Berkeley's Blueprint for GenAI Readiness
Before a city can deploy generative AI, it needs to know the truth about its data. A deep audit of Berkeley's data infrastructure surfaced critical gaps - and delivered a governance roadmap that makes responsible AI adoption actually possible.
Aspen Tech Policy Hub x City of Berkeley
A capstone engagement through the Aspen Tech Policy Hub with the City of Berkeley. The team brought together a policy PhD, a physician, and an engineer - three disciplines rarely in the same room. All technical work - the data analysis, the EDA, the automation architecture - was designed and delivered by a single engineer.
Project Overview
City of Berkeley - GenAI Readiness Assessment
The City of Berkeley was exploring how to adopt generative artificial intelligence into its government services - a forward-thinking ambition shared by municipalities across the country. The engagement started with one question but quickly surfaced a more important one: the city wasn't asking the right question yet. The real starting point wasn't "how do we deploy GenAI?" - it was "is our data actually ready for GenAI?" A deep audit surfaced the foundational steps needed before AI could deliver on its promise - and that early clarity is what made the engagement genuinely valuable.
The Challenge
Ambition Ahead of Infrastructure
City leadership wanted to deploy GenAI to improve resident services - a legitimate goal. But the data infrastructure needed to support AI hadn't been assessed, audited, or standardized. The ambition was real; the foundation wasn't ready.
Critical Data Quality Gaps
Exploratory data analysis revealed that 90% of "UCR Return A Code" values in police incident data were missing - meaning incidents weren't being properly classified. Data fed into an AI system in this state produces unreliable, potentially harmful outputs.
Siloed, Misidentified Data
In the city's 311 system, residents were entering "Berkeley" as their neighborhood rather than the actual neighborhood name - systematically breaking geographic cross-referencing across departments and making inter-departmental collaboration nearly impossible.
No Governance Framework
Without a city-wide data governance framework, quality problems compound over time. Each department managed data independently with no shared standards, no accountability for data quality, and no roadmap toward certification or best practices.
Approach
The engagement began with an honest reframe. The city's original question - "how do we deploy GenAI?" - couldn't be answered responsibly without first establishing whether the underlying data could support it. That reframe shaped everything that followed.
Honest Scope Reframe
Telling a client what they want to hear is easy. Telling them what they need to hear is what creates real value. The project scope was re-framed from "how to deploy GenAI" to "how to prepare for GenAI" - because building AI on poor data produces poor AI. Garbage in, garbage out.
Data & Statistical Analysis
A deep EDA and statistical analysis was performed across real city datasets - police incident records, 311 service requests, and departmental systems. Statistical methods quantified the severity of gaps, turning vague "data quality concerns" into concrete, evidence-backed findings.
Stakeholder Interviews
Key decision-makers across city departments were interviewed to understand how data actually flows through the organization - not just the technical architecture, but the human workflows, the pain points, and what "working well" would look like to the people who depend on it every day. Most data audits skip this layer entirely.
Near-Term Wins While Governing
Data governance takes time. Rather than leaving the city with only a long-term roadmap, concrete automation examples using Zapier, IFTTT, and Workato were demonstrated - showing how tasks like ordinance drafting could be streamlined immediately, without waiting for the deeper infrastructure work to complete.
What Was Built
Data Audit & Analysis Report
Comprehensive exploratory data analysis and statistical analysis across Berkeley's real city datasets - surfacing specific, quantified quality gaps with visual evidence and stakeholder interview findings, accessible to both technical staff and city leadership.
Governance Roadmap
A city-wide data governance framework aligned with the Bloomberg What Works Cities certification - a structured, externally-validated pathway covering 43 criteria across eight foundational data practice areas.
Automation Recommendations
Concrete sample workflows using Zapier, IFTTT, and Workato were demonstrated to city leadership - showing how ordinance drafting and other high-friction tasks could be streamlined immediately, giving decision-makers a tangible picture of near-term ROI alongside the longer governance roadmap.
Technologies Used
Value Delivered
The city wanted to deploy GenAI but had no clear picture of whether its data could support it. Critical datasets had systematic gaps and misclassifications that would have made any AI system built on them unreliable - or worse, confidently wrong.
A clear-eyed audit surfaced what actually needed to be fixed first. The city received a concrete governance roadmap, near-term automation wins, and an externally-validated certification pathway - the real prerequisites for responsible AI adoption.
Right Foundation First
A data audit before AI deployment isn't a detour - it's the right sequence. Understanding the true state of the city's data meant any future AI investment would be built on a foundation that could actually support it, rather than amplifying existing gaps at scale.
A Concrete Path to Quick Wins
Demonstrated automation examples gave city leadership a clear picture of what efficiency gains were achievable immediately - without waiting for multi-year governance improvements - turning abstract recommendations into something decision-makers could actually visualize and act on.
A Roadmap That Holds
The What Works Cities certification framework gives Berkeley's leadership a structured, peer-validated pathway used by 50+ cities - not just recommendations, but a proven system for becoming genuinely data-driven.