Why the Shadow Inc. App Failed

Building an app is more than shipping some code and telling people to use it. It is adding a new factor to a complex social system. It requires planning, training, and care. And yet the Shadow Inc. app used in the Iowa caucuses was built and shipped in three months.

Officials from the 68th caucus precinct overlook the results of the first referendum count during a caucus event on February 3, 2020, at Drake University in Des Moines, Iowa. Tom Brenner / Getty

The Democratic race has begun with a major technical debacle. The final results in the Iowa caucus have been delayed while data is manually verified and are, at the time of writing, only reporting 97 percent of responses. Days after Pete Buttigieg essentially declared victory, his lead on Bernie Sanders has narrowed to one-tenth of a percentage point, with some precincts and satellite sites not yet reported — satellite sites that the Sanders campaign deliberately heavily organized.

Theories have been circulating on social media since Monday. The app used to communicate results between the individual caucuses and the statewide party was produced by a privately owned company called Shadow Inc. Shadow Inc. is staffed by a number of former members of Hillary Clinton’s digital campaigning team. Shadow Inc. is partially funded by a nonprofit called Acronym. Former employees of Acronym are now senior staffers on the campaign of Pete Buttigieg. The founder of Acronym is married to a senior strategist in Buttigieg’s campaign.

Supporters of the Sanders campaign are suggesting foul play from the Democratic Party bureaucracy, and #MajorCheat trended most of Tuesday on Twitter. Meanwhile, the Iowa Democrats’ Twitter account has been tweeting out results, then quickly correcting them, only fueling further doubt.

Whatever the truth of the matter, the clearest takeaway is the revelation of how technology in general is produced and how specifically political technology is made.

The production of the app shows an insensitivity to local context, the problem at hand, and who would be using it. It was not only that the app worked poorly. Many precinct captains phoned in their results on Monday night. This caused the hotlines of the local Democratic Party to jam, which caused the slowdown in reporting and chaos. This is because volunteers at the counts had been using this phone-based methodology for decades.

As the New York Times reports, many of the volunteers are older and less comfortable with technology. They did not use the app at all, or they ditched it quickly when they encountered problems with it. They had no training on how to use the app, outside of hastily issued guidelines on the eve of the caucus. The problems seen with the app were the result of it not being tested in the field, where mobile coverage was spotty. This is a common problem with modern mobile apps. However, considering where and how it would be used, the app should have been designed with this situation in mind.

Why were these factors not considered: who would be using the app, where they would be using it, and how likely the uptake was going to be? A more minor intervention, using a phone, instant messaging, and a spreadsheet, would not have caused these problems. Though it is unclear how the decision was made, and by whom, at some point, someone decided the way in which results were entered for decades was “a problem” and that the solution was “a smartphone app.” The way in which this question is framed and the reflex to solve it with technology is indicative of the tech solutionism that pervades responses to political and social situations more widely. Maybe you don’t need an app?

Building an app is more than shipping some code and telling people to use it. It is adding a new factor to a complex social system. It requires planning, training, and care. This app was built and shipped in three months.

Technology companies are well versed in the disciplines of user research, designing for user experience (UX) and following user needs. Whole frameworks of these disciplines tell technologists to attend closely to how real people in real situations will use their work.

Those at Shadow Inc. would have been familiar with these best practices when they began building. They form part of a wider discipline, pioneered in the digital public sector, called service design. However, when budgets and timelines are tight, these are the first to go. Someone, at some point, said, “We know what to build.” They probably said, “It’s trivial” or “It’s just a form.” And this work was dropped.

The same goes for testing at scale. Few engineers have the luxury of testing their technology as rigorously as they might like. Like carefully trying to meet actual user needs instead of guessing, the patterns are well established. But when push came to shove, and crunch time hits, the product went out the door untested.

The Iowa app compounds these issues. It was distributed not through app stores but, according to a teardown by Vice’s Motherboard, through more complex tools normally used only for distributing test versions of apps. It was not ready in time to go through the processes of vetting that Apple applies to software released through its App Store. Shadow Inc. therefore used the lowest, free tier of the platform, which caps at around 200 users.

When dealing with a single-day event, there are naturally contingencies and unexpected situations that occur. These can be planned for, but only if there is thought and time.

There are important considerations about how the political economy of campaign technology operates. As this excellent Twitter thread notes, usually technologies used in campaigns are produced primarily over the course of the campaign and thrown away when the campaign is over. Campaigns and the organizations that run them shed all the contractor developers they employed in the days after the result. The institutional knowledge of technology within organizations is completely removed, only to begin again in the next electoral cycle. With few exceptions, the code itself is kept closed source for fear of enabling political rivals, even after campaigns.

