Scaling Auditable Agentic Workflows in Financial Services – with Leaders from Moody’s and Prudential Insurance

This article is sponsored by Moody’s and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.

Regulated industries face compounding operational and regulatory pressures that steadily erode speed, accuracy, and competitiveness. Analysts lose hours to manual reporting because critical data remains scattered across siloed systems, and outdated processes slow decision cycles and heighten risk exposure amid shifting regulations.

The increasing burden of regulatory reporting signals the scale of this challenge. In the United States, compliance remains overwhelmingly labor-driven: a Cato Institute Research Brief in Economic Policy finds that U.S. firms devote 1.3-3.3% of their total wage bill to regulatory compliance, while the underlying CESifo analysis shows that 93.9% of compliance costs in the U.S. financial sector are labor-related, rather than attributed to automation or equipment.  

The burden is even heavier for smaller institutions. A decade of CSBS Community Bank Survey data shows the smallest banks devote ~11–15.5% of payroll to compliance versus ~5.6–9.6% at larger institutions, reflecting fixed, largely manual obligations that don’t scale down with asset size.

Despite significant investments in automation and AI, organizational performance remains hampered by fragmented data and manual work. S&P Global Market Intelligence research shows that data analysts spend 54% of their time just finding and preparing data for analysis, reducing capacity for consistent KPI reporting and strategic insight generation.

The expanding use of AI introduces additional governance complexity when deployed on top of these labor-intensive, fragmented foundations, revealing widespread gaps in achieving mature and auditable KPI standards. Meeting these challenges requires agentic frameworks that integrate compliance into day-to-day decision-making and align closely with the needs of IT, risk, and business units.

Emerj recently hosted a special series of the ‘AI in Business’ podcast focused on agentic AI solutions and human-led automation in regulated industries, featuring executives from Moody’s and Prudential Insurance. Pavlé Sabic, Senior Director of Generative AI Solutions and Strategy at Moody’s, and Nina Edwards, Vice President of Emerging Technology and Innovation at Prudential Insurance, joined Emerj Editorial Director Matthew DeMello to unpack the intricacies of agentic AI deployment challenges.

This article examines the following key insights from their discussions for financial leaders aiming to deploy agentic AI effectively, ensure regulatory compliance, and scale pilots into measurable outcomes:

• Accelerating credit workflows with agentic AI: RAG systems turn fragmented data into memos under analyst supervision.
• Leveraging proprietary data for regulatory-ready AI: Proprietary ratings ensure precise, auditable memos under regulation.
• Redesigning metrics for AI velocity: Outcome charters reveal trapped productivity — enabling visible enterprise ROI.
• Unifying language for enterprise scale: KPI glossaries standardize terms — enabling comparable executive reporting.

Listen to the full episodes from this series below:

Episode 1: From Manual Reports to Generative and Agentic AI Automation in Finance – with Pavlé Sabic of Moody’s

Guest: Pavlé Sabic, Senior Director, Generative AI Solutions and Strategy at Moody’s

Expertise: Generative AI Strategy, Enterprise Agentic Systems, Data & Analytics Leadership, Fintech & Risk Intelligence

Brief Recognition: Pavlé Sabic is a Senior Director leading generative AI solutions and strategy at Moody’s, where he helps global financial and risk organizations adopt enterprise-grade AI and agentic workflows. He previously held senior commercial and analytics leadership roles at S&P Global Market Intelligence, and his insights have been featured in the Wall Street Journal, CNBC, Fortune, and other global outlets. Sabic also serves as an Associate Partner at the University of Edinburgh Business School.

Episode 2: Rewiring Systems to Scale AI From Demos to Deliverables – Nina Edwards of Prudential Insurance

Guest: Nina Edwards, Vice President of Emerging Technology and Innovation at Prudential Insurance

Expertise: Enterprise AI Strategy, ROI Metrics & Scaling, Emerging Technology Leadership, Applied Intelligence, Financial Services Innovation

Brief Recognition: Nina Edwards is Vice President of Emerging Technology & Applied Innovation at Prudential Financial, where she drives AI strategy and scaling initiatives across the enterprise. She previously served as Global Chief of Staff for Accenture’s Applied Intelligence practice, supporting growth and strategy in data, AI, machine learning and more. Her career spans strategy development, partnerships, financial planning, performance metrics, executive reporting, and operations across financial services.

Accelerating Credit Workflows with Agentic AI

Pavlé pinpoints the core bind for financial leaders: analysts buried in manual work and fragmented data, grappling with frequently changing regulatory demands atop complex, outdated systems. These pressures stretch credit decisions from days to weeks, compounding inefficiency and risk exposure.​

Sabic contrasts low-risk, customer-facing agentic workflows such as appointment booking with higher-stakes workflows like credit origination. For banks processing hundreds of industry-specific documents, these systems aggregate company data, sector trends, and firm news into compliant credit memos that can reduce production time by 60% under analyst oversight.

