There Are High-Value, Low-Risk AI Opportunities for Government

Introducing AI into government operations comes down to a fundamental calculus: risk and reward. Government leaders are seeking out high reward, low risk ways to implement AI into their agencies. And with good reason — a Harvard Business School study showed AI users completed tasks more than 25% faster with more than 40% higher quality. 

The assumption that every AI application with high potential for impact must also carry higher risk isn't a scientific argument. It's an extrapolation of a truism, and it can be overturned by getting specific about AI capabilities and their deployment context.

The risk posed by foundational AI models can essentially be boiled down to two areas.

One: These models sometimes provide incorrect or made up information or take incorrect actions, frequently referred to as “hallucinations.” On the lower end, OpenAI’s GPT 5.4 has a hallucination rate of 3.1%, meaning it’s factually accurate 96.9% of the time. xAI’s Grok 4.1 model, on the other side, has a hallucination rate between 17.8% and 19.2% of the time. 

For some aspects of government operations, a 3.1% risk of error is not worth the efficiency gains AI brings. Few people would accept more veterans healthcare claims being processed if it meant nearly 3% of veterans didn’t get benefits they were due. If the IRS could process more tax returns, but 3% of Americans — about 10 million people — had to pay the IRS more than they owed because of an AI-driven error, that is not a compromise we should be willing to make. 

Two: AI models that ingest personally identifiable information of constituents trigger — rightly — data governance and policy challenges for many agencies. Who “owns” PII once it’s fed into an AI model? Is an AI provider using PII to train its models? Who is responsible for mistakes when PII is used? These are questions worth answering. These governance realities mean that AI's ingestion of PII naturally slows the pace of adoption—and for good reason.

High-value, low-risk workflows

Agency leaders can bifurcate applications of AI into external use cases like those articulated above — AI that is directly involved in service delivery to constituents, where risk of errors and misuse of PII is relatively high — and internal use cases.

We can think of internal use cases as “the work to do the work,” meaning the essential functions that are necessary before a policy or program reaches the public, or the coordination work inside government operations before constituent-facing activities happen.

While state and local governments have varied responsibilities — from public health programs to emergency response, from supporting people experiencing homelessness to managing elections — they all require coordination. One agency’s scope of responsibility may intersect with dozens of institutions, including other state and local government agencies, the federal government, and non-governmental partners like non-profits. 

This coordination work requires agencies to continuously:

• Plan and map out an ecosystem of personnel needed to achieve a mission
• Communicate across a geographically and organizationally dispersed group of people with updates, information, resources, and guidance
• Work together with their partners
• Organize cross-agency, historical and real-time information
• Measure the efficacy of initiatives to inform future work

This coordination work represents a low-risk, high-impact opportunity to apply AI to help public servants work better and faster together.

Workflows ripe for AI

Proactive resource management

AI can be used to automate the end-to-end workflows to collect data from government agencies on the front lines to help agencies proactively plan what resources they need. Local offices of emergency management (OEM) are a prime example: AI can automate the initiation of a poll (or other method through which it can gather structured data) about a standard set of materials that partners in a region have — think fuel, batteries, blankets, and sandbags. 

After autonomously collecting data, AI can identify which organizations haven’t responded and compile the responses into a weekly summary, enabling decision-makers at HQ to identify gaps. 

The risk of hallucination is low because the AI is initiating the collection and analysis of data provided by people. Similarly, there’s no PII involved. But it’s automating the work to do the work — in this case, ensuring the OEM and its partners have the requisite materials to respond to an emergency if and when it occurs (i.e., the work). 

Information gathering and synthesis

AI can accelerate the recurring, labor-intensive process of assembling information from across an agency's operational landscape into coherent collateral. Take state public health departments: They have to track disease spread, understand local healthcare provider capacity, and manage county- and city-level data from different siloed systems and agencies regularly. 

AI can collate this data — not unlike the step after collection for OEMs — into salient updates from local health districts, flagging missing or incomplete information, drawing attention to anomalous results, and supporting the overall delivery of public health response. Here, too, risk is relatively low. This process can be functionally automated without patient information, and the output is internal, reviewed by public health personnel before a decision is made and constituents engage with the department. 

Cross-agency coordination

Public servants will tell you that coordinating initiatives across multiple government agencies and non-governmental partners like non-profits across different jurisdictions is a significant amount of manual work. This burden is felt acutely at organizations like county or state housing and homelessness offices. 

Such programs, like much of government services, are exceedingly cross-functional. Housing authorities, school districts, public health departments, homelessness shelters, and even law enforcement are often involved in helping people find temporary shelter and long-term housing. AI can draft routine coordination information — for example, partner updates or meeting summaries — ensuring the most vital updates don’t fall between the cracks of a collaborative, multi-agency network. 

The risk posed by AI here is manageable. AI is drafting and distributing information to help public servants and other officials stay aligned. It’s not making decisions. And government employees — rather than spending time keeping everyone on the same page — can stay focused on the work itself, helping people experiencing homelessness get back on their feet. 

Simplicity is key

Many AI-supported workflows can feel daunting because implementation is overly tech-heavy or the workflow being automated is sensitive and complex. There is a cognitive burden associated with automating overly complex workflows, and that makes AI adoption harder. 

Starting with internal tasks makes safe and effective AI implementation manageable. And the results are tangible: Public servants spend less time on the work to do the work, and more time on the actual work. Their quality of life goes up. The quality of service delivery for constituents, too, goes up.

In the age of AI, win-wins are rare. But not impossible.

See what AI built for government can do — Reach out.