From Experimentation to Operation: The Key Transformation of City Government AI Leadership

Introduction: When AI Experiments Encounter the Complexity of Urban Systems

Hundreds of cities around the world are running AI pilot projects—from traffic flow optimization to automated service processes. Yet, very few have truly moved from pilot to large-scale deployment. In a recent interview, Karthik Anbalagan, General Manager of Emerging Technologies at Granicus, pointed out that the problem lies not with the technology itself, but with how city governments define AI’s role in their operations.

"The difference lies in whether leaders see AI as a technology project or an operating system." Anbalagan’s view cuts to the core contradiction in current smart city development: cities often treat AI as an independent tool rather than a capability embedded in public service infrastructure.

Step One: Start with the End in Mind, Not with Technology

Many cities launch AI experiments without clear objectives. Anbalagan recommends reverse-engineering from business goals: first define two to three use cases and clearly identify metrics for success. This approach may seem simple, yet it is the polar opposite of the common pattern of "buying tools first, then looking for problems."

From a city technology perspective, this involves a fundamental shift: AI is no longer a procurement item on the IT department’s list, but a public service delivery model that city managers need to redesign. For example, the success criteria for an AI system designed to shorten approval processes should not be how many requests it processed, but whether residents’ wait times have decreased and whether civil servants’ job satisfaction has improved.

Establish a Governance Framework: Set Boundaries Before Experimentation

Data security, privacy protection, and ethical frameworks are often hastily discussed only after a pilot emerges. Anbalagan emphasizes: "Don’t wait until after the fact to define the protective boundaries for data security and observability."

For cities, governance is particularly critical. City governments manage large amounts of sensitive citizen data—from medical records to tax information. A system that lacks upfront governance and then produces errors could destroy public trust in digital government. This requires cities to establish cross-departmental governance committees early in AI deployment, covering legal, security, and business leaders.

Avoiding AI Sprawl: Dual Integration of Infrastructure and Roles

"AI sprawl" is a typical problem facing city governments: different departments independently purchase different AI tools, leading to data silos and duplicate investments. Anbalagan suggests deciding in advance which tools, infrastructure, and data layers can be reused across the organization.

This recommendation points to an architectural upgrade of the city’s digital infrastructure. The future city needs a shared AI platform layer—similar to an urban operating system—that allows various departments to call upon unified capability modules. At the same time, new roles will emerge: a governance lead responsible for compliance, and an experience lead responsible for identifying pain points in business processes. This organizational transformation is key to elevating AI from the project level to the system level.

The Budget Dilemma: Shifting from "Pay per Token" to "Invest per Outcome"When budgets are uncertain, cities often fall into short-term procurement cycles. Anbalagan proposes three strategies:

1. Shift the conversation from token volume to business outcomes — for example, whether the cost per transaction is sustainable. If a vendor cannot link usage to business results, that's a red flag.
2. Understand cost predictability rather than mere cost efficiency — Are there usage caps? Price protection mechanisms? If not, you face unlimited exposure risk.
3. Choose optimized partners — those that do not simply pass through raw AI model costs but make architectural decisions based on the use case.

From the city's perspective, this is essentially building a "pay-as-you-go" yet "outcome-driven" spending model. Cities need to structure the financial model of AI services like they build water and power grids: stable, predictable, and based on actual service volume rather than compute consumption.