Best AI Tools for Investment Banking in 2026

For investment banking in 2026, Rogo is the most complete AI product built for bankers, and it tops this list: hundreds of agents turn a prompt into finished pitch books, models and documents on licensed data. Kepler is our pick for numbers a client will see. Every multiple in a comps table or precedent set traces to a line item in a filing, on each company’s own adjusted EBITDA, computed in code and cited, and the pages come out in the bank’s own format.

This list is for coverage and M&A bankers, ECM and DCM teams, restructuring groups and boutique advisory firms. The work it covers is pitch books, trading comps, precedent transactions, CIMs, buyer lists, data rooms and fairness backup. Tools built mainly for the buy side, or for sales and trading, sit on our other lists.

We build Kepler and rank it second on purpose: for the deliverable itself, Rogo is further along. Facts about other tools come from their own websites, launch posts and public reporting as of November 2026, linked inline. Pricing appears only where it is published or publicly reported.

How we picked

We asked every tool the questions a banking team asks before rollout:

• Where does each number come from? Pulled from a filing or a licensed feed and computed in code, or written by a model.
• What does a click on a multiple open? The filing line, period and definition, or a document in general.
• Whose adjusted EBITDA? The company’s own reconciliation, a vendor’s standardized figure, or the bank’s definition.
• Does the output match the bank’s template? Pitch pages, comps sheets and models in house format.
• What data comes with it? Filings, transcripts, market data, private-company and deal data, and which of those need your own licenses.
• Does it work on live deals? Data rooms, the deal team’s files, and walls between deals.
• Does the same request give the same answer? A comps table that changes between runs can’t go in a book.

For the comps test in detail, see how to build trading comps with AI.

Quick comparison

Tool	Best for	Data it brings	Output	Where numbers come from
Rogo	Finished banker deliverables from one prompt	51 listed integrations, from S&P Capital IQ and FactSet to PitchBook and Preqin	Excel, PowerPoint, Word	Agents build from connected feeds
Kepler	Numbers a client will see	26M+ SEC filings with XBRL, 40M+ IR documents and transcripts, market data, your deal files	Excel, PowerPoint, Word, Claude, ChatGPT	Pulled from the filing and computed by code, cited to the line
Claude for Financial Services	A frontier assistant on licensed data	LSEG, FactSet, S&P Global, Morningstar connectors	Excel, PowerPoint, Word, Outlook	Model-written, linked to sources
ChatGPT for Financial Services	Banks standardizing on OpenAI	Daloopa, PitchBook, LSEG News, Crunchbase included	Firm templates in Excel, Word, PowerPoint	Model-written, with citations
Hebbia	CIMs, data rooms and buyer mapping	Filings, transcripts, partner feeds, Datasite rooms	Matrix grids, memos, decks	A model answers each cell, with sources
AlphaSense	Research, buyer universes and pitch decks from a library	Licensed library, Canalyst models, expert calls	PowerPoint and Excel add-ins	AI answers over library content; its agent writes and runs code
Model ML	Pitch books in house format	S&P Capital IQ, FactSet, PitchBook, Preqin	Excel, PowerPoint, Word, Outlook, Teams	Agent-built, checked by AutoCheck
Microsoft Copilot in Excel	Model mechanics inside the workbook	Connectors such as LSEG, FactSet, PitchBook, Daloopa	Excel	Workbook data and connected feeds
Datasite (Blueflame AI)	Sell-side processes run in Datasite	Live data-room content, Grata data	Inside Datasite; MCP for assistants	Answers drawn from the room
Daloopa	Source-linked historicals for models	Fundamentals and KPIs for 6,000+ companies	Excel agent, MCP	Extracted data points, each linked
Shortcut	An AI agent that builds in Excel	Filings and the workbook	Excel, Google Sheets	Agent-built formulas

The 11 best AI tools for investment banking

1. Rogo

Best for: coverage and M&A teams that want an agent to turn one prompt into a finished deck, model or document on licensed data.

Rogo was built for bankers, and its product shows it. Felix, which Rogo calls its financial agent harness, sits behind an Agent Library of hundreds of agents, each returning a finished Excel model, PowerPoint presentation or Word document from a single prompt. The banking set covers fairness-opinion-style DCFs and LBOs, public comps, precedent transactions, CIM analysis and PIBs, and teams can reshape any agent with their own precedents, benchmarks and house formatting. Rogo’s integrations page lists 51 sources, including S&P Capital IQ, FactSet, LSEG, PitchBook, Preqin and Daloopa, plus SharePoint, DealCloud and data rooms. Deal Room (August 2026) keeps models, diligence responses and presentations current as deal information arrives, and Rogo bought Rivanna for data-room diligence and Arvo for meetings. It was a launch partner for Copilot in Excel built for finance. Sacra reports pricing of about $3,300 per seat a year.

