Artificial intelligence (AI) could unleash a productivity windfall for the U.S. government, worth up to $532 billion annually by 2028. Before this can happen, however, federal workers and executives need to be empowered to get the most from AI.

The rapid spread of AI technologies promises huge productivity gains for workers. This is especially true in the public sector, where red tape and other institutional constraints hamper productivity.

AI can empower government workers in two ways. The first is by automating repetitive tasks, freeing up time for higher value-added activities. As just one example, consider how tax compliance examiners might benefit. In much of the world, examiners’ daily routines still involve laborious steps—interviewing taxpayers, combing through mountains of tax returns, and drafting audit reports. If AI were applied to such activities, examiners could instead focus on complex, high-stakes compliance issues that made better use of their expertise.

The second way that AI can make public-sector workers more productive is by augmenting their capabilities. Historically, the ability to detect tax fraud was too often limited to analysis of existing filings. Today, the IRS is using AI to identify unforeseen connections in disparate external data, allowing the agency to target even non-filers as high-value tax cheats (In one 2019 trial run, 84% of fraudulent returns detected by an experimental AI were subsequently missed by examiners).

The U.S. government, which already spends more on AI than any other government, is keen to extend its advantage. In February 2019, President Trump unveiled the “American AI Initiative”, an executive order to scale investment in AI across federal agencies. According to market researcher IDC, U.S. government spending on AI will rise from about $250 million, in 2018, to almost $1 billion in 2023.

The return on the government’s AI investments could be enormous: productivity gains worth up to $532 billion annually by 2028, by our estimates.¹ To reap such rewards, federal agencies must begin taking actions—in the areas of training, strategy, and data—that empower workers and executives to make the most of AI.

¹ We concluded our modelling before the COVID-19 outbreak shuttered much of the global economy. Nevertheless, we believe the key insights discussed in this paper remain true.