Beyond credit origination, Pavlé notes similar agentic applications in sales intelligence and KYC screening workflows and distills operational priorities for regulated credit environments down to three key points:

• Auditability over raw speed: Traceable outputs for compliance
• Human-led supervision: Agents are responsible for drafting while analysts validate liability
• Repetitive data tasks: Automate aggregation for strategic focus

Leveraging Proprietary Data for Regulatory-Ready AI

Pavlé goes out of his way to caution leaders in industries like financial services and healthcare that face an increased regulatory burden across the board, but especially in their use of new GenAI tools and models:

“The industry is going through a bifurcation where regulated and non-regulated use LLMs differently. Non-regulated can get away with off-the-shelf LLMs, but regulated has to be careful with audit, security, and switching costs. Every time a new LLM comes, they have to change the stack and re-implement.”

– Pavlé Sabic, Senior Director, Generative AI Solutions and Strategy at Moody’s

Public datasets create audit gaps and inconsistencies for agentic AI in finance. Pavlé notes that proprietary datasets — credit ratings, firmographics, risk analytics — serve as vetted foundations fused with external signals for enterprise-grade outputs. These datasets, he tells the Emerj executive audience, are essential for thoroughly auditable AI systems:

“Proprietary data is foundational context for an AI agent: you can’t have a system censoring all possible information or it becomes unmanageable. Unlike a half-decade ago where everything was ingested, regulated industries now require specific, client/industry/sector data. This specificity enables AI to pass audit tests where general internet-sourced systems fail.”

– Pavlé Sabic, Senior Director, Generative AI Solutions and Strategy at Moody’s

Retrieval-Augmented Generation (RAG) is a technique that helps enterprises enhance LLMs by retrieving real-time information from an external knowledge base (usually in the form of a database or another repository digitized documents). The system then utilizes the context from these resources to generate more accurate and verifiable responses that reduce hallucinations and enable domain-specific answers with less direct training.

RAG-powered systems trace every step for full auditability while accelerating credit memo production rooted in precise internal sources over web risks. From this data foundation, Pavlé outlines compliance priorities for enterprise deployments:

• Maintain grounding in internal data: Promote precision over public sources.
• Mandate step-level logging: Enforce full traceability for audits.
• Human-supervised origination: Ensure analysts own high-stakes calls.

Redesigning Metrics for AI Velocity

Nina argues that one of the biggest reasons AI pilots stall, even when they deliver real gains, is that outdated metrics cannot capture the velocity that gets swallowed by legacy processes. Engineers deliver code at incredible speed, service teams draft replies in seconds, yet rigid quarterly gates and approval chains erase this impact.

“System readiness means a human-centered operating model where people shift from doing to deciding, from gathering to governing, processing to prioritizing, checking to sequencing. AI handles high-frequency, repetitive, deterministic work while humans focus on judgment. An insurer moved underwriters from data gathering to decision sequencing — tracking time-to-decision instead of documents reviewed — and cut cycle times from days to hours.”

– Nina Edwards, Vice President of Emerging Technology and Innovation at Prudential Insurance

Nina recommends ‘outcome charters’ that lock in targets like reduced rework and shorter cycle times, and pairing them with weekly operational reviews that track exceptions, customer impacts, and the amount of rework avoided, in order to transform local speed into enterprise-visible proof.

To make these shifts measurable, Edwards points to transitional metrics now surfacing in large financial firms’ digital‑transformation programs. Measures such as “rework avoidance rate,” “time‑to‑decision,” “cost‑to‑comply,” and “exception resolution time” create a bridge between quarterly financial KPIs and continuous operational outputs.

These benchmarks, already appearing in consulting and regulatory frameworks, help translate agentic speed into CFO‑ready indicators of resilience and productivity. Nina and Pavlé outline complementary priorities for turning AI velocity into measurable outcomes:

• Shift metrics from hours saved to end-to-end flow: Track cycle time, exceptions, and rework avoided — not isolated task speed.
• Standardize human-AI roles: Humans should govern/sequence while agents execute repetitive data tasks under supervision.
• Embed weekly operational cadences: Replace model reviews with decision-time tracking and customer impact logging.
• Embed auditable workflows: Human-led origination with proprietary data ensures regulated ROI visibility.

Unifying Language for Enterprise Scale

Nina notes that fragmented definitions for these metrics and more compound enterprise communication gaps — ‘analyst productivity’ might signal the volume of output to operations, but cost efficiency to finance, causing confusion and burying AI impact.

“Language fragmentation is a real challenge. Analyst productivity in one area versus another can have completely different meanings. Operations sees it as claims processed per analyst. Finance measures cost per analyst or cost to serve. Those numbers can’t be compared or aggregated enterprise-wide.”

– Nina Edwards, Vice President of Emerging Technology and Innovation at Prudential Insurance

Nina’s solution: enterprise-wide KPI glossaries that standardize cycle time, exception rates, automation percentage, and rework avoided — and embedding these directly into tools such as Workday and Jira for more consistent enterprise reporting.

She outlines three practical rollout steps to achieve these outcomes:

• Map inconsistencies: Create a shared truth document auditing how teams define core metrics (one retail bank reduced approval cycles from months to days in this manner).
• Codify with examples: Build a KPI glossary with real use cases — claims teams should define “exception rates” from the start, enabling clear financial stories within one quarter.
• Embed in tools: Insert standardized definitions into daily platforms so that staffing, funding, and evaluations all speak the same ROI language.

Pavlé complements this point by noting that agentic systems must deliver consistent, auditable outputs across all organizational tools, helping to unify workflows and outputs even when teams use disparate programs like Word, email, and analytics platforms.