What users say: “Tend to just have it churn data out into a simple format, force it to put checks on everything to verify it did it correctly, then I have excel pull the rest,” wrote a Wall Street Oasis user in a thread about Rogo.

How Kepler differs: Rogo is built to produce the deliverable. We built Kepler around the numbers inside it: each one computed in code from a cited filing line, on each company’s own definitions. See Kepler vs Rogo and Rogo alternatives.

2. Kepler (our pick for numbers a client will see)

We build Kepler.

Best for: bankers and deal teams whose comps, precedents, fairness backup and pitch pages go in front of a client or a board, and who need every multiple to hold up when someone asks where it came from.

A banker’s AI problem is rarely the slide. It’s the number on the slide. A model writes figures as text, so an EV/EBITDA multiple can be off by a quarter’s worth of EBITDA and look fine. Vendor feeds standardize adjusted EBITDA, so the multiple in the book may not match the reconciliation the company published. LTM periods drift, fiscal years don’t line up, and a second run gives a different table the night before the meeting.

We built Kepler so that can’t happen. Code pulls and computes every figure, and the model never writes one. Each company’s adjusted EBITDA comes from the non-GAAP reconciliation in its own earnings release, summed over four quarters in code. The EV bridge builds from share count, price, debt, cash, minorities, preferreds and leases, each input cited to its line. Statements assemble from the filings’ own XBRL, as filed or as restated, and fiscal years calendarize with the rule written into the formula. Click any multiple and it opens a small sheet with the formula and each cited input. Ask again tomorrow and the table is identical, with any new quarter added.

The public record is already read: 26M+ SEC filings across 20+ form types (including 8-K exhibits, proxies and merger filings) and 40M+ investor relations documents and earnings call transcripts for 14,000+ companies in 120+ countries. Our forward-deployed engineers, who built this layer before at Palantir, also connect the bank’s own files read-only and build your firm’s ontology: coverage universes, clients, past mandates, prior books and the bank’s definitions, each resolved to one record. Agents refresh the comps the hour a peer reports and flag a restatement. Pages arrive in the bank’s PowerPoint template and models in Excel with live formulas, and the same engine answers inside Claude, ChatGPT and Slack or through our API.

Why it’s our pick for numbers a client will see

• Every multiple traces to a line item. The filing, the period and the formula, one click from the cell.
• Each filer’s own adjusted EBITDA. Taken from its published reconciliation, or the bank’s definition where the book needs one.
• The same table every run. Reproducible comps and precedents, with restatements flagged.
• Complete lists. “Every software deal since 2023 with a go-shop provision,” read from the merger agreements themselves.
• Your templates, built in. Comps sheets, football fields and pitch pages saved once in house style and applied every time.
• Walls between deals. One access check on every read, every query logged, SOC 2 Type II, our cloud or yours (security).

Questions bankers ask Kepler

• “Trading comps for US cybersecurity, EV/EBITDA on each company’s own adjusted EBITDA, LTM and NTM, every input cited.”
• “Precedent take-privates in vertical software since 2022 with premiums to the unaffected price and the termination fee in each merger agreement.”
• “Refresh the comps page in our Project Cedar book when any peer files its 10-Q, and flag what moved.”

“Kepler’s numbers match the filings. My team stopped re-checking every figure before it goes to a client.” (Vice President, Global Investment Bank)

Not the right fit if: your team mostly needs a licensed deal database and a live market terminal, or a writing assistant for emails and internal notes.

Try it on a peer set at app.kepler.ai, or book a demo and we’ll build comps on your coverage list.

3. Claude for Financial Services

Best for: banks that want a frontier model working across the data licenses they already hold.

Claude for Financial Services works inside Excel, PowerPoint, Word and Outlook, plus Google Docs, Sheets and Slides, with pre-built integrations for LSEG, FactSet, S&P Global and Morningstar. Anthropic offers agent templates for pitch books and credit memos as plugins in Claude and Claude Code, and its finance Agent Skills cover comps, DCF models, due diligence data packs, and company teasers and profiles for pitch books and buyer lists. Premium connectors run on the bank’s own license with each provider.

How Kepler differs: a frontier model on its own writes the figures and learns the bank’s definitions from what it’s given. Kepler runs inside Claude through MCP, so a banker can stay in the same conversation and get multiples computed in code and cited; Anthropic published a case study on how we built Kepler with Claude. For the head-to-head with Rogo, see Rogo vs Claude.

