AI Skills for Finance Jobs: What Banks and Funds Expect of Candidates

The AI skills for finance jobs that banks and funds describe are working habits more than technical credentials. Five recur in their own words: checking a machine's draft before anyone relies on it, tracing each figure back to its source, asking for work inside the firm's approved tools, keeping client and deal information out of prompts, and, for data and AI seats, Python.

The firms teach the tools themselves once a new hire arrives. What a candidate brings is the judgment to use them, and one checked piece of work is the clearest way to show it. The firms that publish a rule on AI in hiring draw the same line from the other side: prepare with it, then answer in your own words.

What AI skills do finance jobs require?

Each of the five has a firm's own words behind it: an executive describing how junior staff now work, a training programme, or a careers page. Here they are side by side, with the evidence and its date.

Four of the five belong to anyone who works with a model at a bank or fund; only the last depends on the seat. Mastery of a particular tool is not on the list. The habits that make a tool's output safe to use are.

Checking a model's work, seat by seat

The first two rows are one habit with two halves: judge the draft as a whole, then prove its parts. What that looks like depends on the desk.

On a deal team

The draft is a page: a comps table, a company profile, the first cut of a pitch. Checking it means testing the choices the model made, which companies it treated as peers and what it left out, as well as the arithmetic it did. Those questions only occur to someone who could have built the page without the tool.

In research and advisory work

Here the draft is a summary of a filing, an earnings call or a library of research. Morgan Stanley's grading, in the table above, is the institutional form of the habit: experts score the machine's answers before anyone relies on them. The individual form is simpler and just as strict. Open the source, find the passage the summary rests on, and read it.

In a data or AI seat

In a data seat, checking stops being a step and becomes the job. An August 2026 posting for BlackRock's AI lab asks its data scientists for "benchmark construction, metric design, and statistically sound measurement of quality, safety, and reliability": deciding what a right answer is before the system runs, then measuring how often it reaches one. The data scientist seat at BlackRock shows what that work involves week to week.

What banks teach new hires about AI

If checking is what the firms want, prompting is what they teach. The prompting row in the table is two training programmes, and graduate schemes carry AI tracks of their own: UBS builds an AI Fluency Pathway into its Graduate Talent Program. Citi's mandatory course took experts under ten minutes and beginners about 30, and the firms that train new analysts teach the tools alongside the manual work.

The timing is the point. The mechanics of a firm's tools are taught after the offer, often in minutes, and nobody outside a bank can practise on its own assistant anyway. What no short course installs is the judgment to see when an answer is wrong. That rests on having done the work without the tool, and it is the part a candidate has to bring through the door.

What should you never put into an AI tool at work?

Of the five skills, one can be failed before the first day. The banks' first answer to ChatGPT, in early 2023, was to restrict it; what replaced the block was access on the bank's own terms, with controls on what goes in.

At a bank that screens prompts the way Goldman does, a prompt belongs in the same category as an email: a communication that someone in compliance may read. That settles most questions about what goes in. These stay out of consumer apps entirely, and go into the bank's own tool only where its policy allows:

• client names and deal terms
• anything from a data room
• anything a deal team told you

Where the line falls in detail, from the legal meaning of material non-public information to what the consumer apps do with conversations, is set out on the page on ChatGPT and Claude in banking.

Do you need to code for a finance job?

Only where the posting says so, and the careers pages are specific about it:

• JPMorgan's investment banking analysts, in the bank's description, "write reports, build updated financial models, and support multi-billion dollar transactions." The work is models and documents.
• JPMorgan's Data & AI analysts need, for AI-focused roles, Python and "familiarity with common analytical stacks," with SQL and visualization tools such as Tableau for data-focused roles.
• BlackRock's campus Analytics & Modeling function asks for "intermediate coding and programming skills, e.g., R and/or Python," alongside knowledge of finance, econometrics, statistics or advanced math.

For someone without a technical degree, the CFA Institute's practical skills modules are one structured way in: Python Programming Fundamentals at Level I, and Python, Data Science, and AI at Level II, each taking 10 to 20 hours.

Where a posting names code, the code is part of the job. Where it names models and reports, the spreadsheet is, and the AI skill that matters is checking what a model puts into it.

Can you use AI in a finance job application?

Code is where AI help and your own work blur most easily, and three firms have set out where they draw that line for candidates: BlackRock and HSBC on their careers sites, and Goldman Sachs in an email to campus applicants that Fortune reported in June 2025. They agree more than they differ.

These rules govern how a candidate answers, not what they are asked. BlackRock and HSBC both let candidates prepare with AI; all three want the answers in the room to be the candidate's own. Goldman's spokesperson told Fortune: "We want to hear from our applicants in their own voice."

Is the entry-level finance job shrinking?

Every rule above assumes there is a seat to apply for. Whether banks are cutting those seats is answered bank by bank in Will AI replace investment banking analysts, which reads what executives have said on the record, the hiring numbers banks have put their names to, and where the far larger figures in circulation come from.

It is worth reading before an interview, because a confident number with no source behind it is the easiest thing to repeat and the hardest to defend.

How do you show AI skills before your first finance job?

A candidate cannot set the size of the class, only the evidence brought to it. The strongest evidence of the checking habit is a finished piece of work in which a model met a known right answer on public data, with a record of where it failed. Five projects built that way, from grading a model's answers against an annual report to replicating a published recession model, set out the data, the build and what a finished version looks like.

How to prepare, step by step

Each step turns one of the five skills, or the hiring rules above, into something you can point to:

1. Read the posting for its skill words. Code, models and reports, responsible use: match your evidence to the words the firm chose, and if it names a programming language, reach the level it asks for before you apply.
2. Keep a log of one task done twice, if a full project is more than you have time for. Do the task by hand, give it to a public chatbot, and write down every difference and which version was right. Use public filings, transcripts and your own work only, never anything from an internship or a client, and be ready to say why in a sentence.
3. Find the firm's rule on AI before you apply. Prepare with AI where it allows, and answer without it in the interview.
4. Write the result, not the tool. On a CV, a line that names the data, the check and what it found is evidence. A tool name on its own is a claim.

The difference shows on the page:

Claims the skill

• Proficient in AI chatbots and Copilot
• Passionate about AI in finance
• Completed an online AI course
• Familiar with prompt engineering

Shows the skill

• Checked an AI-built comps table against the filings and logged each error
• Built a valuation by hand, then compared the AI version line by line
• Tested an AI summary tool against the source filings, with an error rate
• Traced each quote in an AI call summary back to the transcript

From log to CV line, worked

Say the log covers a company profile a chatbot drafted from a public annual report, with 25 figures in it. Checked one by one against the report, 4 do not match, and 2 of those turn out to be figures from the previous year's report. That is an error rate of 16%, 4 divided by 25, and a finding with a cause attached.

The CV line comes straight from the log: an AI-drafted company profile checked against the annual report, 4 of 25 figures wrong, each traced to its source. The numbers are illustrative. The shape is the point, because it gives an interviewer something concrete to ask about.

AI skills for finance jobs, in short

The AI skills for finance jobs that banks and funds have described are habits of supervision: check the machine's draft, trace what it says to its source, ask for the work inside the firm's tools, keep what is confidential out of them, and write code where the seat calls for it.

The firms teach the tools. What they look for is the judgment, and the firms with written hiring rules want the answers in the interview to be the candidate's own. A single checked piece of work, and the ability to explain every line of it, is how a candidate shows that judgment.