Cognitive Sovereignty Self-Audit for Investment Bankers

1. When you build a financial model in Excel or a Bloomberg terminal, how do you approach the key assumptions about revenue growth, margins, and terminal value?

I accept the assumptions that Capital IQ AI or ChatGPT suggest and use them to complete the model quickly. I use AI suggestions as a starting point but adjust them without going back to verify them against company reports or competitor data. I test AI suggestions against my notes from earnings calls and peer analysis, then decide which to keep or change. I build my own assumption set first based on the business, then compare it to what AI suggests to spot what I may have missed.

2. A junior banker on your team has used Copilot to draft a 40-page deal memo. When you review it, how do you decide whether it is accurate?

I scan it for tone and structure, assume the analysis is sound, and send it to the client with minor edits. I check a few key sections against the source documents but rely on the AI's logic for sections I know are complex. I verify the main findings against deal documents, industry reports, and legal opinions to catch any gaps. I trace every material claim back to its source, identify what the AI missed or oversimplified, and rewrite those sections myself.

3. You are assessing whether to recommend a deal to a client. The financial model shows a strong IRR. How do you form your view on deal risk?

I rely on the risk summary that Kensho or ChatGPT generated from industry data and historical deal outcomes. I read the AI risk summary and add my own thoughts based on conversations with the management team. I form my own judgement about competitive threats and execution risk, then compare it to what the AI flagged. I spend time with the business plan, competitive landscape, and management track record to build my own view before looking at what AI surfaces.

4. When you pitch a deal to a client, how much of the original analysis and insight in your presentation came directly from you versus from AI tools?

Most of the analysis came from Bloomberg AI or ChatGPT. I assembled it and added client-specific context. AI tools generated the bulk of the analysis. I rewrote some sections and added my own perspective on key issues. I did the main analysis myself and used AI to help shape the narrative and graphics. I generated the core insights and recommendations first, then used AI to strengthen the presentation without changing my view.

5. A junior banker brings you a three-statement model built with Copilot assistance. You spot a circular reference error. What does this tell you about the model?

AI made an error. I fix it and trust the rest of the model is sound. AI may have made other errors I cannot see. I ask the junior banker to rebuild it and spot-check a few more line items. The error suggests I need to walk through the build logic with the junior banker to understand what went wrong. I assume the junior banker did not verify the AI output properly. I ask them to rebuild the model from scratch without AI and show me the logic step by step.

6. During due diligence, you need to assess the quality of a target company's customer contracts. How do you approach this?

I upload the contracts to ChatGPT and use its summary to identify contract terms and renewal risk. I use AI to summarise the contracts, then I read a sample of them myself to verify the summary is correct. I read key contracts myself, note what concerns me, then ask AI to check the full population for similar patterns. I read and score the material contracts myself based on criteria I define, then I use AI to flag outliers I should review.

7. Your client is considering whether to make an offer at a price your model supports. They ask you a question the model does not answer. What do you do?

I ask ChatGPT or Copilot what it would recommend and relay that to the client. I look at what the AI suggests, think about it for a moment, and give my initial reaction to the client. I take time to think through the question based on my experience and what I learned in diligence, then I answer. I tell the client I need to think about it, I form my own view separately from any AI input, and I come back with a reasoned recommendation.

8. You are training a new analyst on how to use AI tools for financial modelling. What is your main guidance?

Use AI to build models fast. It knows best practise. Trust its outputs and move on. Use AI to speed up the work. Always sanity-check the results against what you know about the business. Build your model logic yourself first, then use AI to check your work and spot inefficiencies. Learn to build models by hand before you use AI. Only then can you spot when AI gets it wrong and why.

Cognitive Sovereignty Self-Audit for Investment Bankers

Cognitive Sovereignty: How To Think For Yourself When AI Thinks For You