4. ChatGPT for Financial Services

Best for: banks whose teams already work in ChatGPT Enterprise.

OpenAI launched ChatGPT for Financial Services on September 10, 2026, starting with investment banking and equity research. It includes datasets from Daloopa, PitchBook, LSEG News and Crunchbase covering transcripts, financial statements, fundamentals and private companies, and more than 50 connectors, among them Datasite, Box, Preqin, FactSet and Intapp, reach the rest; OpenAI says it is working with S&P Capital IQ, LSEG, MSCI, Factiva and Moody’s on shared sign-in. Administrators publish Excel, Word and PowerPoint templates from an admin page, so teams can turn analysis into valuation models and pitch books in the bank’s format.

How Kepler differs: strong for drafting in a familiar interface, with the model still writing the numbers. Kepler runs inside ChatGPT too, with each figure computed and cited.

5. Hebbia

Best for: deal teams that want one set of questions run across a CIM, a data room or a peer set’s filings.

Hebbia’s investment banking page covers CIM creation, data-room analysis, comps, precedent benchmarking, origination signals and buyer and investor mapping. Its Matrix grid puts documents in rows and questions in columns; Matrix 2.0 (August 2026) carries the work through to a model, memo, deck or email, with a person signing off at each step. Max, which Hebbia describes as an AI team member that builds slides in house style, began reaching a small set of firms in July, and Headless Hebbia (October 1) opened an API and MCP server. Hebbia added Daloopa data in September and a Datasite partnership in October.

How Kepler differs: Matrix runs a model over the documents you select, question by question. We read every filing and deal document once at ingest, and Kepler computes each multiple in code. See Rogo vs Hebbia.

6. AlphaSense

Best for: bankers who research sectors, buyers and targets in a licensed library before building the deck.

AlphaSense’s banking page pitches buyer and target search across 40M+ entities, Canalyst models (part of AlphaSense since its Tegus acquisition) and in-document analysis of CIMs and data rooms. Its PowerPoint and Excel add-ins (July 2026) build decks of up to 40 slides in the firm’s template, update comps tables and precedent transactions, and, for early beta customers, build LBO, DCF and M&A models from firm templates. SuperAnalyst, its agent, entered early access in June 2026 and got a major release on October 6, with skills that include building a buyer universe.

What users say: a TrustRadius reviewer found it “not useful for generating a synthesis that reads through these reports and summarizes what the sentiment actually is in a reliable way. Still need to manually go through to develop a reliable synthesis.”

How Kepler differs: AlphaSense is built to search the content it licenses. Kepler is built around computing the numbers in the book from filings, on the bank’s definitions. Expert-call transcripts and broker research are licensed content we don’t carry, so teams that read them keep a library for that.

7. Model ML

Best for: banking teams that want an agent to produce long decks and linked models in house format.

Model ML calls itself an agent harness for financial services. It works by voice, text and email and inside Excel, PowerPoint, Word, Outlook and Teams, and its own example is handing the agent a target company and getting back a 50-plus page deck in house format. Agents build multi-tab workbooks from S&P Capital IQ, FactSet, filings and transcripts, with PitchBook, Third Bridge and Preqin among its data partners. AutoCheck, from its Flippr acquisition, reviews decks for numbers that don’t tie, factual errors and formatting. Model ML doesn’t publish pricing.

How Kepler differs: Model ML checks a finished deck. Kepler computes each figure in code before it reaches the deck.

8. Microsoft Copilot in Excel

Best for: analysts building and auditing models inside Excel on a Microsoft 365 license.

Microsoft’s June 2026 finance release added pre-built skills for DCF and comparable company models, custom skills written as SKILL.md files, Plan with Copilot, and attribution of Copilot’s edits in the Show Changes pane. Data arrives through connectors from LSEG, Moody’s, Daloopa, FactSet, Morningstar, PitchBook and S&P Global (through Kensho), some of which need separate licenses. Rogo is among the first skill partners.

What users say: “As far as having Copilot put together a full pitch deck, credit memo, model, or other deliverable, it’s a grand waste of time,” wrote a Wall Street Oasis user in investment banking.

How Kepler differs: Copilot is strong on workbook mechanics, with sourcing the inputs left to the analyst. Kepler sends cited inputs into the same workbook.

9. Datasite (Blueflame AI)

Best for: sell-side teams that run their processes in Datasite rooms.

