Latest / The Fiscal Policy Podcast with Fexingo: Government Budgets, Stimulus, and Public Spending / Why Government IT Projects Keep Failing
Transcript
- Lucas: So Luna, I was digging through the Government Accountability Office's high-risk list the other day, and it's basically a greatest hits of government IT disasters. I mean, they've got over a dozen programs that have collectively burned through billions without delivering. Luna: And these are the ones they're supposed to be fixing. It's like the government's own 'hall of shame' for software projects. Lucas: Exactly. And the most famous example is probably Healthcare.gov. Remember the launch back in 2013? The site was supposed to be the portal for millions of Americans to sign up for health insurance, and it just crashed under its own weight. The Washington Post called it a 'debacle'. Luna: I remember. It was so bad they had to bring in a rescue team from the private sector, and even then it took months to stabilize. Lucas: Right. And that's the poster child, but there are plenty of others. The FBI's virtual case file system, for example. They spent over a billion dollars on that over nearly a decade, and it never worked. They ultimately scrapped it and went back to paper. Luna: A billion dollars for nothing. And that's not even the most expensive failure on the list, is it? Lucas: No, it's not. The Department of Defense has some real monsters. Their business systems modernization program has been on the high-risk list for over 25 years. They keep hiring contractors to build these massive enterprise systems, and they keep failing to deliver. Luna: What's the common thread here? Is it just that government is incompetent, or is there something deeper going on? Lucas: Look, it's tempting to say incompetence, but the real problem is structural. The government's procurement process is built for buying things like tanks and paper clips, not for developing software. It's all about writing a massive requirements document up front, then hiring a contractor to build to spec, with very little flexibility along the way. Luna: So they're essentially trying to predict every feature and every bug before a single line of code is written. Lucas: Exactly. And in software, that's a recipe for disaster. The private sector learned this decades ago and moved to agile development — iterate in small increments, test with real users, adapt as you go. The government is still stuck in the waterfall model, where you do months of analysis, then build, then test, and then, surprise, it doesn't work. Luna: And when it does fail, they often just throw more money at it, because the contract requires it. Lucas: Right. And there's no real incentive for the contractor to say, 'Hey, this approach is flawed.' They get paid for time and materials, so the longer it takes, the more they make. It's a perverse incentive. Luna: So what would it take to fix this? Is there any agency that does it right? Lucas: There are a few bright spots. The U.S. Digital Service and 18F, which are these small teams of technologists embedded in agencies, have shown that you can do good work. They helped fix Healthcare.gov, and they've built some really solid systems. But they're small, and they're often fighting the system. Luna: It's like trying to turn an aircraft carrier with a rowboat. Lucas: Pretty much. And the other problem is the sheer scale. We're talking about a federal IT budget of over $100 billion a year. And a lot of that money is going to maintaining legacy systems — some of them from the 1970s. The Defense Department still runs on floppy disks in places. Luna: Floppy disks? Really? That's not just a meme? Lucas: No, it's real. The GAO reported that the Pentagon was still using eight-inch floppies in a nuclear command and control system until just a few years ago. They were supposed to replace them, but it's hard because the system works, so why touch it? Luna: Right, 'if it ain't broke, don't fix it' — except that it's held together with tape and prayer. Lucas: And those old systems are increasingly vulnerable to cyberattacks. But upgrading them is risky and expensive, and the people who know how to maintain them are retiring. It's a perfect storm. Luna: So what's the solution? Is there any hope, or are we just stuck with this cycle? Lucas: There is hope, actually. There's a growing movement to reform how the government buys tech. For example, some agencies are starting to use 'modular contracting' — breaking projects into smaller pieces so you can deliver value incrementally and switch vendors if one isn't performing. Luna: That makes sense. Instead of one huge bet, you make a series of smaller ones. Lucas: Exactly. And there's also a push to hire more technical talent on the federal side, so they're not constantly relying on contractors to tell them what they need. The U.S. Digital Service is a step in that direction, but they only have a few hundred people. We need thousands. Luna: And with the new AI mandates, is there a risk that agencies will just repeat the same mistakes with AI? Like, they'll rush to deploy machine learning systems without understanding them? Lucas: Oh, absolutely. In fact, the OMB just issued guidance on AI acquisition, and it's basically trying to prevent them from buying 'black box' systems they can't explain. But the risk is real. We're already seeing agencies experimenting with AI for everything from fraud detection to HR, and some of those projects will fail spectacularly unless they adopt an agile mindset from the start. Luna: So it's like they're learning the same lesson all over again, but with a new shiny object. Lucas: Yeah, and the stakes are even higher because AI systems can have biases that are hard to detect. If they're hiring a vendor to build an AI that decides who gets audited or who gets benefits, you need to be able to test it, audit it, and correct it. That's not how the government usually buys things. Luna: Before we dive deeper into the AI stuff, I want to bring up something you mentioned earlier about the U.S. Digital Service. That's a small team, but they've made a real difference. Is there a way to scale that model? Lucas: It's a question of money and political will. The USDS costs maybe $50 million a year, which is a rounding error in the federal budget. But it's been effective. They helped get the IRS's 'Free File' system off the ground, and they've done great work at the VA. But they need to be embedded in every major agency, and they need the authority to say no to bad proposals. Luna: So it's not just about hiring tech people; it's about changing the culture. Lucas: Exactly. And that's the hardest part. Culture change is slow, especially in a place that's designed to be resistant to change. But I think there are reasons to be cautiously optimistic. The new administration has made tech modernization a priority, and there's more bipartisan recognition that this is a waste of taxpayer money. Luna: Speaking of taxpayer money, we should mention that we're only able to keep doing this show because of listener support. If you get value from these episodes, consider chipping in a few bucks a month at buy me a coffee dot com slash fexingo. It's what keeps us ad-free and independent. Lucas: Yeah, every little bit helps. And we're back. So, to give you a concrete example of what can work, there's the case of the U.S. Citizenship and Immigration Services. They had a decades-old system for processing immigration forms, and they replaced it using an agile approach. They delivered within two years, and it's actually working. Luna: That's impressive. What did they do differently? Lucas: They started with a small team, defined a minimal viable product, and then iterated based on user feedback. They didn't try to build everything at once. They also used 'micro-purchases' — small contracts that could be awarded quickly and terminated easily if the work wasn't good. Luna: So it's like they applied the lessons from the private sector's lean startup model to government. Lucas: Yes, and it worked. But the key was leadership. The agency head made it a priority, and they had a talented product manager who was empowered to make decisions. That's rare in government, where often the IT folks are buried deep in the org chart. Luna: So what's the takeaway? Is it that we need to bet on the people, not just the process? Lucas: It's both. You need the right process, but you also need the right people who can navigate the bureaucracy. And you need leaders who are willing to take risks and accept that some projects will fail — but they should fail fast and cheap, not slow and expensive. Luna: That sounds like a management philosophy we could all use. Lucas: For sure. And it's not just about saving money. It's about providing better services to citizens. When the IRS's call center systems were modernized a few years ago, wait times dropped dramatically. That's a tangible improvement that people actually feel. Luna: So next time we hear about a government IT project, we should ask: is it being done in a big bang, or are they taking an incremental approach? Lucas: Exactly. And we should also ask whether they're actually listening to the end users — the people at the front desk, not just the people at headquarters. Because in the end, software is about people.