AI moves at breakneck speed, government operating models don’t. Here’s how to close the gap.

Government doesn’t have an artificial intelligence adoption problem; it has an AI endurance problem, with pilot after promising pilot fading out before it can make agencies meaningfully faster and impactful for the people it serves.

New research from Adobe’s latest State of AI Readiness in the Public Sector report found that roughly 90% of proposed public sector AI use cases never reach actual production.

The study highlights a fundamental disconnect in government as agencies are investing heavily in AI, however, their operating models often haven’t evolved enough to support it. For many agencies, scaling AI depends less on the tools they adopt and more on how effectively they align their people, processes and decision-making around its implementation.

Today’s Piloting and Scaling Gap

Of the 502 active public-sector respondents, 62% say they are currently in the piloting or deploying stages of AI adoption, underscoring the fact that many agencies have indeed moved beyond exploration, yet remain far from expanding AI across their organizations.

For state and local governments, a successful pilot may demonstrate that a tool can summarize documents, streamline internal workflows or help employees find information. But scaling that success requires integrating the technology into everyday operations, establishing consistent oversight and demonstrating measurable value.

Agencies must establish these key foundations to avoid AI pilots that, while promising, are prone to stall as they move through IT reviews, security assessments and department priorities. While each review may be necessary, unclear responsibilities and decision-making authority make the path from pilot to production difficult to navigate.

Avoid the Silo Pitfall

For state and local agencies, cross-functional collaboration is king. Breaking down silos between business and tech teams is critical to successfully scaling AI pilots. However, 92% of active public-sector respondents described the relationship between business functions and IT as siloed or transactional.

Program teams, technology leaders and oversight functions must work collaboratively rather than sequentially to avoid any issues involving security, privacy and accessibility surfacing late in development and forcing costly revisions or halting projects altogether. 

Agencies should start with one AI use case tied to a clear citizen need. From there, bring all necessary program stakeholders, including IT, security, privacy and accessibility teams into the process, while the solution is still being designed, to identify early risks and align on priorities.

Define Impact Baselines

The study shows that only 4% of public sector respondents reported comprehensive measurement of AI’s impact.

Tracking usage or time saved statistics can be useful, but agencies must be able to define metrics that further establish whether a service has improved, whether employees can serve residents more effectively and whether the investment delivers meaningful public value.

Defining the intended outcome and establishing a baseline against which results can be measured can more effectively convey to state leaders whether an AI initiative should expand, be revised or stop.

Designate a Program Owner

As public sector organizations look to scale AI responsibly, clear ownership and decision-making authority are essential to turning promising pilots into lasting, organization-wide impact.

Just 7% of public sector respondents reported having a formal AI owner with a defined title and budget authority. Committees and informal working groups provide valuable guidance, but they may lack the authority to resolve competing priorities, allocate resources or make final decisions about expanding a successful use case and vice versa. Scaling requires clear, accountable leadership that can bring stakeholders together and move decisions forward.

Likewise, agencies should create an environment where teams can share lessons from unsuccessful experiments. Properly documenting what didn’t work helps prevent recurring mistakes and gives other departments a smoother path to success. Internal champions who demonstrate practical results can further encourage adoption by showing colleagues how AI fits into their daily practices.

True government AI readiness will be measured by an agency’s ability to turn promising AI use cases into reliable and repeatable outcomes across departments. As AI initiatives continue to evolve at breakneck speed, agencies that establish clear ownership, collaborative governance and meaningful performance measures will be better positioned to adapt. And that will make AI useful in ways that truly improve how government works for the people it serves.

Brian Chidester is Global Head of Industry Strategy for Public Sector at Adobe.