Skyline National Bank Strengthens Fraud Operations with AI
How a Community Bank Used Microsoft Copilot Studio Agents to Enhance Fraud Analysis While Keeping Humans in Control
Executive Summary
Artificial intelligence is creating new opportunities for banks to improve efficiency, uncover insights, and support employees in high-value and time-consuming operational workflows. For Skyline National Bank, one of those opportunities emerged within the fraud department, where analysts spent significant time reviewing daily fraud dispute reports and searching for patterns hidden within large volumes of transaction data.
To address this challenge, Skyline partnered with ABM to develop a Fraud Dispute Review Agent using the Microsoft Copilot Studio platform. The work grew from Skyline’s broader Copilot and data assessment discussions with ABM and was shaped by a formal proof-of-concept scope of work focused on utilizing CM20 fraud dispute reports, secure SharePoint-based knowledge sources, and natural language review of historical fraud activity.
Today, Skyline reports strong adoption signals, active testing feedback, and early-stage business value. The Skyline team has been uploading current and historical fraud files, validating agent responses, and using feedback controls to help ABM evaluate answer quality, adoption, and readiness for next-step conversations.
The Challenge
Like many financial institutions, Skyline National Bank relies on daily fraud and dispute reporting to identify potential losses, investigate suspicious activity, and protect customers. In Skyline’s case, the initial challenge centered on credit card fraud dispute review processes. Prior to the project, analysts manually reviewed daily cases, closed cases, general ledger files, and related dispute data while trying to spot repeat activity across time and other potential fraud trends.
Analysts were tasked with utilizing vendor-generated CM20 reports as a primary source and additional internal and external reporting sources considered as part of their analysis and discovery process. Skyline is looking for ways to reduce analyst’s time and effort, and increase accuracy and overall analyst productivity.
The bank was looking for a way to:
- Reduce manual review effort
- Improve visibility into fraud trends and recurring activity
- Identify patterns that might otherwise go unnoticed
- Support analyst productivity without sacrificing oversight
- Maintain security, compliance, and governance controls
The Solution
ABM worked with Skyline National Bank to design and implement a Fraud Dispute Review Agent powered by Microsoft Copilot Studio, aligning the agent configuration to Skyline’s existing Microsoft 365 environment and operational fraud review workflow.
While the term “agentic AI” generally refers to AI systems taking autonomous action, this solution was built to serve as an intelligent assistant for fraud analysts. An agent was designed that would ingest information from knowledge sources across a newly created Sharepoint site. As Skyline coordinated file uploads and testing access, the ABM project team focused on making sure the agent could work from the right data, in the right locations, with the right governance boundaries before expanding usage. ABM migrated approved file shares as agent knowledge sources, published updated versions during testing, and ran delta migrations so newly uploaded Skyline data could be available to the agent for review.
Key capabilities include:
- Pattern and trend analysis across historical CM20 fraud dispute reports
- Identification of repeat offenders, recurring activity, and over-time dispute trends
- Structured summaries and operational reporting
- Support for consumer and commercial workflows, including BIN-aware review where definitions are supplied by Skyline
- Natural language interaction through Microsoft Teams
- Human-in-the-loop review for all business-critical decisions
Because the solution operates within the bank’s Microsoft environment and approved SharePoint locations, Skyline maintains control of its data, permissions, and governance policies. The project also reflected Skyline’s security posture in practice, including coordinated tenant access, conditional access controls, and careful handling of administrative setup.
Throughout testing, Skyline’s team helped validate that the agent was grounded in authorized source content by uploading the relevant daily, closed-case, repository and ledger files, confirming when new data was ready, and providing response feedback that could be used to evaluate adoption and answer usefulness.
Responsible AI in Banking
From the beginning, both organizations agreed that AI should support bank employees, not replace their judgment.
The Fraud Dispute Review Agent was intentionally designed to provide recommendations, observations, and insights while leaving all final decisions to trained banking professionals. The solution does not approve or deny disputes, make account decisions, perform customer-facing tasks, or take any autonomous action. Instead, it helps analysts work more efficiently and consistently while maintaining accountability and regulatory compliance while multiplicatively increasing their ability to discover, process, and analyze large quantities of dispute records.
This approach reflects Skyline National Bank’s commitment to adopting AI responsibly and demonstrates a practical model for community banks exploring similar technologies.
Results
Although initially deployed as a proof-of-concept, the solution quickly demonstrated meaningful business value. The test group of analysts have started using the agent solution daily and have reported back strong levels of positive feedback. ABM encouraged testers to use thumbs-up, thumbs-down, and written feedback so the project team could measure adoption, evaluate response quality, and tune the experience based on real user input. Analysts can now ask questions of historical fraud data and quickly identify relationships, trends, and recurring activity that were previously difficult to detect through manual processes alone. Skyline continues to report positive adoption and growing confidence in the agent’s output as fraud personnel incorporate the solution into their daily review processes.
Discovery of New Fraud Patterns
Perhaps most importantly, the solution is helping the bank uncover fraud patterns that were previously undiscovered, creating opportunities to strengthen internal procedures and improve fraud response strategies. The feedback from Skyline indicates that the agent is not merely accelerating review; it is surfacing relationships and questions that may influence how the bank updates its internal fraud analysis SOPs. The agent is noted to be learning both internal procedures and noticing patterns that have not been brought forward by analysts at Skyline to date.
“The agent is starting to notice fraud patterns that we have never even considered and is causing us to amend internal fraud analysis SOPs.”
Agent Next Steps
The future of the fraud analysis agent will be centered around expansion of adoption across analyst teams, expanding historical data stores, and streamlining access to the most current data on a daily basis. As data quantity begins to scale, agent performance will remain closely monitored. Skyline is considering the implementation of an additional background agent whose primary responsibility will be to organize and normalize data before becoming available to the analysis agent. This will continue to ensure accurate performance and analysis by the front-end, employee facing agentic tool.
Looking Ahead
What began as a focused fraud analysis initiative has created a foundation for broader conversations around AI-enabled banking operations. After the fraud agent work, Skyline invited ABM to explore additional AI use cases and ways Skyline could leverage Copilot more broadly across the organization.
By starting with a clearly defined business problem, leveraging Microsoft Copilot Studio, and implementing strong governance controls, Skyline National Bank has demonstrated how community banks can responsibly apply AI to augment human expertise, improve operational effectiveness, and strengthen risk management practices. The project also gives Skyline a practical reference model of a sensible path to implementing future AI initiatives: begin with governed data, validate through controlled testing, preserve human decision-making, and expand only after measurable value and user confidence are established.
If you’re looking to streamline operations, curious on how you can unlock new efficiencies, or just want to discover new opportunities in the AI space, we’re here to help. Contact us today to begin!