Avoiding common pitfalls in public sector AI adoption

Public sector organisations are in an AI race, striving to maximise the value of their investments and unlock their potential to deliver on mission outcomes. These investments, however, often fail to deliver ROI due to a narrow focus on short-term efficiency or productivity gains and ill-defined goals, exposing them to backlash from political leaders.

According to a recent Gartner survey, 74% of government CIOs have already deployed or plan to deploy AI within the next 12 months, with 80% indicating they will increase their investment in AI.

Despite this momentum, many government CIOs still approach AI as a technology deployment problem — ‘pick a platform, run pilots’ or ‘deploy an AI tool and see what people use it for’ — rather than as a longer-term ecosystem-building challenge. This often leads to reactive strategies, shallow skills and capability, and increased risk of under-delivering.

In contrast, those who embrace an ecosystem approach are better positioned for success; that is, where AI-ready data, platforms, models, protocols, hardware, and AI-ready workforce are managed together to deploy AI solutions and deliver outcomes.

Gartner research shows that across all industries, organisations with robust data and analytics foundations and an agile AI strategy will achieve twice the impact in the enterprise AI race compared with those that don’t.

Achieving this level of impact, however, isn’t just about technology choices or operational improvements. AI adoption also requires navigating complex political cycles and meeting the governance and control expectations of the community — factors that can significantly influence how quickly and effectively solutions can scale.

In this environment of high expectations and unique challenges, public sector leaders must recognise — and avoid — the most common pitfalls that can undermine efforts to realise meaningful value from AI investments.

Overlooking AI’s evolving nature in contracts

Many public sector organisations approach AI procurement as if it’s a static service, expecting vendors to deliver and maintain solutions within basic service-level agreements. Metrics such as uptime and support response overlook the fact that AI solutions require continuous monitoring and updating due to their dynamic nature and evolving regulatory requirements.

Engaging vendors in this manner can result in degraded performance, reduced transparency, loss of sovereignty or data privacy, and even vendor lock-in.

AI contracts must specify model lifecycle needs, such as performance thresholds, retraining requirements, transparency measures and sovereignty protections. This leaves public sector organisations in a strong position to maximise value and maintain trust in their AI solutions.

Limited AI expertise

Talent is an area where the size and ambitions of a public sector organisation will impact the challenges it faces and the approaches needed to overcome them.

Many public sector organisations have workforces that use AI platforms as consumers but lack deeper AI literacy for government use. Without these skills, governments can’t effectively evaluate or govern their AI solutions, increasing risks of poor performance, bias or drift.

A lack of AI literacy across the organisation makes it difficult for employees to act as the discerning user that is needed to be an effective human in the loop — knowing when to trust AI’s recommendations and when to challenge them.

Insufficient internal talent also means they must rely heavily on vendors, increasing the risk of vendor lock-in and reducing control over mission-critical AI.

Developing an AI talent plan that builds technical AI literacy across the organisation supports the ability to govern, adapt and secure AI models. This can facilitate the rapid development of not only AI talent across the organisation, but also with related agency partners. It also avoids traditional silos of activity in government, developing cohorts of learners and practitioners in centres of excellence.

Governance as an afterthought

Governance in government agencies is often complicated by hierarchical decision-making, siloed funding streams and disparate agency missions.

The use of siloed data or technology governance to govern AI fails to capture all of the risks and opportunities. It can also be inadequate to address the exploratory nature of AI. Without early, integrated governance, it can be harder to identify bias within AI models or assign responsibility for outcomes. This is particularly of concern to procurement, vendor oversight and ongoing operations. Public sector adoption of AI will be subject to internal and independent audit to ensure they can evidence their responsible use of AI. Governance will be a critical part of the audit process.

Gartner predicts at least 80% of governments will have their adoption and ongoing monitoring of AI independently audited by 2028.

Enterprise governance committees must be upskilled on AI risks and opportunities, and dedicated AI governance must be established to align with whole-of-government regulation and policies. It is also important to educate teams and end users across the organisation about policies and escalation paths as new use cases emerge.

Fragmented AI roadmaps

Many AI roadmaps in public sector organisations are reactive or piecemeal. They are often focused on short-term pilots or tools, with little clarity on buy-versus-build decisions, or how AI ties into broader mission and data strategies.

Without a clear strategic view for using AI, initiatives are often short-term or siloed, lacking integration with AI-ready data, modern data management and governance, and workforce upskilling plans. Buy-versus-build blind spots also arise, leading to overreliance on commercial solutions and missed opportunities to develop sovereign or mission-specific capabilities.

The focus must be shifted from isolated AI success stories to use cases that directly support mission objectives. Multilayered AI roadmaps that balance immediate productivity gains with long-term strategic goals must also be developed, prioritising enterprise-wide capabilities over standalone, redundant ones.

*Dean Lacheca is a VP analyst at Gartner, focused on supporting public sector CIOs and technology leaders on the transition to digital government and the potential impact of AI.

Image credit: iStock.com/wildpixel