Building a Culture of Innovation and Governance in Public Service

Anthony Fisher, Data Governance Manager, Department of Revenue, the State of Colorado

Through this article, Anthony Fisher, Data Governance and Artificial Intelligence Program Manager at the Colorado Department of Revenue shares his career journey from investment banking to public service, highlighting how a purpose-driven mindset and data expertise have shaped his impact across agencies. He outlines the evolution of data governance at DOR, the creative use of Tableau to drive insights with limited resources, and the importance of a “fail forward” culture to build a resilient data strategy. Anthony also explores how AI can support customer service and policy development while reinforcing the importance of ethical implementation. His message is clear: progress begins with clean data, strong governance, and the courage to innovate.

A Career Coded in Purpose

My career began in finance as an investment banker at a Chicago firm specializing in middle-market mergers and acquisitions, restructuring, and capital markets transactions. This role strengthened my analytical skills and underscored the importance of data integrity and effective communication for high-level decision-making. I transitioned to consulting with a global firm in Denver, gaining valuable insights into data management practices and the implementation of robust data governance processes within large corporations.

Driven by a desire for public service, I pursued a career with the FBI. I was drawn to this opportunity to become a Special Agent due to my fascination with seeking the truth and wanting to protect others; I felt my data skills would fit right in. During the application process, I maintained my analytical expertise through freelance consulting, working with clients on performance analytics for the military, and optimizing Unmanned Aerial Vehicle (UAV) missions in agriculture.

My continued interest in data analytics led to an opportunity with the Colorado Department of Revenue (DOR) Marijuana Enforcement Division (MED) to spearhead the division’s data analytics program. For more than three years, I built and implemented a data lake, a Large Language Model (LLM), and other technical tools to support DOR. Recognizing my expertise, DOR's Executive Director's Office (EDO) hired me to help establish the DOR data strategy. For the past two years at the EDO, I've focused on leading DOR's data governance program and developing AI policies and use cases.

We are implementing Gemini for Google Workspace internally for our staff. Users are required to take general AI training provided by the Office of Information Technology (OIT) and additional training developed by my team at DOR.

Dashboards and Breakthroughs

Tableau has significantly impacted the DOR over the past few years. We have nine divisions, each with a mission and the freedom to build their infrastructure as long as our data governance policies are followed. Some divisions don't have the staff or budget for large data teams, but Division of Motor Vehicles (DMV), has done something extraordinary. Even without a formal data team, this group within Division of Motor Vehicles (DMV), uses Tableau fully. They've created internal dashboards to help with decisions and understand what's happening in Division of Motor Vehicles (DMV) offices across the state. Next year, they plan to launch public dashboards with which customers can interact. It all started with a conversation and brainstorming, then getting people trained on Tableau. A year later, they show everyone what's possible with supportive leadership and a vision. These dashboards not only improve internal decision-making but will ultimately lead to better services for the citizens of Colorado.

“We can establish test environments and engage stakeholders to gain clarity and move forward. Although these actions require effort, the rewards of embracing innovation are boundless.”

Lessons in Trial and Dashboard Error

 I firmly believe in the “fail forward” philosophy, which our Chief Strategy Officer, Michael Arrington, also champions. In my two-plus years at the EDO, we've implemented numerous strategies to build a strong data culture across our divisions. Some, like our monthly Data Management Group, have been successful. Others haven't worked as intended. However, we believe in the value of these experiences. They teach us what works and doesn't, allowing us to adapt, learn, and progress. This 'fail forward' approach is particularly important in data governance, where experimentation and adaptation are key to finding the best solutions.

Scoring Progress, One Pillar at a Time

Our data governance program utilizes a maturity model and scoring mechanism to assess each division's current state across the foundational pillars of our data governance policy and guidelines. We conduct scoring every six months, and I collaborate with each division's data steward(s) to develop tailored strategies for program progression, considering their unique needs, resources, and workload constraints. In our initial assessment in early 2024, aggregate data, on a scale of one to five (one being least mature, five being most mature), revealed that most divisions averaged scores of two or three across the data governance pillars. A year later, we observed an increase in scores to three and four, indicating growing leadership support and a higher state of program maturity. This advancement can be attributed to various factors, such as divisions refining their data inventories, securing funding for new data infrastructure, or enhancing data integrity through dataset standardization for unified reporting.

AI on Call: Helping Hands, Not Replacing Them

AI has significant potential to enhance our operations in several key areas, particularly customer interactions and policy management. DOR has two divisions with support centers that handle many public inquiries. As part of a broader strategy, we are expanding our online services to provide customers with more interaction options. Both these areas rely heavily on technology, creating ideal opportunities for AI integration.

AI can assist our staff in managing inquiries in our support centers. While automating initial intake entirely with 'robo' systems is not ideal, as research suggests customers value human interaction. AI support would enable our limited staff to handle more customer interactions effectively.

For our expanding online services, an AI-powered chatbot on our service portal could guide customers through online applications, particularly benefiting those in remote areas of the state.

Finally, AI can benefit our policy management team, which excels at creating and managing DOR policies and ensuring compliance and efficient operation. However, the complexity and volume of policies can lead to overlaps or inconsistencies, requiring frequent revisions. AI can assist by analyzing policy documents, identifying potential overlaps, and suggesting revisions. AI can also streamline policy tracking, version control, and accessibility. By automating these tasks, we can free up our policy team to collaborate with stakeholders to enhance the policies and strategic development.

As we integrate AI into our operations, we are firmly committed to ensuring its ethical and responsible use. Our approach prioritizes privacy and fairness and is fundamentally underpinned by the principle of maintaining human oversight in all critical processes.

Face the Data, Not the Fear

I have two quotes for this:

“Fear is the mind-killer." This quote from Frank Herbet’s “Dune” resonates with me, as it speaks to the potential pitfalls of fearing new technologies. While innovation naturally brings risk aversion, too much caution can lead to paralysis by analysis. We can establish test environments and engage stakeholders to gain clarity and move forward. Although these actions require effort, the rewards of embracing innovation are boundless.

“Do your homework.” Many organizations are eager to adopt AI, but their data is disorganized and lacks formal governance. Just as a parent advises a child to complete homework before play, ensuring data is clean and secure before deploying AI is crucial. Establishing robust data governance is essential prior to AI adoption and implementation, as rectifying data issues afterward is significantly more challenging.

AI

Data Management

Test

Information Technology