Credit analysts spend hours reviewing financial documents to determine whether a business can repay a loan. While modern platforms help aggregate and “spread” financial data, the analysis phase remains largely manual. Analysts must interpret years of financial history, identify risk signals, understand ratios, and ultimately produce a credit memo summarizing the borrower’s financial story.
To support Moody’s broader push toward agentic AI experiences, I began exploring how AI could assist analysts in interpreting financial spreads and accelerate the path toward generating a credit memo. The result was an AI financial analysis orchestrator: a contextual AI assistant that sits alongside financial data and helps analysts extract insights, visualize trends, and generate summaries.
Lead UX Designer
Research, interaction design, conversation design
Developing patterns for agentic experiences
I led early concept exploration, working with product stakeholders, engineers, and internal financial experts to define how AI could support analyst workflows.
Ongoing (Exploration phase)
The project is still in exploration so formal metrics are not yet available. However, early feedback suggests the concept could reduce time spent analyzing financial spreads, help analysts identify trends faster, and accelerate the creation of credit memos
Credit analysts review financial statements to determine a borrower’s creditworthiness and answer a central question:
“Will the bank get its money back?”
After financial documents are structured into financial spreads, analysts must review three or more years of financial data to identify trends, assess risk, and understand the financial story of the borrower.
This analysis typically involves:
- Identifying revenue and margin trends
- Understanding liquidity and leverage ratios
- Evaluating cash flow sustainability
- Projecting potential financial scenarios
Despite modern financial platforms, this process remains highly manual, time intensive, and mentally demanding, requiring analysts to sift through large financial tables before translating insights into a credit memo used by loan committees to approve or reject lending decisions.
Current workflow:
Financial Documents
(PDFs, spreadsheets)
↓
Financial Spreading Tool
(structured financial tables)
↓
Manual Analysis
• Identify trends
• Calculate ratios
• Create charts
• Interpret financial performance
↓
Notes + External Tools
(Excel, calculators, internal docs)
↓
Credit Memo Creation
Pain points you should highlight:
• Manual pattern detection
• Repeated calculations
• Context switching between tools
• Time spent translating numbers into narrative
To better understand analyst workflows, I conducted 4 interviews with credit analysts and internal SMEs. Key insights include:
- Why did DSCR drop in 2023?
- What happens if revenue declines 10%?
- How does leverage trend over time?
This makes the workflow well suited to conversational exploration.
- Revenue growth or decline
- Margin compression
- Cash flow coverage
- Increasing leverage
Each time analysts must manually create charts and interpret the data.
- What changed?
- Why did it change?
- What risks does it introduce?
The final deliverable — the credit memo — translates financial data into a narrative explaining the borrower’s strengths, risks, and repayment capacity.
Instead of asking analysts to manually interpret large financial tables, an AI system could:
- Extract patterns
- Generate summaries
- Visualize trends
- Simulate scenarios
While still allowing analysts to guide the analysis and verify conclusions. The design introduces a contextual AI assistant embedded beside the financial summary table. This assistant acts as a financial analysis orchestrator that can route requests to specialized agents.
Instead of presenting a blank prompt, the design introduces task-oriented starting points based on common analysis activities. These prompts help analysts quickly initiate high-value workflows:
“Generate best-case and worst-case scenarios”
This triggers a projection agent trained to model financial scenarios.
“Show the DSCR trend over the last 3 years”
This activates a graph generation agent.
“Create a two-paragraph financial summary”
This uses a summary agent trained on analyst reporting patterns.
This approach lowers the barrier to entry while still allowing analysts to ask custom questions. Future iterations could generate prompts dynamically based on detected financial signals (e.g., declining DSCR or rising leverage).
Agentic System Architecture (Conceptual)
Financial Documents
↓
Financial Spreading Tool
↓
AI Financial Analysis Assistant
↓
User asks questions or selects prompts
↓
AI Orchestrator routes request
↓
Specialized Agents
• Projection Agent
• Graph Agent
• Summary Agent
↓
Insights Generated
(charts, projections, summaries)
↓
Saved to Entity Overview
↓
Credit Memo Creation
Key improvements to label:
• Faster insight discovery
• Reduced manual analysis
• Reusable outputs for documentation
• Guided exploration of financial data
Analysts remain responsible for interpreting insights and making the final credit recommendation.
The AI orchestrator interprets the request
It selects the appropriate specialized agent
The agent performs the analysis
Results are returned to the analyst
Agents explored in the concept:
Projection generation
Graph generation
Financial summarization
This approach allows the system to expand with additional analysis capabilities over time.
Saving Insights
Analysts can save AI-generated outputs such as:
Charts
Financial summaries
Scenario analyses
Saved items are collected in an Entity Overview page, allowing analysts to assemble a structured view of the borrower. From here they can export insights to support credit memo creation.
Why an AI Assistant?
During early exploration we considered several ways AI could support financial analysis.
Full automationAutomatically generate a complete credit memo from financial data.
Insight dashboardsPre-generate charts and risk signals without user interaction.
Conversational analysisAllow analysts to explore financial data through prompts and guided questions.
We chose the assistant approach because financial analysis is inherently investigative. Analysts rarely follow a fixed checklist — they explore the data, ask questions, and investigate anomalies.A conversational interface allows analysts to guide the investigation while maintaining trust and control over the final decision.
Key Design Challenges
Designing AI for expert financial workflows introduced several challenges:
Balancing automation with analyst trust - Financial decisions require defensible reasoning. The system needed to surface insights without obscuring how conclusions were generated.
Avoiding generic AI chat - A blank prompt risks forcing analysts to guess what the system can do. Structured prompts were introduced to guide high-value workflows.
Maintaining financial context - Analysts must always see the underlying financial data when interpreting insights. The assistant was intentionally placed beside the financial spread rather than replacing it.
Several principles guided the design:
Analysts maintain control over interpretation and final recommendations.
Instead of generic AI chat, the design focuses on high-value financial analysis workflows.
Analysts frequently need to reuse charts and summaries in documentation. Saving outputs reduces duplicate work.
Because the project is still in exploration, formal metrics are not yet available. However, early feedback suggests the concept could:
- Reduce time spent analyzing financial spreads
- Help analysts identify trends faster
- Accelerate the creation of credit memos
Future validation will focus on measuring:
- Time required to identify key financial signals
- Usefulness of AI-generated summaries and graphs in credit memo creation
- Trust in AI-generated insights
