Identifying what government services, functions and tasks AI will transform first, and how deeply, is crucial and urgent for public decision-makers. It is also one of the hardest questions to answer.

After thinking hard about this, I found a candidate ranking high: citizen services.

In the business sector, customer services are already one of the categories most impacted by generative AI. This is a quote from Eoghan McCabe, co-founder of Intercom, one of the leading customer service companies whose AI agent, Fin, is topping the evals in the category:

[Customer] Service is arguably one of only two or three categories where generative AI has thus far had a material commercial, economic, real world impact. The other being coding, and perhaps the other being the legal industry. The TAM [total addressable market] in each of these spaces is insanely large ($250B-$1T?) and as a result they are hotly contested by multiple companies, which have been aggressively capitalized.

If the potential is clear in the private sector, it’s at least as compelling in government. But whether AI actually transforms citizen services or becomes another overpromised tech deployment depends on the design choices governments make now.

Why citizen services are a good fit for AI in government

There are several reasons why this potential impact can translate to government-run citizen services.

First, giving citizens information, assisting them with services and handling requests is a large segment of what governments, particularly local ones, do.

And the numbers are staggering.

In New York City alone, the 311 system handled 38.2 million contacts in fiscal year 2024 across phone, web, mobile, text, and social media.

These interactions are not only high in volume, but also relatively repetitive. This is why rules-based chatbots were already widely adopted in government citizen services, providing 24/7 availability and automating many of these interactions.

But traditional chatbots had clear limitations in addressing questions or interactions that did not directly fit into their constrained option menu. We’ve all been trapped in a chatbot that can’t parse a nuanced question.

That has now changed with LLMs. Interactions with the new generation of AI agents are faster, more accurate, and - occasionally - genuinely impressive.

Widespread use of generative AI products like ChatGPT, Gemini or Claude has profoundly changed our perception of what it is like to interact with AI agents, and this shift in citizen attitudes towards AI matters.

Citizens across age groups and backgrounds are now much more open to engaging with AI agents, though this doesn’t mean that everyone will use these services equally. Younger and digitally native citizens will find it more natural to use these speedier and more convenient services, and bridging that gap for other groups must be a design priority. With advances in voice-modes and the integration of these AI capabilities across channels (phone, WhatsApp, websites, etc.), though, these tools can now reach a wider share of the population.

Another important change is that AI agents don’t just provide information. They can already trigger actions, use tools and, in some cases, handle a process end-to-end.

This is still frontier territory — though advancing rapidly and attracting more players by the month — and there will always be complex and edge cases that need to be handed over to humans. According to Intercom, their AI Agent “Fin” achieved an average 67% resolution rate, with its mean resolution rates improving 1% every month. That’s impressive for the commercial sector, but may not be sufficient for government services where inaccurate information can have legal consequences. Yet, given the rate at which these technologies are improving, we can expect confidence levels to reach acceptable thresholds for the most repetitive interactions.

Beyond efficiency: improving trust, equity, and government performance

Making millions of citizen interactions available at all times and with faster resolution rates brings clear efficiency gains. But those are not the only benefits. Having a responsive, effective and smooth experience when asking for information or trying to book an appointment will likely increase citizen trust in that government too. It’s also a good way to reduce widespread citizen dissatisfaction.

The OECD’s 2024 Survey on Drivers of Trust in Public Institutions, based on nearly 60,000 responses across 30 countries, found that satisfaction with day to day interactions is one of the key drivers of trust in the civil service. The report explicitly notes that “improving the speed and ease of administrative service delivery” would contribute to trust, especially in civil servants and local government. In a context where 44% of OECD citizens report low or no trust in their national government, the quality of routine interactions is not a minor issue.

A clunky chatbot that can’t handle a straightforward complaint about an unfair fee reinforces the suspicion that government doesn’t care. Closing the gap between how citizens experience private-sector service and how they interact with government could be one of the most powerful trust-building moves available.

It’s also an area where many governments have a good deal of experience with innovations. The 311 services were a crucial new way to handle citizen interactions. They have brought the government closer to citizens, and yielded additional and valuable spillovers. For example, all the data generated by NYC’s 311 system has helped the city understand what happens in its neighborhoods, allocate resources, and improve services. It can also enhance transparency: the NY Office of the State Comptroller recently launched an interactive 311 Monitoring Tool to track complaint trends by neighborhood.

Academic researchers are using this open data to investigate key urban governance questions too. My colleagues at the Bloomberg Harvard City Leadership Initiative, for example, recently published an interesting article using 311 and 911 data to better understand how place-based interventions in Buffalo, NY, impacted citizens’ engagement levels with government departments.

Despite the potential of generative AI to transform citizen services, three serious objections deserve engagement. First, equity: AI agents risk widening the digital divide if they replace rather than supplement existing channels. One answer is multichannel deployment. Second, accuracy: a wrong chatbot answer about building permits or benefit eligibility carries legal consequences that a wrong product recommendation does not, so governments need clear escalation and accountability protocols. Third, labor: if AI handles the majority of routine queries, current workforce must be explicitly transitioned toward complex cases requiring judgment and empathy.

Governments thus need to treat equity, accuracy, and workforce impact as first-order design requirements. If they are conceived and implemented thoughtfully, however, the next wave of AI-powered citizen service agents can become a tool to significantly improve government performance.

Critical design choices for governments

A strategic, conscious and careful implementation of AI in citizen services demands more than what this post can deliver, but here are some key design choices that any government thinking about leveraging these tools needs to consider:

1. Channel coverage: What is the channel mix (WhatsApp, phone, website, in-person kiosks) needed to reach the full range of citizens, including those who are not digitally native?

2. Text, voice, or both: Voice-mode AI is advancing rapidly and can close digital gaps, but deploying it in a government context, where accuracy has legal consequences, raises different stakes than a commercial helpdesk.

3. Platform readiness: Can existing citizen service platforms and tools incorporate top-performing agentic capabilities and products? If not, what internal changes (data, integrations, training) are needed to roll it out?

4. Pricing and procurement: Some of the most innovative AI service companies are developing new business models where you only pay for resolved queries. This may be appealing but can escalate costs quickly and unpredictably, a hard sell in fiscally constrained governments with strict budget planning and execution rules.

5. Where to start: Which services to pilot first will depend on each context, but the choice matters. Starting with a high-visibility, high-frustration service (like permit inquiries or appointment booking) can build momentum and political support for wider rollout.

A research and policy agenda worth pursuing

Beyond implementation, there are important questions that researchers and policy evaluators should be investigating:

— How do AI-powered citizen services impact citizens’ perceptions and trust in government?

— Do faster resolution rates translate into faster outcomes (houses built sooner, benefits obtained earlier, new businesses opened faster)?

— Which departments integrate these technologies first? Why? And how do they change frontline workers’ understanding of their own effectiveness, identity, and job satisfaction?

— What equity gaps emerge or close when AI agents handle the bulk of citizen interactions?

**

I find all this to be genuinely exciting. It sits at the intersection of a rapidly maturing technology and one of the largest surface areas in government operations. And it has potential for an enormous impact on citizens’ experience with government and their life outcomes.

But this potential comes with significant challenges. The governments that get this right will be the ones that treat AI in citizen services as a public service redesign that puts equity, accuracy, and citizen trust at the center. The governments that get it wrong will likely join the long list of failed digital modernization promises, and will have deepened the very distrust they were trying to reduce.

The window for thoughtful design is now, before procurement decisions are locked in and the defaults harden. If you’re working on this - whether designing policy, deploying tools, studying the impact, or building the companies that will serve this market - I would love to hear from you.