3 Lessons for Scaling Mission AI: Insights from Public Sector Leaders

The pressure to adopt and show real value with AI is real. So is the gap between AI ambition and the results that drive real impact.

At Snowflake Summit, we asked public sector leaders: how do you move from the market pressure to adopt AI toward mission-scale impact? Two public sector sessions shared a consistent answer: The path to valuable AI production starts well before deploying a model. 

That means the data foundation, governance architecture and organizational alignment that make AI production possible must come first.

Below we dive into three lessons from public sector leaders on how to close the AI value gap.

Lesson 1: Build a solid data foundation first

AI doesn't fix broken data infrastructure. Rather, it exposes it.

When the University of Florida began experimenting with AI, they discovered only 10-15% of their data was semantically defined, making any insights generated from their data incomplete and therefore unreliable.¹ Their fix: The team developed a process using NaviGator AI and components within Snowflake to generate natural language definitions at scale. Had they done the work manually, they estimate it would’ve taken a human 83 weeks to complete² (Based on internal University of Florida analytics, as of June 2026).

James Martinez, Data Engineering Manager for the University of Florida says, “Now that we've built the foundational semantic layer, we’ve built a product called Navigator Insights — conversational analytics sitting in Snowflake, using the MCP server and exposing Cortex Analyst and search capabilities.” With this, the team can now ask questions in natural language and discover data trends that directly impact the students within their institution. 

AI readiness requires a solid data foundation. Data governance, security and interoperability aren't prerequisites an organization can revisit later. Rather they’re elements of the essential foundation an organization needs before they can expect value from AI innovation. 

Lesson 2: Move from isolated pilots to enterprise-grade intelligence

Most organizations aren't short on AI pilots. They're short on AI that scales.

Alan Sim holds a Ph.D. in epidemiology from Drexel University and serves as Chief Data and Analytics officer at Leidos, a defense and government services company with AI deployments spanning cyber, veterans benefits and logistics. He says, “The shift is moving from very focused, siloed one-offs to what does it mean to be enterprise-grade?”

Leidos has been using AI and machine learning for years, and their challenge hasn’t been building. It’s been getting buy-in across the organization and connecting AI capabilities to the overarching leadership priorities.

To better manage the overflow of requests for new AI pilots and tools, Sim has stopped asking teams at Leidos what tools they want and has started asking them what they’re trying to solve. He says, “Sometimes it’s as simple as saying, ‘Are you sure you need an agent? Maybe you just need automation. Maybe this is a process re-engineering activity.’" He’s found that that discipline of matching the right solution to the right problem is what separates AI pilots from real AI production that drives real value. 

Joe Larson, VP of OpenAI for Government, reinforced this sentiment from the model side. He says, "It is more compelling, and I think will deliver much higher value, to have the government be able to marry the frontier model with its data infrastructure and to create something that's more comprehensive and well-architected."

Making models available is a starting point. But making them operational by integrating them with enterprise data, workflows and governance is the real goal that will drive the most impact. 

Lesson 3: Align AI to mission outcomes, not technology metrics

Many AI projects fail mostly because they aren’t tied to tangible organizational outcomes. In the public sector, where success isn't measured in revenue but rather in impact, this challenge can be more severe. 

Stephen Moon, Snowflake's public sector field CTO says, "In the public sector, it's about mission. Are we driving mission outcomes with these AI initiatives?"

For the team at General Dynamics Mission Systems, their sales quote processing system was ingesting thousands of vendor PDFs all with varying formats, layouts and terminology. The team attempted to implement a training-based approach to combat these inconsistencies but failed. They then moved to zero-shot prompting with Snowflake Cortex AI and went from 50% accuracy to 85% accuracy in their recent go-live to production³ (Based on internal General Dynamics Mission Systems analytics, as of June 2026). The outcome wasn't measured in model performance. It was measured in speed, accuracy and cost savings on a workflow that directly affects mission delivery. 

From blueprint to action

Across every organization represented at Snowflake Summit, including General Dynamics, Leidos, OpenAI for Government and the University of Florida, a common thread emerged: Organizations that are delivering real AI outcomes build the right foundation first and then move deliberately. They take the time to set up:

• Unified, governed data before models
• Enterprise-grade use cases before scaling AI pilots
• Clear mission alignment before measuring ROI

AI technology is ready. The question for most agencies is whether the data foundation, governance architecture and organizational alignment are ready too.

Explore Snowflake's public sector resources to assess your AI readiness and build toward mission-scale outcomes.

_{¹  Snowflake Summit 26 Session IN227: Scaling AI in the Public Sector Showcase: From Pilots to Real-World Impact}

_{² Snowflake Summit 26 Session IN227: Scaling AI in the Public Sector Showcase: From Pilots to Real-World Impact}

_{³ Snowflake Summit 26 Session IN227: Scaling AI in the Public Sector Showcase: From Pilots to Real-World Impact}