40 Questions Investment Bankers Should Ask Before Trusting AI
AI can build a financial model in minutes, but it cannot tell you which assumptions will break the deal or which competitor move will kill the thesis. Your job is to interrogate what the machine produced and rebuild your own judgement on top of it. These questions separate bankers who use AI as a research tool from those who outsource their thinking to it.
These are suggestions. Use the ones that fit your situation.
1If Capital IQ pulled revenue growth assumptions from comparable companies, do those comparables actually operate in the same market segment as the target, or are they just similar in size?
2When ChatGPT populated the WACC calculation, which discount rate did it use for the risk-free rate and why, and does that rate still hold given current gilt yields?
3The model shows a 15-year cash flow projection. What specifically changes in years 10 to 15 that the AI identified as different from years 1 to 5, or did it just extrapolate linearly?
4If Bloomberg's AI recommended a trading comp multiple, how many of those comparables are actually trading at that multiple today versus how many have moved 20 percent or more since the data was pulled?
5When Kensho suggested a merger risk adjustment to the valuation, what specific deal outcome data did it analyse to reach that percentage, and does that data include recent failed deals in this sector?
6The AI model assumes working capital stays at 8 percent of sales. Have you physically traced that assumption back to the target's actual cash cycle over the past three years, or are you inheriting a default?
7If the sensitivity table shows a range of outcomes, can you explain to the client without looking at the model which three assumptions matter most and why, or have you lost the reasoning?
8When Copilot filled in the tax rate assumption, did it account for the specific jurisdiction the target operates in, or did it use a generic blended rate?
9The AI calculated a terminal value using perpetuity growth. What happens to your valuation if growth is actually half what the model assumes in year 20?
10Have you stress-tested the model against the worst-case scenario from the last deal you closed in this sector, or are you only running scenarios the AI suggested?
Deal Sourcing and Due Diligence
11When Capital IQ's AI flagged a company as a target based on financial metrics, did it also flag that the target's largest customer represents 35 percent of revenue and has a contract expiring in nine months?
12If ChatGPT built your initial due diligence checklist by scraping standard frameworks, what critical questions specific to this target's business model did it miss?
13The AI deal memo says the target has strong margins. Have you verified whether those margins are real or inflated by one-time benefits that will disappear post-close?
14When Kensho identified synergy opportunities, did it base that analysis on this buyer's actual track record of synergy realisation across previous deals, or did it use generic benchmarks?
15Has the AI tool flagged the specific regulatory change that came into force three months ago which will affect how this target operates, or was it trained on older data?
16If Bloomberg's screening suggested this target is undervalued versus peers, can you identify the one or two reasons why the market has already priced it lower that the AI might have missed?
17The deal memo highlights growth prospects in a new market. Did the AI research actual customer demand in that market, or did it infer demand from industry reports about market size?
18When the AI flagged management as a risk, did it actually analyse the management team's performance in previous roles, or did it just note that tenure is short?
19Have you identified which five employees would walk if the deal closes, and does your synergy case assume they stay?
20If the AI said acquisition risk is low because debt levels are reasonable, have you looked at what happens to debt covenants under a recession scenario the model shows?
Client Advisory and Presentation
21When Copilot drafted the investment committee paper, did it include the specific precedent transaction in this sector from two years ago where the buyer overpaid and wrote down the asset, or did it only cite deals that support your thesis?
22Your client is asking why your valuation range is 20 percent higher than the seller's valuation. Can you articulate your answer without reading from what the AI generated?
23The presentation shows a one-page summary of regulatory risk. Are there two or three specific regulatory scenarios that could kill the deal, and did you name them or let the AI describe them generically?
24When ChatGPT polished the deal rationale, did it soften language around execution risk because the tone needed to be positive, even though execution risk is real?
25Your client wants to know the probability this deal closes at your recommended price. What actual data or experience is that probability based on, or is it the AI's statistical output interpreted as confidence?
26The AI-drafted memo says the seller is under pressure to exit. What specifically created that pressure, and can you confirm it with your own conversations, or are you relying on what you read in an article the AI cited?
27If the presentation recommends proceeding with the deal, what is the one piece of information that would reverse that recommendation, and have you built that into your diligence plan?
28When your client pushes back on a key assumption in the model, can you defend that assumption as something you stress-tested, or will you need to go back to the AI to understand how it was set?
29The AI suggested three strategic rationales for the deal. Which one do you actually believe will drive value for your client, and which one are you including because the AI listed it?
30Your client asks what will happen to the target's business in a recession. Does your AI-generated scenario analysis include this client's actual recession experience from the last crisis, or generic models?
Junior Banker Development and Firm Capability
31If you give a junior banker an AI-generated model without asking them to rebuild the key sections, what will they learn about how to interrogate assumptions when they present to a client?
32When was the last time you asked a junior banker to build a financial model from scratch without AI assistance, and what gaps in their thinking did you discover?
33Can a junior banker on your team explain why a particular comparable company is the right one to use, or do they just know the peer set came from Bloomberg?
34If a client challenges a valuation multiple you derived from AI-suggested comparables, could a junior banker on your deal point to the specific reasons those comparables are relevant?
35When you use Kensho to identify deal targets, are you teaching junior bankers to think about which companies actually fit your strategy, or just to trust what the algorithm flagged?
36Your firm's competitive advantage in M&A advisory rests on deal instinct. Is AI handling so much of the initial analysis that new junior bankers never develop that instinct?
37In your last deal debrief, did you ask the team what the AI got wrong and what it missed, or did you skip that discussion because the deal closed?
38If your firm switched to a different AI tool tomorrow, how much of your deal analysis capability would transfer, and how much is locked into knowing how to work around this tool's limitations?
39When a junior banker receives an AI-drafted due diligence checklist, do they know which questions on it actually matter for this specific target, or are they treating the checklist as exhaustive?
40If a major client fired your firm and hired another bank, what part of your advisory work could they replicate using the same AI tools, and what part depends on your people's judgement?
How to use these questions
After the AI builds a financial model, ask a junior banker to identify the three most sensitive assumptions and explain in their own words why they matter. If they cannot, you have not really understood the model either.
When Capital IQ or Kensho flags a target or comparable, spend 20 minutes researching why the market has already priced that company the way it has. The AI shows you what was true last quarter. Your judgement is what happens next quarter.
Before presenting a deal memo to a client, remove the AI-generated language from one section and rewrite it yourself. If you struggle to explain the idea, the AI may have written something that sounds rigorous but is not actually coherent.
Test your deal instinct once a month by predicting the outcome of a deal without running a model. Then build the model and compare. If the model output surprises you, find out why before you trust it.
Set a rule: every financial model must have a written section, in plain language, explaining which assumptions would need to change by how much for the deal recommendation to flip. If the AI cannot articulate that, you do not really know what you are recommending.