A practical blueprint for AI transformation in the public sector

With use cases increasing 105% in a single year, it goes without saying that federal artificial intelligence adoption is accelerating rapidly. The significant increase signals that the government has moved beyond experimentation and into real-world implementation.

Public sector leaders no longer struggle to understand what AI can do; that phase of the conversation is largely over. Today, the challenge is purely operational: How do agencies deliver on mission requirements amid dwindling headcount, compressed timelines and increasingly complex environments?

Across the federal landscape, pressure is reaching a breaking point. While workforce capacity has fluctuated, the mission has not slowed. Leaders don’t have the luxury of scaling back objectives simply because they have fewer people. In this environment, AI is no longer evaluated as a speculative "innovation" initiative; it is considered a fundamental tool for capacity development.

The conversations I have with agency leaders today reflect this shift. The question is no longer "Should we use AI?" but rather, "How do I maintain service levels with 30% fewer staff and half the time?"

We are already seeing this play out in real time. In one recent case, a major civilian agency facing a sharp reduction in back-office personnel moved quickly to pilot AI-driven support capabilities to help absorb the workload.

The real constraint: operational reality, not technology

Expectations have shifted from multi-year transformation cycles to "months-to-value." Leaders need to see impact where friction is highest — specifically in IT and HR, where roughly 66% of all employee requests originate. These are natural starting points where AI can provide immediate relief to an overburdened workforce.

However, a significant hurdle remains: legacy systems. In many mission areas, the gating factor isn't the sophistication of the AI model; it’s the underlying technology stack. You cannot effectively apply modern AI to a fragmented, archaic environment. To move forward, we must rethink the systems that support how work gets done. 

A practical blueprint: systems over models

This is where many AI initiatives hit a wall. Organizations often treat AI as a "model problem" — something to test and pilot in a vacuum. But real transformation does not come from models alone; it comes from systems.

You cannot simply "force" your way through an AI transformation by layering a model on top of a broken process. Real outcomes require a strategic marriage of two distinct capabilities:

• Probabilistic AI (The Models): providing the judgment, prediction and natural language processing necessary to understand intent. 
• Deterministic Workflows (The Execution): ensuring compliance, repeatability and the actual completion of a task.

AI can recommend the next best action. But workflows are what actually complete the work — ensuring consistency, compliance and accountability.

From insight to execution: embedding AI into government workflows

In the public sector, "close enough" is never an acceptable standard. Government systems must follow strict policy, maintain airtight audit trails and produce consistent outcomes. AI might suggest the "next best action," but without a deterministic workflow to carry that action through to completion, it is merely expensive advice.

The agencies making the most significant progress are moving away from fragmented experimentation. Instead, they are taking an integrated approach: connecting AI models directly to existing data environments (like established data lakes) and embedding them into the workflows where work actually happens.

By building governance into the architecture from day one, these agencies are moving faster than those working in silos. They aren't just "doing AI" — they are integrating AI into their core operating model.

The path forward

The transition from pilot to production requires us to stop viewing AI as a standalone miracle and start viewing it as the engine within a larger machine. When models, data, workflows and governance work in concert, "potential" finally turns into "impact."

The path forward is not about adopting AI in isolation; it is about integrating it into the systems that support the mission. When done well, AI does more than improve efficiency — it helps agencies sustain performance, even under constraint.

Mike Hurt is group vice president for the U.S. public sector at ServiceNow.