TL;DR GPT-5 Tips for Finance
If you prompt GPT-5 like GPT-4, you’re leaving money and insight on the table.
Written by ChatGPT-5, curated by Ryion Pun
Bottom line: GPT-5 is no longer just a Q&A bot. Think of it as your strategic finance co-pilot, capable of reasoning, planning, and executing multi-stage analysis like a senior member of your team. Here’s how to get the most out of it in high-stakes finance work.
1. Treat It Like an Agent, Not a Utility
Give GPT-5 a clear endpoint, constraints, and context — then let it work the steps.
Example:
“You are my strategic finance AI. Goal: Prepare a Board-ready analysis on the impact of a 200 bps interest rate increase on our debt portfolio. Phase 1: Identify assumptions. Phase 2: Run scenario modeling. Phase 3: Draft a 5-slide Board deck. Stop after each phase for my input.”
2. Use Multi-Phase Prompts
Move away from one-shot requests. Break complex work into research → outline → draft → review → polish.
Example:
“You are my strategic finance partner. We’re preparing a Board entry strategy for Southeast Asia.
Phase 1: Research macro trends and regulations.
Phase 2: Outline 3–4 market entry models.
Phase 3: Draft P&L, cash flow, and ROI for top 2 models.
Phase 4: Risk/opportunity matrix.
Phase 5: Convert into a 6-slide Board deck.”
3. Add Validation Checkpoints (COT-SEP)
Avoid bad assumptions snowballing into bad outputs. Chain-of-Thought with Separators (COT-SEP) breaks reasoning into steps with review points.
Example:
“You are my senior analyst. Evaluate refinancing $200M in debt. Phase 1: List all assumptions. Phase 2: Model cash flows under three scenarios. Phase 3: Calculate NPV, IRR, payback. Phase 4: Strategic & risk assessment. Pause after each phase for approval.”
4. Control Depth with verbosity and reasoning_effort
High verbosity: For deep Board papers and investor memos.
Low verbosity: For quick exec updates.
High reasoning effort: For complex multi-variable analysis.
Example:
“Evaluate 3 financing options for our $200M capex project. Use
verbosity=highandreasoning_effort=highto include modeling, risk scoring, tax implications, and strategic fit.”
5. Match Tone to Audience (Personalities)
Choose a preset style: Robot (concise), Nerd (detailed), Listener, or Cynic.
Example:
Robot for Board decks: “Summarize Q2 variance drivers in 3 bullets, 15 words each.”
Nerd for FP&A deep dive: “Explain step-by-step the NPV calculation, showing all formulas.”
6. Leverage Coding, Voice, and Tool Calls
Integrate GPT-5 with your data stack for instant analysis.
Example:
“Pull last 8 quarters of revenue from Snowflake via SQL, run a regression forecast, and preview the chart in Canvas before adding to the investor deck.”
7. Break Complex Trade-offs into Labeled Steps
Example:
“Step 1: Define problem — SG&A up 18% YoY.
Step 2: Identify cost drivers.
Step 3: Evaluate 3 reduction options.
Step 4: Recommend best option with ROI and payback.”
8. Use Long Context Windows
Feed GPT-5 full Board packs, budgets, and prior forecasts in one thread.
Example:
“Here’s our full 2024–2025 budget (45 pages) + last 4 Board packs. Identify budget-to-actual variances, emerging risks, and 3 priority areas for the Audit Committee.”
9. Trust but Verify
Even with fewer hallucinations, always cross-check high-stakes outputs.
Example:
“Project EBITDA under a 10% revenue drop. Cross-check via contribution margin method and full P&L model.”
This is how I use AI to learn AI. I hope this brings you some food for thought. Enjoy!
This is AICFOHK.
Ryion Pun
Founder, AICFOHK.org
Lead with joy. Choose simplicity. Create wonder.