The ground rule: your firm's policy comes first
Before anything else: investment banks handle material non-public information (MNPI) about clients and deals. Pasting a client's financials, a deal codename or a draft CIM into a public AI chatbot can breach confidentiality agreements, securities rules and your employment terms.
Only use AI tools your firm has approved, in the way it has approved them. Many banks now provide internal AI assistants that run inside their own secure environment. If you're not sure whether a tool is allowed, ask your staffer or compliance before you use it, not after.
Where AI helps an analyst most
1. Getting up to speed on a company or industry
Before a pitch, you might need to understand an industry you've never seen. With approved tools, AI can summarize public sources such as annual reports, earnings call transcripts and industry primers into a first-pass overview: key players, how they make money, the main metrics and recent themes. You still read the sources for anything that goes into a deliverable.
2. First drafts of written content
Company profiles, “situation overview” pages, email drafts to clients and talking points all start as blank pages. AI is good at producing a structured first draft from notes you provide. Your job is editing it for accuracy, tone and the house style of your group.
3. Excel and formula help
Stuck on a nested formula, a dynamic array or a macro? Describe what you need in plain English and ask for a formula and an explanation of how it works. It's especially useful for building checks, cleaning messy data exports and writing lookups across large tables. Always test the formula on cases where you already know the answer.
4. Reviewing your own work
AI can help proofread slides for typos, inconsistent tense and formatting, or flag places where a number in the text doesn't match a chart label. It's a second pair of eyes, not a replacement for the checklist your VP expects you to run.
5. Meeting and call notes
Where your firm allows transcription or note tools, AI can turn a long internal call into action items. Confirm what was agreed with your team; summaries can miss nuance.
Where AI can hurt you
- Made-up numbers and facts. AI tools can produce confident, specific and wrong figures, such as revenue, deal multiples or dates. Every number in a client deliverable needs a source you can point to: a filing, a database or the client's data.
- Stale information. Market data, share prices and deal terms change constantly. Use your firm's data platforms for anything market-related.
- Hidden errors in formulas. A formula that looks right can anchor the wrong cell or ignore an edge case, like a year with negative income. Test it.
- Losing the skill. Analysts are promoted for judgment and technical mastery. If AI writes every formula for you, you won't be able to build a model in a timed test or catch a junior's mistake later.
A simple workflow that works
- Check the policy and use an approved tool.
- Give context, not confidential data. Describe the structure you need (“a three-statement model with a revolver”) rather than pasting the client's numbers into an unapproved tool.
- Ask for reasoning, not just answers, so you can follow the logic.
- Verify everything against a source or a hand calculation.
- Own the output. If it goes out with your name on it, it's your work.
Talking about AI in interviews
Interviewers increasingly ask how you'd use new tools. A strong answer shows enthusiasm and judgment:
“I use AI to get up to speed faster, for example summarizing an industry or helping me write a tricky Excel formula, but I always verify numbers against the source and I'd only use tools the firm has approved, because client information is confidential. I see it as a way to spend less time on formatting and more time on analysis.”
Avoid claiming AI can do an analyst's job. The work still depends on people who understand the numbers, and that's what interviews test.
AI can't take your modeling test for you
Build the core skills yourself, one formula at a time, with instant feedback.