A Case Study in AI-Powered Process Automation
January 2, 2026
CogAbility recently partnered with the Maricopa County AZ Treasurer’s Office to deploy an AI-powered Mailing Address Research Tool that eliminated 1,000s of hours of manual work, streamlined service delivery, and saved over $1 million for Maricopa County residents.
Public sector agencies across the country face familiar challenges: increasing service demands, dealing with massive and growing data volumes, and constantly re-prioritizing workloads due to limited staff availability.
Nowhere is this challenge more visible than in mission-critical but tedious administrative tasks like correcting addresses for undelivered mail.
When undelivered mail backlogs grow, the impact is felt not only by government employees, but by citizens who depend on timely and accurate communication.
Tasked by the County to address a property tax accounting issue, the Maricopa County Treasurer’s Office demonstrated how responsible use of AI can simultaneously clear a massive backlog of work, streamline operations, and deliver significant ROI.
The Challenge: 200,000 Pieces of Undeliverable Mail
The Treasurer’s Office is responsible for issuing and managing critical taxpayer communications, many of which require the use of physical mail.
These mailings are not optional—they are essential to ensuring residents receive accurate information about property taxes, payments, and their obligations.
Due to a recent legal settlement, the Treasurer’s office was suddenly charged with clearing a backlog of more than 200,000 pieces of mail returned by the U.S. Postal Service as undeliverable, with more returns arriving daily.
Each returned mail piece requires a manual research workflow to locate and validate a current mailing address for the parcel involved. This workflow involves the following steps:
1. Review Property Assessor records to identify other properties owned by the same individual and infer a current address.
2. Search Arizona Corporation Commission records for updated business addresses.
3. Conduct GAMA deed research to identify address updates associated with property transactions.
Performed manually, this process is slow, repetitive, tedious, and error-prone.
Clearing the backlog of 200,000+ bad mailing addresses would have consumed thousands of staff hours and distracted the team from more important issues, introduced rework, and delayed taxpayer communications.
The Treasurer’s Office needed a solution that would accomplish the following:
- Research 200,000 addresses in days, not months
- Deliver better-than-human accuracy, per strict business rules
- Minimize total cost and processing time
- Keep staff free to focus on higher-value work
- Deliver a production tool that can be used to research undelivered mail in the future.
The Solution: CogBot ‘Taxi’
Rather than applying AI to a one-off project, the Treasurer’s Office and CogAbility took a structured, people-centered approach, resulting in a long-term address-correction solution the team is proud to call their own.
Together, Maricopa’s most experienced staff trained their CogBot—nicknamed ‘Taxi’—at CogUniversity, the world’s first university designed to teach machines rather than humans.
Maricopa enrolled Taxi in the CogUniversity course, “Process Automation Lab – TXC7800L”, where CogAbility engineers and Treasurer Office subject-matter experts collaborated to:
- Translate institutional knowledge into clear & auditable business rules
- Design & deploy a fully-automated mailing address research workflow
- Build dashboards and safeguards to monitor accuracy, report key stats, and identify opportunities for continuous improvement.
The result of this effort is a new Mailing Address Research Tool that autonomously executes the research process for every piece of returned mail.
Once trained, Taxi was set loose to research and correct a returned-mail file containing 216,453 records.
How Taxi’s Mailing Address Research Tool Works
Taxi’s AI-powered tool executes the same research steps that a trained employee would—but at a speed, cost, consistency, and scale impossible to achieve manually.
For each returned mailing address, Taxi performs 3 research tasks:
- Search Property Assessor data to identify ownership records.
- Query Arizona Corporation Commission data for the latest business address updates
- Review deed data in the Property Appraiser’s GAMA system to find the most recent mailing addresses.
The Technology Underneath
Taxi’s Mailing Address Research Tool was built with the following components in CogAbility’s AI platform for local government:
- Robotic Process Automation (RPA): Taxi uses a secure RPA tool to access, operate, and extract information from applications and websites during the research process – just like a person would.
- Large Language models (LLMs) are used to analyze and compare address & ownership data.
- A Master Controller architecture guides & monitors Taxi’s research work.
- CogAbility’s Model Context Protocol (MCP) provides Taxi with access to the software, data sources, and tools it needs to perform its job.
- CogAbility’s AI Safety Guardrails for Local Government keep Taxi’s AI-based work safe, secure, and on-topic – and insulated from LLM hallucinations.
