4 Simple Prompt Engineering Techniques That Make AI Think Like a Financial Analyst

Learn how Chain-of-Thought and Tabular Chain-of-Thought prompts can turn raw SEC filings into smart investment insights fast

---

💡 Real Example: From SEC Filing to Smarter Investment Decisions

You feed an AI the 10-K of two tech companies.
It outputs:

Company A: 15% YoY revenue growth, debt ratio 0.25
Company B: 8% YoY revenue growth, debt ratio 0.40
→ Suggests A has better financial resilience under rate hikes.

Now that’s AI reasoning, not guessing.
Let’s break down how you can teach models to think this way using four reasoning & planning techniques.

---

1️⃣ Chain-of-Thought (CoT) Prompting

Example first:

Question: Based on the income statement, did Company X improve profitability?

Step 1: Revenue increased from $1.2M → $1.6M (+33%)
Step 2: Net income rose from $120K → $240K (+100%)
Step 3: Margin improved from 10% → 15%
Answer: Yes, profitability improved significantly.

What it does:

• Forces the model to show intermediate reasoning.

• Helps spot logical errors or weak assumptions.

• Makes answers auditable great for finance and compliance.

Try it yourself:

Prompt: “Analyze Tesla’s 2024 Q2 filing. Show step-by-step how you determine if cash flow supports expansion plans.”

---

2️⃣ Tabular Chain-of-Thought (TCoT)

Example first:
When comparing investments:

| Metric           | Company A | Company B | Reasoning |
|------------------|------------|-----------|------------|
| P/E Ratio        | 18         | 24        | A undervalued |
| Debt-to-Equity   | 0.4        | 0.9       | A lower leverage |
| Growth Forecast  | +12%       | +8%       | A higher growth potential |

What it does:

• Organizes reasoning visually.

• Highlights trade-offs between choices.

• Reduces cognitive load for humans reviewing AI outputs.

Try it yourself:

Prompt: “Compare Q3 performance of Apple, Microsoft, and Google using a reasoning table that includes revenue, margins, and R&D intensity.”

---

3️⃣ Skeleton-of-Thought (SoT)

Example first:
You want a structured investment memo.

Outline:
1. Overview
2. Financial Highlights
3. Risk Factors
4. Recommendation

Now expand each with evidence from SEC filings.

What it does:

• Builds the framework before content.

• Keeps AI from drifting off-topic.

• Great for consistent financial summaries or investor memos.

Try it yourself:

Prompt: “Draft an outline for a portfolio risk assessment report, then fill in each section using ETF performance data.”

---

4️⃣ Plan-and-Solve Prompting

Example first:

Task: Identify if XYZ Corp is undervalued.
Plan:
1. Gather valuation ratios (P/E, P/B)
2. Compare to industry average
3. Interpret in context of growth
→ Execute each step sequentially.

What it does:

• Forces AI to plan before solving.

• Improves accuracy for multi-step reasoning.

• Perfect for investment screeners and forecasting tools.

Try it yourself:

Prompt: “Plan your approach to evaluate if renewable energy stocks are undervalued post-rate cuts. Execute one step at a time.”

---

⚙️ Quick Reference Code Prompt Examples

# 1. Chain-of-Thought
“Show reasoning step-by-step before giving the final result.”

# 2. Tabular CoT
“Summarize reasoning in a markdown table with columns for data, analysis, and insights.”

# 3. Skeleton-of-Thought
“Outline your answer first, then expand each section.”

# 4. Plan-and-Solve
“Write your action plan, then execute it step-by-step.”

# Finance Application Example
“Using SEC 10-Q data, plan and execute a 3-step reasoning process to evaluate liquidity trends.”

# Comparative Prompt
“Compare the same analysis with and without reasoning steps — highlight clarity improvement.”

---

🚀 Key Takeaways

• CoT = Better logic visibility.

• TCoT = Clarity in comparisons.

• SoT = Structured thinking.

• Plan-and-Solve = Multi-step accuracy.

Use these when you want AI to think like an analyst, not just sound like one.

---

Next Step:
Try rewriting your next financial analysis prompt using Tabular Chain-of-Thought then see how much sharper your insights get.