization occur many times. “Why,” he asks, “is so much policy educated guesswork with a feedback loop measured in years?” It doesn’t have to be.

Oracle sees digital talent exercising judgment instead of blindly following process as a threat to its business.

Few in Silicon Valley (by which I mean an industry and a mindset, not a geographic location) see building government capacity as worthwhile—or perhaps even possible—placing all their bets on what the private sector can achieve.

The skills most needed in government are good product management and service design. The work is hard not because the tech is complicated but because the environment is.

Cecilia Muñoz, a recipient of a MacArthur “genius grant” for her work on immigrant rights, who served as the head of Obama’s Domestic Policy Council.

“We need to bring technologists into all levels of government, so digital practices and concerns are built into policy decisions from the beginning. We need technologists to apply user-centered design to policy and put citizens and their needs at the center of the policy-making process.” It turns out that if the tech people you bring in spend time understanding the problem instead of trying to apply predetermined solutions, government actually buys less tech, not more, and at a lower cost, while delivering better outcomes.

California launched a site that let you enter your personal information, looked you up in the state vaccine registry, and sent you a QR code you could keep on your phone to prove you’d been vaccinated. As with covidtests.gov, the process took between a few seconds and a few minutes. The team also decided to make the code freely available, and other US states and other countries have borrowed

Services like the digital vaccination card and covidtests.gov threaten companies like Oracle not just because they leverage more in-house tech talent. They’re a threat because of how small they are. The combined USPS/USDS team got covidtests.gov ready in just three weeks. The initial vaccine record app took six weeks.

Ukraine built a multipurpose mobile app for its citizens that lets them access government records and various social benefit services. During the pandemic, the government offered financial support for those who got vaccinated, and it easily paid out the benefit within the app, checking each applicant’s vaccination status automatically. When Russia invaded, being able to prove your Ukrainian citizenship without paper documents suddenly became a lifesaver for hundreds of thousands of refugees, and support became critical again.

tax prep is all but automatic in many countries, including most of Europe, Japan, and New Zealand.

We spend $10 billion on tax preparation services like H&R Block and an additional $2 billion on tax preparation software like Intuit’s TurboTax.18

the situation might be changing, thanks to the recent demise of a long-standing agreement between the IRS and a consortium of companies called the Free File Alliance. Under that agreement, the companies had promised to provide free tax filing for people with low incomes, and in return the IRS promised not to build anything that would compete with consumer software products like TurboTax. In 2019, when investigative journalists revealed that the companies were intentionally hiding their free option—and in fact tricking people who would have qualified for it into using a paid product instead—the conversation started to change.19 The consortium eventually fell apart, opening the door, theoretically, for the US to catch up with its peer countries and make things much easier for the public. The Inflation Reduction Act of 2022 includes $15 million for the IRS to study how it might implement such a system.

There are good reasons some services, like felony expungements and pandemic food benefits, should be delivered automatically while others, like COVID tests, should not. For one, when COVID tests are in short supply, sending them to people who wouldn’t use them keeps them out of the hands of those who will. (Indeed, it’s likely that a portion of the population would actively resent being sent COVID tests.) For another, the burden of “applying” for these tests was minimal, and there is value in reminding the public of the government’s role. Government could use some good PR.

other government entities could also build on one another’s work. “Other agencies already rely on the IRS to verify applicant information,” she points out. And the IRS already has data on low-income families who claim the earned income tax credit and the child tax credit.

Product management is key to avoiding the trap of “find all the requirements, fulfill all the requirements,” but savvy vendors can and do provide excellent product managers—as long as there is an empowered and engaged “product owner” within government who can work with them to decide on and authorize smart tradeoffs.

How do you break down barriers between policy and implementation teams so that you don’t leave programmers guessing about policy intent?

And we have tried policy changes, though some would say not enough. We’ve increased the power of agency CIOs,5 added alternate purchasing paths that are not subject to burdensome acquisition regulations,6 introduced a “technology exchange program” for government staff to spend a year at tech companies to learn their practices,7 and so forth. The latter two programs have gone largely unused, victims of a culture in which if even one person prefers safe over sorry, we’re back to “the way it’s always been done.”

The USDS tallied up $3.5 billion in savings and cost avoidance (that is, helping agencies find better, less expensive ways to address their needs) in 2019 alone. Given the unit’s relatively small budget, that came out to a seventeenfold return on investment.11 More importantly, it helped create less-intrusive government interactions and a lower regulatory burden. It’s a government that feels smaller but gets more done.

the way those contracts are structured also needs to change. What we need is not more megaprojects but incremental, stage-based funding.

VCs fund startups to start small and learn quickly, then invest more when the startups get traction and show value. (That’s the underlying concept, at least; in recent years the venture world has had more money than it knows what to do with, so funding has been less disciplined.) Incremental, value-based funding pairs perfectly with small, cross-functional teams that practice agile development and focus on understanding and meeting user needs.