Part I: Lecturer: Andrea Cremonino

1- Intro to Artificial Intelligence as a modelling concept and application in bank and corporate Treasury

Origins of AI and its application to banking and finance

Distinguishing between AI, Machine Learning, and Generative AI

Agentic AI and AI in decision making

2- AI and Treasury

Setting up your AI, data integration, and integrating AI initiatives into Treasury and ALCO processes

Gen AI and Large Language Models: uses and applications

Examples:

• Machine Learning (ML) and Large Language Models (LLMs) in
  • cash management, (in/outflows, intraday, zero balance, payments cutoffs
  • risk management (exposure identification. Payment breach, cash ladder),treasury policy automation.
• Cash Flow Forecasting: Using predictive analytics to improve accuracy.
• Middle office: Data checks
• FX Risk Management: Utilizing AI as a “hedging co-pilot” to analyse market trends
  • Structural fx
  • tactical (short term) hedges
• Automation: Using LLMs to automate treasury policy, and reporting (pt 1)
• Reporting pt.2: LLMs and ALCO Dashboard production

Small Language Models:

• Cheaper to train, fine-tune, and run. SLMs trained on your bank’s past ALCO decks

Financial report drafting, scenario stress testing, and using Retrieval-Augmented Generation (RAG) to ensure accuracy.

Writing strong prompts: Examples and refining

Issues

• Data availability
• System integration
• Black box
• Explainable AI

3- Case Study: Cash Flow Forecast

Engineering AI model for finance, mitigating AI hallucinations: a practical example

• Definition of objective
• Identification of CF sources
• Integration of flows
• Forecast 1:Time series extrapolation
• Forecast 2 Causal model by AI
  • Target definition
  • Variables selection
  • Model testing
  • Backtesting
• Running ongoing monitoring

4- Transition to Agentic AI

Moving beyond passive AI to systems that can autonomously initiate actions

• Executing interest-rate risk exposure hedging
• Middle office: contract confirmation
• Margin calls

Good practice in Agent design functional side

---

Part 2: Lecturer: Tom Ho

5-  AI Governance

Appropriate governance for AI in  terms of

• Model development
• oversight
• Centralised developments vs repositories of prompt
• Data infrastructure

6- Example illustrations

Demonstration of the AI model in action as part of the routine ALCO process:

1. A) MI package for core balance sheet data produced as an automated process by AI:
2. B) Key takeaways from the MI report produced by AI (the “CAMELS”)

What are the strengths and weaknesses? What are the optimal strategies to balance the CAMELS risk exposure? Note: the few proactive strategy can address each CAMELS risk in isolation from the others.

1. C) Decision Support and Decision Making Action items from the AI ALCO report (demonstrated in detail for each item below):

Strategic  / Tactical recommendations from the AI ALCO report

• Capital Structure Optimisation
• Interest-Rate Risk Calibration
• Credit Risk Management
• Funding and Liquidity Resilience
• Capital Efficiency Improvements
• Scenario-Driven Portfolio Rebalancing

7- Case study

“Why we should incorporate GenAI and Agentic AI in our tactical and strategic processes.”

Use case for AI-driven MI.

Why are these reports essential? They allow for integrated, Strategic ALM ALCO processes:

“Signature Bank embraced crypto, not Liquidity. New York Community Bank (NYCB) focused on earnings rather than the asset quality of its commercial real estate loans. First Republic Bank extended its duration during COVID-19, not raising capital. Republic First Bank focused on growth, not on Management’s internal control. Decisions are seldom confined to a risk source.”

8- Application example: Demonstration of relevant applications

1. Generation of ALCO and CAMELS reports that present the bank’s strategies with clarity and their strategic goals.
2. Collaboration between Buy and Sell agents, Risk Agents, and the CFO agent to generate NII before and after simulations.
3. “Real-time” collaboration: Agentic AI enabling ALCO to evaluate What-If Simulations.
4. Demonstration of product pricing and profitability while meeting long-term risk-based performance.
5. Selection of optimal risk-appetite adjusted strategies (with and/or without benchmarking against peers)

Part III: Online assessment

Multiple-Choice Test & Certificate:

The BTRM Advanced Treasury Masterclass Series ends with a Multiple-Choice Test. This test is available until Wednesday 15th July.