Prompt Engineering for Financial Professionals

The five elements

Every effective prompt has up to five elements. You won't always need all five, but knowing the framework lets you diagnose why a prompt isn't working.

1. Role — Tell the AI who it is.

You are a senior credit analyst at a European investment bank.

This isn't small talk. It activates domain-specific knowledge and adjusts what the AI focuses on.

Context and task

2. Context — The background the AI needs.

I'm reviewing a quarterly earnings transcript for Acme Corp (Q3 2025). We have a $50M credit facility maturing in 18 months.

The more specific your context, the more relevant the output. "Analyse this document" → generic. "Analyse this given we have credit exposure maturing in 18 months" → useful.

3. Task — What you actually want.

Identify the three most significant credit risks mentioned or implied in this transcript, with supporting quotes.

Not "analyse this" — but what kind of analysis, how many things to find, and what evidence to include.

Format and constraints

4. Format — How to structure the output.

Present as a table with columns: Risk Factor, Evidence (direct quote), Severity (High/Medium/Low), Recommended Action.

Without this, the AI guesses. For work going into an IC memo or client deliverable, you want the format right the first time.

5. Constraints — Set boundaries.

Do not speculate beyond what is stated or clearly implied. If the evidence is ambiguous, say so.

This is critical for financial services. It prevents hallucination and forces transparency about uncertainty.

Putting it together

Here's the full prompt, assembled from all five elements:

You are a senior credit analyst at a European investment bank. I'm reviewing the attached quarterly earnings transcript for Acme Corp (Q3 2025). We have a $50M credit facility with them maturing in 18 months.

Identify the three most significant credit risks mentioned or implied in this transcript, with supporting quotes. Present as a table with columns: Risk Factor, Evidence (direct quote), Severity (High/Medium/Low), and Recommended Action.

Do not speculate beyond what is stated or clearly implied. If the evidence is ambiguous, say so.

This takes 30 seconds to write. The output is ready for a credit committee discussion.

Templates for your role

Copy these. Replace the brackets with your specifics.

Earnings analysis:

Review this transcript for [Company]. Extract: (1) Guidance changes vs. prior quarter, (2) Management commentary on [topic], (3) Questions management deflected, (4) Statements diverging from consensus. Format as an investment memo with one-paragraph executive summary.

Due diligence review:

You are conducting buy-side DD on [Target]. Review the attached [document type]. Flag: inconsistencies, unsupported claims, information gaps, key valuation assumptions. Organise by priority: Critical / Important / Minor.

Comp table:

Build a comp table for [companies]: EV/Revenue, EV/EBITDA, P/E, revenue growth, EBITDA margin, net debt/EBITDA. Highlight where [Target] sits vs. peer median. Note exclusions and why.

The team multiplier

Build a prompt library:

1. Identify your top 10 recurring tasks
2. Write and test a prompt for each
3. Share where the team can find them
4. Update based on what works

Quick fixes

Output too generic → You didn't provide the actual document. Attach it.

AI made up facts → Add: "Only reference the attached documents."

Too long → Add: "Maximum 500 words" or "One-page executive summary."

Ignored instructions → Number them. Put the most important one first.