Why Most Initiatives Stall and How to Avoid It
AI adoption across the U.S. public sector is growing rapidly, and for many reasons. Government agencies are being asked to do more with less while expectations for responsive public services continue to rise. At the same time, agencies are looking for new ways to increase workforce capacity and deliver better outcomes with limited resources.
Yet as AI adoption rises, governments are still building the capabilities required to govern it effectively. A 2026 report by the Organization for Economic Cooperation and Development (OECD), found that most countries in the OECD are still lagging in areas such as role-specific AI training, service user feedback, pre-deployment risk assessments, and post-deployment audits.
The same pattern is visible in the United States. A 2025 report by the U.S. Government Accountability Office found that generative AI use cases among the agencies it reviewed increased nearly ninefold between 2023 and 2024. Yet agency officials also reported challenges related to data privacy requirements, policy compliance, technical resources, and keeping governance running at the pace of AI innovation.
Government agencies aren’t shying away from AI, but the challenge is how to scale AI in ways that improve services and support mission outcomes while meeting the standards expected of the public sector. We take a closer look from our perspective at the reasons many AI initiatives fail to move from an experimentation phase, along with the foundational steps agencies can take to adopt AI successfully.
Why Public-Sector AI Initiatives Stall Before Delivering Value
Generally, what we see is projects losing momentum because organizations focus on deployment before they establish the conditions needed for long-term success.
Across the public sector, successful AI programs tend to encounter the same implementation challenges:
Technology Looking for a Problem
Sometimes the technology arrives before a clear business objective. Agencies invest in AI because of its potential, then struggle to connect it to a measurable outcome.
Example: A county government purchased a generative AI platform after seeing interest from multiple departments. Six months later, teams had experimented with content generation and research support, but no one had identified a specific service objective or metric for success.
Data Not Prepared
In other cases, the data foundation isn't ready. Information may be fragmented across systems, inconsistently managed, or difficult to access.
Example: A state agency launched an AI initiative to help employees find policy documents and procedural guidance faster. The project stalled when the team discovered information was spread across SharePoint sites, network drives, email archives, and legacy systems with inconsistent naming conventions and permissions.
No Governance Guardrails
Questions around privacy, security, procurement, compliance, and oversight often emerge after a project is already underway.
Example: A department piloted an AI assistant to help draft constituent communications. Early results were positive, but broader deployment paused when leaders realized there were no policies defining what information could be entered into the tool or who was responsible for approving new use cases.
Wrong First Use Case
Even well-governed projects can struggle if the first use case is too ambitious. Large, highly visible initiatives carry greater complexity and risk, whereas starting with a focused use case can create the momentum needed to support broader adoption.
Example: An agency's first AI initiative focused on assisting with eligibility determinations for a public benefits program. The effort quickly became mired in legal review, oversight requirements, and concerns about explainability. A later pilot focused on document summarization and knowledge retrieval delivered value much faster and helped build organizational confidence.
No Roadmap
Finally, many organizations launch pilots without a strategy for scaling them. A proof of concept is only the first step. Without a roadmap for governance, workforce adoption, funding, and operational ownership, projects can remain isolated experiments rather than becoming lasting capabilities.
Example: A pilot reduced the time required to process procurement documents by several hours per week. The project demonstrated clear value, but no plan existed for user training, governance, funding, support, or expansion to other teams. The tool remained useful within a single department but never became an enterprise capability.
Fortunately, each of these challenges points to an opportunity that agencies can improve. Organizations that successfully scale AI tend to establish five foundational elements before pursuing broader adoption.
The Five Foundations of AI Adoption Success
Many of the barriers that stall AI initiatives can be traced back to missing foundations. Organizations that successfully move from experimentation to measurable outcomes tend to follow a common sequence:
- Mission and service outcomes
Start with the problem, not the platform. Define the outcome you want to improve, whether that's reducing processing times, improving service delivery, lowering backlogs, or increasing workforce capacity.
Once outcomes are clear, agencies can assess whether the underlying information exists to support them.
- Data readiness
AI can only create value from the information available to it. Agencies should inventory critical data and establish standards for access and ownership. For many agencies, readiness also means connecting information that remains distributed across legacy applications, departmental repositories, and siloed systems.
Understanding these fundamentals helps organizations identify the right starting points for AI while reducing the risk of unreliable outputs and compliance issues. - Responsible governance
Governance is often viewed as the step that slows AI adoption, when it's what makes sustainable adoption possible. Public agencies operate under a different set of expectations than private organizations. AI outputs may be subject to public-records requests, influence decisions with legal or regulatory implications, and process information that carries privacy, security, and compliance obligations.
Establishing policies for data usage, privacy, security, risk management, and human oversight before deployment helps agencies address these responsibilities while reducing the risk of compliance issues or reputational harm. Governance provides the guardrails that allow agencies to move faster while maintaining public trust. With guardrails established, agencies can prioritize the use cases most likely to deliver value. - High-value use cases
Not all AI use cases are equal. Agencies should prioritize initiatives that deliver measurable value while carrying manageable levels of risk. That means starting with productivity-focused use cases such as document summarization, constituent self-service, knowledge discovery, or administrative support.
Higher-stakes applications such as eligibility determinations, fraud detection, forecasting, or caseworker decision support can create significant value, but they also place greater demands on data quality and oversight. As governance and data maturity improve, agencies can expand into use cases that directly influence service delivery and mission performance. Individual wins create momentum, but scaling requires a longer-term plan. - A realistic roadmap
AI adoption is more of a journey than a single deployment. Organizations that see lasting success take a crawl, walk, run approach. They start with quick wins, such as establishing acceptable-use policies, launching a measured productivity pilot, and building AI literacy across the workforce. As confidence grows, they scale successful use cases, strengthen governance processes, and address the data gaps that matter most.
Only then do they move toward applications that influence decisions and operations.
The sequence matters. The first three foundations create the conditions for success, while the last two translate that foundation into business value. When organizations skip ahead to use cases before addressing outcomes, data, or governance, progress often slows and AI initiatives struggle to scale.
Taking the Next Step in AI Implementation
AI offers an opportunity to improve service delivery and help agencies make better use of the information they already manage. Realizing those benefits requires the right foundation.
To explore these concepts in more detail, watch the recording of our webinar From AI Interest to AI Impact: A Practical Roadmap for Public Agencies, where we discuss common adoption challenges and practical strategies for building an AI-ready organization. Complete our AI Readiness Assessment at the conclusion of the webinar to benchmark your current capabilities and identify the areas that will have the greatest impact on your AI journey.
Together, these resources provide a starting point for moving from AI interest to measurable results.
/1.%20Visuals/Website/Icons/x-twitter.svg){kind=icon}