Prompt Patterns for FIG (Financial Institutions Group) Investment Banking
Below is a collection of prompt patterns tailored to FIG Investment Banking. Each pattern includes the purpose, the problem it addresses, the data it requires, and a customized prompt. The collection below is geared towards understanding AI Prompts through the lens of FIG, instead of performing in-depth analysis.
1. Meta Language Creation Pattern
Purpose: Create a shorthand language to standardize inputs for faster communication.
Problem: Repeatedly defining financial metrics, key ratios, or terms increases workload.
Data Needed: A glossary of terms or metrics commonly used in banking.
Prompt: “From now on, whenever I say ‘NPL:G’, I mean ‘Non-performing loans growth as a percentage of total assets’. Use this notation in your responses.”
2. Output Automater Pattern
Purpose: Automate repetitive workflows, such as generating comparable company analysis (Comps).
Problem: Manually assembling financial metrics across multiple companies is time-consuming.
Data Needed: A dataset with financial metrics for comparable banks.
Prompt: “Whenever you analyze a dataset of comparable banks, generate a Python script that extracts revenue, net income, and key ratios, then outputs the data as a sorted table in Excel.”
3. Flipped Interaction Pattern
Purpose: Let the AI guide the conversation to gather all necessary inputs for financial models.
Problem: Remembering all variables needed for model-building is challenging.
Data Needed: None initially; the AI will query you for inputs.
Prompt: “You are an investment banker preparing a merger model for two regional banks. Ask me all the necessary questions to build the model, including synergies, financing, and market conditions.”
4. Persona Pattern
Purpose: Customize AI responses to align with a specific role or expertise.
Problem: Generic responses may lack the depth needed for niche analyses.
Data Needed: Context about the type of analysis required.
Prompt: “From now on, act as a senior investment banker specializing in the Financial Institutions Group. Focus on M&A opportunities within regional banks and provide in-depth insights into valuation drivers.”
5. Question Refinement Pattern
Purpose: Help refine ambiguous or incomplete questions for better answers.
Problem: Initial questions may miss critical details.
Data Needed: Your initial query.
Prompt: “Whenever I ask a question about valuation, suggest a refined version that incorporates relevant financial metrics like ROE, P/E ratio, and growth rates.”
6. Alternative Approaches Pattern
Purpose: Generate alternative solutions for achieving a goal.
Problem: Being locked into a single analytical method limits creativity.
Data Needed: The objective you want to achieve.
Prompt: “Suggest three different approaches to valuing a distressed bank, comparing DCF, precedent transactions, and liquidation value methods.”
7. Cognitive Verifier Pattern
Purpose: Subdivide complex questions into manageable steps.
Problem: Complex problems are harder to tackle in one go.
Data Needed: The original question or task.
Prompt: “When I ask a question about projecting revenue, break it into sub-questions such as ‘What is the historical growth rate?’ and ‘What macroeconomic factors could impact future growth?’ Combine the answers to provide a detailed projection.”
8. Fact Check List Pattern
Purpose: Highlight assumptions or key facts for validation.
Problem: AI outputs can include inaccurate data or unverified assumptions.
Data Needed: The output that requires validation.
Prompt: “After providing an analysis of a bank’s credit risk, list the key assumptions and facts that should be validated, such as delinquency rates and asset quality trends.”
9. Template Pattern
Purpose: Enforce a specific output format for reports or presentations.
Problem: Inconsistent formatting can create inefficiencies.
Data Needed: A predefined template or format.
Prompt: “Use the following template for all outputs: ‘Company Name: XYZ Bank | Metric: ROE | Analysis: Details | Recommendations: Details.”
10. Infinite Generation Pattern
Purpose: Continuously generate outputs based on a single prompt.
Problem: Repeatedly entering prompts for similar tasks wastes time.
Data Needed: The initial dataset or problem scope.
Prompt: “Keep generating valuation summaries for regional banks in the dataset, focusing on P/B ratios and ROEs, until I say ‘stop’.”
11. Visualization Generator Pattern
Purpose: Create visualizations to support presentations or analyses.
Problem: Some insights are better conveyed visually.
Data Needed: A dataset relevant to the analysis.
Prompt: “Generate a visualization of Tier 1 capital ratios across top 10 regional banks. Highlight any significant changes over the past 5 years.”
12. Game Play Pattern
Purpose: Make learning or scenario analysis more engaging.
Problem: Traditional analysis can be dry and uninspiring.
Data Needed: A scenario or dataset.
Prompt: “Create a game where I act as a regulatory analyst and uncover risks in a bank’s balance sheet based on clues provided by quarterly filings.”
13. Reflection Pattern
Purpose: Explain the reasoning behind AI outputs.
Problem: Users may not understand how outputs were derived.
Data Needed: The AI-generated response.
Prompt: “When analyzing a bank’s liquidity position, explain the reasoning behind your conclusions, including any assumptions about interest rate movements or asset quality.”
14. Refusal Breaker Pattern
Purpose: Rephrase queries when AI refuses to respond.
Problem: Queries might hit guardrails unnecessarily.
Data Needed: The original question.
Prompt: “Whenever you cannot answer a question, explain why and suggest alternative ways to phrase it so I can get the information I need.”
15. Context Manager Pattern
Purpose: Focus or reset the conversation context.
Problem: AI may include irrelevant or outdated information.
Data Needed: The context to include or exclude.
Prompt: “Focus on analyzing credit risk trends in mid-sized banks. Ignore profitability metrics unless directly relevant.”
16. Recipe Pattern
Purpose: Provide a detailed sequence of steps to achieve a goal.
Problem: Completing complex tasks requires clear guidance.
Data Needed: An end goal and any known steps.
Prompt: “Provide a step-by-step guide for conducting a credit quality assessment, starting with asset review and ending with stress testing scenarios.”