There are a number of more permanent technologies, like NGP VAN or Act Blue, but they tend to be around fundraising or compliance with electoral rules. The nearer to the money, the more successful they are. Outside of this, technologies like the Action Network or Salsa do exist, but they struggle with financial sustainability. The software company NationBuilder takes one approach to solving this problem, which is, lamentably, to make technology that can be used by any political campaign, regardless of politics, including Donald Trump.

Where there is sustained work done on political technology, it is often done by teams spun out by campaigns. Here Shadow Inc. is no different: the technologists who worked on Hillary Clinton’s campaign may have wanted, in the aftermath of Trump’s election, to continue working on political problems, bound together by the intensity of a campaign and the needs of the political moment. We do not share their politics, but we understand why they would want to leave mainstream technology to make a difference.

Certainly, though, it doesn’t look good to have a team that provides technology for different campaigns to also provide the technology for counting the results. Shadow Inc. also provides software for the campaigns of presidential candidates Kirsten Gillibrand, Joe Biden, and Pete Buttigieg.

During this electoral cycle, Facebook will be used by all political candidates in all parties. It is even more opaque and wholly beholden to shareholder value as the primary measure of success. In reorienting to be more focused on “meaningful discussion” and groups, with private communication taking precedence over the public feed, we can imagine a situation where Facebook build this electoral technology themselves. They already encourage voter registration. In 2015, Microsoft built a successful mobile app for reporting the Iowa caucuses. Is this better or worse than technology built by Shadow Inc., an organization with nominal commitment to giving “permanent advantage for progressive campaigns and causes through technology”? Microsoft proudly works with ICE (Immigration and Customs Enforcement) in the United States, against the protests of their own workers.

The phrase “progressive campaigns” is a thin political description, but building technology for ICE is also a political act. Shadow Inc. used React Native and Firebase to build the app, deploying it to users with TestFairy. React Native is an open source technology first developed by Facebook. Firebase is a web application development platform owned by Google. TestFairy is hosted on Amazon Web Services. Even when not directly involved, the FAANG group of technology companies (Facebook, Apple, Amazon, Netflix, and Google) are not far away. They are not merely building the hosting infrastructure but the methodologies of software development themselves.

Microsoft worked with UX consultancy InterKnowlogy on their Iowa app. It took a year to build. Doubtless, it would have used the best research techniques available and the most thorough rollout and training, leaving nothing to chance — Microsoft has the time and resources to do this. It was bipartisan: “Some of our team was sitting with the Republican Party, others with the Democratic Party.” It was not continued in the next election, nor made open source, so it counts more as a PR experiment than contributing to permanent, well-funded democratic infrastructure.

Three technologies — Hustle, GetThru, and Spoke — that enable text messaging supporters, or allowing them to text one another, make interesting, contrasting stories.

Hustle was begun by Obama-era campaign workers. It was one of the tools used by Bernie Sanders’s 2016 presidential campaign. It raised money through traditional start-up venture capital. In its last round of investment, it raised $30 million, with money coming from Google Ventures (GV) and Insight Partners, the latter of whom are investors in Twitter, Shopify, and Tumblr. However, in early 2019, Hustle laid off forty members of staff. It had struggled, outside of electoral cycles, to find a business model that worked or, more accurately, worked in a way that venture capitalists perceived as successful. Hustle is still being extensively used during this election, by both PACs and Elizabeth Warren’s campaign.

GetThru (formerly called Relay) was a spinout of Bernie’s 2016 presidential campaign. Senior people at GetThru — Daniel Souweine, Jon Warnow, and Catherine Aronson — were all volunteers or organizers in 2016 who may well have used Hustle themselves and decided to form their own company. Taking a more modest approach to funding, combined with diversifying into serving educational institutions as well, means that GetThru is still afloat and was used by Alexandria Ocasio-Cortez’s campaign. In this election, thus far, GetThru is having limited use.

Finally, Spoke was developed by Saikat Chakrabarti and Sheena Pakanati for Bernie’s 2016 campaign as well, and it’s now maintained by MoveOn. Spoke, in contrast to Hustle and GetThru, is open source and freely available. Chakrabarti himself emerged as Alexandria Ocasio-Cortez’s chief of staff until 2019. However, this is not to naively suggest that just by making technology open source, it will be successful. It is not clear how widespread the use of the tool is, and whether campaigns can commit resources to developing, hosting, and securing their own installations of it.

In the twentieth century, social movements committed themselves to building the infrastructure needed for reproducing themselves. In the twenty-first century, they must commit to working out how to do the same in a new digital context. There are no easy answers. In examining a case like the Iowa caucuses, we can begin by asking the right questions, and also turn to considering them outside of electoral cycles.

We need to examine the issues of infrastructure more broadly: looking at the appropriateness of technology to certain political tasks, and the funding models and institutional forms that underpin them.