Datasite acquired Blueflame AI in July 2025, and Blueflame now runs as Datasite’s AI agent for M&A. On April 28, 2026, Datasite launched an MCP server that lets Claude, ChatGPT, Copilot and Blueflame work on live content in Datasite Prepare and Diligence rooms with permissions intact. Sell-side teams can create rooms, organize content and manage permissions from their AI tool, and work through diligence Q&A.

What users say: “Work in PE and just got blue flame it’s pretty disappointing,” wrote a Wall Street Oasis user in private equity in late 2025.

How Kepler differs: Blueflame works inside the Datasite room. Kepler reads the room into a model of the deal, next to the target’s public peers and the bank’s prior work.

10. Daloopa

Best for: analysts who want historical fundamentals in the model with every figure linked to its source.

Daloopa covers 6,000+ public companies with up to 14 years of history. Scout, its Excel agent, builds three-statement models and post-earnings updates, and a read-only MCP server with OAuth sign-in links each data point to its source. Daloopa data is built into ChatGPT for Financial Services, Copilot in Excel, Rogo and Hebbia.

How Kepler differs: a data layer for another tool to reason over. Kepler builds the statements from XBRL with every cell cited, then computes the comps and pages on top. See Daloopa alternatives.

11. Shortcut

Best for: analysts who want an AI agent building directly in Excel or Google Sheets.

Shortcut is an Excel agent from Fundamental Research Labs, available as an Excel add-in, on the web, as a Windows app, as a CLI and in Google Sheets. Its banking use cases include accretion/dilution, comps from filings, LBOs and coverage-model updates, and in July 2026 it introduced its own model, Pivot. Published pricing runs from a free tier with 20 weekly credits to Pro at $100 a month billed annually and Teams at $320 a month plus $100 per seat.

How Kepler differs: Shortcut builds the workbook. Kepler supplies the figures that go into it, each computed and cited.

How to choose AI for a banking team

1. Run one live comps request. Use a real peer set and ask for EV/EBITDA on each company’s own adjusted EBITDA. Check the reconciliation behind three multiples.
2. Click ten cells. Each should open the filing line, period and units, with the formula for anything calculated. Our guide to checking AI-generated numbers lists the full set of tests.
3. Run it twice. A table that changes overnight can’t go in a book.
4. Load your template. The output should come back in your pitch-page format, footnotes included.
5. Ask for a complete list. Every precedent with a specific provision, read from the merger agreements.
6. Check the walls. Deal teams should see only their deals, with every query logged.
7. Count the licenses. Many tools run premium data on your own subscriptions; price the full stack.

For the agents behind these tools, see the best AI agents for finance, and for model building, the best AI financial modeling tools.

Build comps on your own coverage list at app.kepler.ai, or book a demo and we’ll show every multiple traced to its filing.

Frequently asked questions

What is the best AI tool for investment banking?

Rogo is the most complete banker product: hundreds of agents that turn a prompt into finished pitch books, models and documents on licensed data. For numbers a client will see, our pick is Kepler: every multiple in a comps table or precedent set traces to a line item in a filing, on each company's own adjusted EBITDA, computed in code and cited, and the pages come out in the bank's format. We build Kepler.

Can AI build a pitch book?

Several tools now draft one: Rogo, Model ML, AlphaSense, Hebbia and ChatGPT for Financial Services all produce decks in a firm's own template, and Claude has agent templates for pitch books. Drafting the pages is the easier half. The hard half is the numbers on them, so click every multiple and check that it opens the filing line, the period and the definition the bank uses.

Can ChatGPT or Claude build trading comps?

Both can assemble a comps table, and both reach market data through connectors such as FactSet, S&P Capital IQ, LSEG and Daloopa, usually on the bank's own licenses. On its own, a frontier model still writes or transcribes the figures, so each input needs checking. Kepler runs inside both assistants and returns comps where every multiple is computed in code from cited filing lines.

What is the difference between a banker AI agent and a data tool like Daloopa?

A banker agent such as Rogo or Model ML takes a request and returns a finished file: a deck, a model or a memo. A data tool such as Daloopa supplies historical figures, each linked to its source document, for another tool or analyst to work with. Kepler covers both jobs: the data layer, with every figure traced to the filing, and the finished comps, models and pages on top of it.

Is it safe to put client and deal information into an AI tool?

Ask every vendor how it keeps deal teams apart, whether permissions mirror your own systems, whether your data trains any shared model, and what it logs. Kepler runs one access check on every read with walls between deals, logs every query and calculation, holds SOC 2 Type II, never lets your data train models other firms use, and deploys in our cloud or yours.