AI Tools in Finance: What a Junior Analyst Uses and What Each Gets Wrong

The AI a junior in finance is most likely to open at work comes in four kinds: Microsoft's Copilot inside Excel, AlphaSense for research documents, Bloomberg's ASKB on the Terminal, and the general models, OpenAI's ChatGPT and Anthropic's Claude, usually reached through a tool the bank has approved rather than the public app. Each one drafts a different piece of the same job.

What they share matters more than what separates them. The first version of a model, a research summary or a pitch page can now come from a prompt, and the junior's work moves to asking for the right thing and checking what comes back. The employer, not the analyst, decides which of the four sit on the desk.

Which AI tools do finance analysts actually use?

Sorted by where they sit in the work, the four look like this. Read the last column first: at a firm of any size, none of them is something an analyst picks for themselves.

The fourth row is the one students know best and meet least in its familiar form. Several of the biggest banks restricted the public ChatGPT in early 2023, and the same models came back to their staff through routes the banks control.

What do AI tools change for a junior analyst?

Read down the column of what each tool does well and one pattern repeats: every one of them produces a first version of work a junior used to build by hand. The model, the research answer, the market read and the pitch page can all start as a draft that arrives in minutes.

That moves the hours to the two ends of the task. Before the tool runs, the junior decides exactly what to ask for and what may go in. After it runs, the junior proves the draft, figure by figure, against the source it came from.

One request, before and after

Take a request that lands on a Monday: a short profile of a listed company, needed before a client call on Wednesday. Done by hand, most of Monday went on finding and typing: the filings, what management said on the last call, how the shares have traded, a first set of comparable companies.

With the tools, a draft of each part can exist by lunchtime. What fills the rest of the two days is the part no tool signs off: whether each figure matches the filing, whether the comparable companies are the right ones, and whether the story on the page is one the numbers support.

Three habits that carry across all four

Whichever tool a firm has bought, the same three habits decide whether its draft helps or hurts:

1. Ask for the finished shape. Name the period, the companies, the conventions and the form of the output, so the draft has something concrete to be wrong against.
2. Prove each figure where it lives. A filing, a transcript page, a cell reference: a number that cannot be traced does not go on the page.
3. Decide what goes in before you type. Client names, deal terms and anything that is material non-public information belong only in a tool the firm has approved for them.

The banks describe the same shift for their own analysts; what it means for the number of seats is weighed in whether AI will replace investment banking analysts.

Where each tool fits in a working day

Each of the four has a page of its own, and each answers a different question a candidate brings to it. Taken in the order a piece of work usually runs, from reading to the finished file:

Reading the documents: AlphaSense

Most projects start with reading, and AlphaSense is built for that stage. Its page explains what sits in the library, whose words stand behind each citation, and why a well-cited answer can still be out of date. For anyone about to spend a first year on research, it is the tool whose limits matter most. How AlphaSense works for a finance analyst.

Checking the market: ASKB on the Bloomberg Terminal

Next comes the market: how the shares have traded, what the Street expects, what the news says. On the Terminal that now starts with a question typed into ASKB rather than a memorised function code. The Bloomberg page covers what counts as fluency once the Terminal takes plain English, how the 2023 research model BloombergGPT relates to the product analysts use, and how to audit the numbers ASKB returns. What ASKB does on the Bloomberg Terminal.

Drafting the deliverable: ChatGPT and Claude

Then the drafting: the summary, the comparable companies, the pages of a pitch. This is the stage the finance editions of ChatGPT and Claude are built around. That page covers the two routes by which banks let staff reach the models, what each lab's finance edition includes, and the line on what must never go into a prompt. How analysts use ChatGPT and Claude at work.

Building the model: Copilot in Excel

Everything ends in a file, and for most juniors that file is a workbook. Copilot in Excel is the AI inside it. Its page walks through what it builds, where its errors hide, which version an employer has to buy, and the arithmetic of checking a valuation it produced. What Copilot in Excel builds, and what to check.

How widely are AI tools used in finance?

None of the four is a niche product, though no single number says which is used most. Each company counts something different, and these are the figures each has put its name to, or that a publication reported:

30m+

Paid Microsoft 365 Copilot seats

Microsoft, July 2026; every Office app, not only Excel

7,000+

Enterprises using AlphaSense

AlphaSense, June 2026

About 1 in 3

Terminal users with ASKB beta access

of 375,000, as WIRED reported in April 2026

120,000+

BBVA staff set to get ChatGPT Enterprise

BBVA, December 2025

Read them as four different things. Microsoft counts paid seats of Microsoft 365 Copilot, and one seat covers Word, PowerPoint and Outlook as much as Excel. AlphaSense counts companies, not people, and its customers run from Amazon and Pfizer to the D. E. Shaw Group. The Terminal figure is access to a beta, not daily use. BBVA's is one bank's plan for its whole workforce.

A figure from any of them is worth repeating only with its owner and its month attached.

Why the tools are starting to look alike

Those four products started in different places: a spreadsheet, a document library, a market-data service and a chatbot. By 2026 they were converging on three things.