- Built-in versioning and analytics monitor performance, validate outcomes, and continuously refine the workflow as rules and data sources evolve.
- An interactive dashboard displays Taxi’s progress in real-time and provides easy start/stop control for Taxi’s team members.
Using this technology, Taxi functions as a digital employee that is trained, supervised, and supported by Maricopa County’s team – just like any other new team member would.
The Result: Backlog Eliminated + $1 Million Saved
The impact of Taxi’s contribution to the team was immediate and substantial:
1. Address Corrections: No Longer a Bottleneck
During the initial phase, Taxi analyzed 216,453 returned-mail records, executing all three research steps for each record—a total of 649,359 research actions—without human intervention.
2. Research Accuracy: Better than Human
To ensure trust and accountability, the Treasurer’s Office reviewed Taxi’s performance and found that Taxi finds the correct address with 90.3% accuracy (±3% at 95% confidence level).
This level of accuracy exceeded expectations for a process historically prone to inconsistency and rework and is considered better-than-human performance.
3. Labor Savings: $1 million
The Treasurer’s Office realized $1,065,923 in estimated labor savings researching the first batch of 200k addresses.
The project achieved payback within the first few months, delivering an ROI exceeding 30x the annual cost of Taxi.
4. Research Backlogs: Gone
Once operational, Taxi was able to research and correct more than 200,000 addresses in about one week.
As a result, Maricopa County tackled its massive 1-time address cleanup without hiring temporary labor.
Moving forward, Taxi can research 1,000s of addresses per hour, so future undeliverable mail issues can be resolved same-day if needed.
5. Other Benefits for the Maricopa County Treasurer
Taxi’s Mailing Address Research Tool delivered more value than the obvious hard benefits, including:
- Automated a mission-critical, countywide task.
- Faster postal mail resends, improving the success rate for taxpayer mailings.
- Reduced human error and rework, increasing confidence in plans and outcomes.
- Staff remains focused on higher-value tasks, improving morale and productivity
6. Benefits for Maricopa County Residents
The benefits for Maricopa County residents and businesses are meaningful, too – including:
- Faster delivery of critical tax communications
- Fewer missed notices due to outdated addresses
- Increased trust in the Treasurer’s Office
Lessons Learned: Best Practices for Government AI
Several key lessons emerged from this effort, including:
1. Treat AI like an employee / coworker.
Because Taxi was trained by Maricopa’s staff at CogUniversity, the team achieved a high-quality outcome with very little rework. In addition, staff ownership and accountability for Taxi was clear on day one..
2. Clearly define the process, business logic & decisioning rules before you build.
Well-documented business rules are essential to maintaining accuracy at scale.
3. Expect to learn as you go.
Process automation – especially with AI – often surfaces hidden issues and complexities in the underlying business process, so team flexibility and rapid iteration are critical.
4. Scale early.
Early deployment at real-world volumes reveals opportunities for improvement that small pilots would have never seen. This also eliminates the need to refactor the process to scale it, later on.
5. Measure everything.
Quantifying the ROI & key operating metrics upfront turned a boring technical project into a leadership-visible AI success story, paving the way for more AI innovation.
Looking Ahead: A Foundation for AI Process Automation
The Mailing Address Research Tool is just the beginning of Maricopa’s AI journey.
The Treasurer’s team is discussing opportunities to expand Taxi’s role in several ways:
1. Deploy Taxi in daily operations to prevent returned mail backlogs from growing.
2. Give Taxi new jobs, such as Customer Service and Employee Support. These are pretrained/tested skills that Taxi can quickly learn by enrolling in CogUniversity courses.
3. Leverage Taxi’s insights to support better decision-making.
This case study also serves as a model for other counties seeking to deploy AI responsibly, transparently, and with a measurable public benefit.
Conclusion: ‘Boring’ AI Can Deliver BIG Results
By using in-house experts to train CogBot Taxi to research and correct undelivered mailing addresses, the Maricopa County Treasurer’s Office transformed a massive backlog of tedious clerical work into a scalable, accurate, and cost-effective automated process the entire county can rely on for years.
For CogAbility, this project reinforces one of our core beliefs, that responsible AI is not about replacing people—it’s about empowering them.
For Maricopa County citizens, CogBot Taxi proves that AI can make government work better for everyone – and save a boatload of money in the process.
To learn more about CogBots, CogUniversity, and AI-powered process automation for local government, contact us at [email protected].