The same files

Every one of them now reaches into the files a junior is judged on:

• Copilot works inside Excel itself.
• AlphaSense's add-ins for Excel and PowerPoint, released in July 2026, extend a model's logic or re-spin a pitch book for a new target.
• Anthropic's Claude add-ins for Excel, PowerPoint and Word became generally available in May 2026.
• OpenAI's banking edition, launched in September 2026, turns analysis into models and pitchbooks in a firm's own templates.
• ASKB hands back the query code behind its numbers, so the analysis can carry on in Excel.

For a junior that raises the value of plain Excel and PowerPoint fluency rather than lowering it. The drafts land in those files, and that is where they are checked.

The same data, bought by the firm

Each tool answers from data, and much of that data is licensed by the firm rather than supplied by the tool. Microsoft says its finance data connectors in Excel may need their own subscription from the provider. AlphaSense shows real-time bank research only where the reader's firm holds an entitlement, which a broker typically grants when the two already do business.

Bloomberg describes ASKB's agents as working over the data a user is entitled to, and OpenAI's banking edition reaches several data sets through a firm's existing subscriptions. An answer is only as complete as the licences behind it, which is why one product can give a fuller answer at one bank than at another.

The model is becoming a setting

The third is the model underneath. Copilot in Excel lets a user switch between OpenAI's and Anthropic's models or leave the choice on Auto. AlphaSense offers a drop-down of models, with a default that picks one per question. Bloomberg built ASKB on several commercial and open-weight models.

When the model is a menu choice, the differences that last are the data behind the tool and the person checking what it says.

How is a bank's own AI different from a vendor's tool?

There is one more kind of AI a junior may meet, and it belongs to the bank. After restricting the public ChatGPT, some of the largest banks built assistants of their own. JPMorgan's LLM Suite is one: a portal the bank built in front of OpenAI's and Anthropic's models, which about 250,000 of its staff could use by September 2025. Goldman Sachs and Citi have built their own versions over outside models.

The difference is what each side brings. A vendor sells one product to many firms and brings something a bank could not easily build alone: Microsoft's spreadsheet, AlphaSense's library, Bloomberg's data. A bank's own assistant brings control instead.

Bought from a vendor

• Sold to many firms at once
• Brings its own software, library or data
• Reached through a seat the employer buys
• Some can be tried through a school or a personal plan

Built by the bank

• Open only to staff of that bank
• Puts outside models behind controls the bank sets
• The bank decides which data and which model answer
• Cannot be tried before you join

For a candidate, the practical difference is honesty. Nobody outside a bank can use its internal assistant, so the credible way to discuss one is through what the bank has said in public. The fullest public record belongs to JPMorgan's LLM Suite: why the bank built it, the models behind it, and what it changes for analysts.

Which AI tool should you learn before a finance job?

The one you can actually use, since access decides the order more than importance does:

1. Copilot in Excel, at home. Microsoft 365 Personal and Family subscribers can get it with an AI credits plan, and Microsoft 365 Premium includes it. Use it on a model you have already built by hand, so every difference between the two versions is visible.
2. A general chatbot, on public material only. The consumer ChatGPT and Claude apps are fine for practising on annual reports, filings and earnings transcripts. Anything from a client or a deal stays out.
3. AlphaSense, through a business-school library. Some schools license it for current students, so check your library's database list before assuming you have no way in.
4. Bloomberg's own courses. Bloomberg Market Concepts is a self-paced introduction to the financial markets taught through Bloomberg's own system, and Bloomberg Spreadsheet Analysis teaches BQL, the query language ASKB hands back with its numbers.

Whatever the tool, work the task by hand first and compare the machine's version afterwards. A bank's own assistant is the one kind nobody can learn in advance.

How to talk about AI tools in a finance interview

Banks and funds do not publish which AI tools, if any, their interviews cover, so there is no list to memorise. What an interviewer can hear is whether you understand the tools the way the job uses them. Four things show that:

• The right name for the right job. Each tool here has a lookalike: GitHub Copilot, the expert network AlphaSights, BloombergGPT, the public ChatGPT app. Mixing one up is an easy tell that you have read about a tool rather than thought about it.
• One error you found. A specific mistake in an AI draft, where it came from, and the check that exposed it.
• Where the data comes from. Knowing that an answer is only as complete as the firm's licences shows you understand why the same tool behaves differently at two employers.
• Only what you have used. Say plainly which tools you have tried, and talk about the rest through what their makers and the banks have said in public.

The interview readiness tool rehearses answers like these, and AI in investment banking covers what each bank has built for itself.

The bottom line

The AI tools in finance that a junior actually meets are four products at four points in the same piece of work: Copilot in the spreadsheet, AlphaSense in the documents, ASKB on the Terminal, and ChatGPT or Claude in the drafting, usually behind a door the bank built. The employer picks them and pays for them.

What the analyst brings is the same whichever one is open: a precise ask, a check of every figure against its source, and a firm sense of what never goes in.