For Analysts and Associates
Financial Modeling With Claude: A Practical Guide for Investment Bankers
The model is still yours. Claude just gets you to a clean first draft faster.
Anthropic's Claude can draft the skeleton of a three-statement model, write the Excel formulas for a DCF, flag a broken link in a merger model, and explain a sensitivity table in plain English before you walk into a review. This guide covers financial modeling with Claude across DCFs, comps, merger models, and LBOs, including where it saves real time and where every number still needs an analyst's sign-off. None of it replaces the modeling skill itself; it replaces the slowest, most repetitive parts of building the file.
Financial modeling with Claude means using it as a drafting and review layer inside the modeling workflow, not as the model's author of record. In practice that means asking Claude to structure a DCF or three-statement model, write the Excel formulas for a section, draft a comps set with sourcing rationale, build an LBO debt schedule, catch broken links or sign errors in an existing model, or explain a sensitivity table in client-ready language. The analyst still builds, tests, and ties out every formula before it reaches a page.
Published
Free lesson · no account neededYour AI Deal Desk: Where Claude Fits in Finance WorkPrivate Equity & Finance with Claude7 minStart the free lessonKey Takeaways
- Claude works best as a structuring and formula-drafting layer: it can lay out a DCF, three-statement model, or LBO debt schedule and write the Excel formulas, but the analyst builds and checks the live file.
- An AI DCF model or AI merger model from Claude is a first draft, not a finished deliverable; every link, circularity, and sign convention still needs to be traced by hand before it goes anywhere near a committee.
- Claude comps analysis is strongest at proposing a peer set and writing the sourcing rationale for each name; the multiples themselves still need to be checked against your data terminal.
- Accretion/dilution math, like any other output, should be explained by Claude but verified against the model's live cells, not trusted because the explanation sounds right.
- Confidential deal information, non-public financials, and anything covered by an NDA should never go into a general-purpose chat without your firm's data-handling policy and IT sign-off.
- Claude and ChatGPT both do financial modeling tasks reasonably well; the practical differences show up in formula precision, long-document handling, and how much the analyst has to re-check.
Does Claude or ChatGPT do better financial modeling?
| Task | Claude | ChatGPT |
|---|---|---|
| Writing Excel formulas | Handles multi-step formulas and explains the logic behind each reference, which makes it easier to spot a wrong row before pasting it in. | Writes comparable formulas; quality is similar, though it can be less explicit about why a reference was chosen. |
| Reading long source documents (CIMs, credit agreements, 10-Ks) | Stronger at holding a long document in context and pulling consistent figures from across it without re-explaining itself. | Capable on shorter excerpts; longer documents are more likely to need the material broken into chunks. |
| Model audits and error-checking | Good at walking through a structured checklist (balance sheet breaks, circularity, sign conventions) step by step. | Covers the same checklist logic; results are comparable when given the same level of detail about the model. |
| Explaining sensitivity tables and outputs in plain language | Reliable at producing a short, client-ready explanation when given the table's actual values. | Also reliable here; this is a strength of both tools when the underlying numbers are supplied correctly. |
| Comps set construction with sourcing rationale | Tends to give a clearer one-line rationale per peer, which is useful for defending the set in a review. | Produces similar peer sets; rationale can be shorter unless specifically asked for. |
Copy and Paste
Prompts You Can Use Today
Structure an AI DCF model before building it in Excel
I'm building a discounted cash flow model for a mid-size industrial company with a ten-year explicit forecast and a Gordon Growth terminal value. Walk me through the full model structure section by section: revenue build, margin assumptions, unlevered free cash flow bridge, WACC calculation, discounting, and terminal value, and tell me exactly which line items feed into which so I can lay out the Excel tabs in the right order before I start building formulas.
Swap the company type, forecast horizon, and terminal value method for your deal.
Draft the Excel formulas for a DCF section
Here is my unlevered free cash flow build with EBIT, taxes, depreciation, capital expenditures, and change in net working capital as separate line items on rows 10 through 15 of a worksheet. Write the Excel formula for unlevered free cash flow on row 16 that correctly adds back depreciation, subtracts capex, and subtracts the increase in net working capital, and flag anywhere my row references look like they might double-count an item.
Paste your actual row and column references so the formula matches your layout exactly.
Build a comps set with sourcing you can defend
I need a comparable company set for a specialty packaging manufacturer with mid-cap revenue and a regional customer base. Propose eight to ten public peers, group them by how close each one is on end market, scale, and margin profile, and for each name explain in one sentence why it belongs in the set so I can defend the selection if a managing director pushes back on any single peer.
Describe the target's sector, scale, and geography in your own words before sending.
Walk through accretion/dilution on a stock-for-stock merger
I'm modeling a stock-for-stock merger between an acquirer and a target where the acquirer is paying a premium to the target's current share price, funded entirely with new acquirer shares. Explain, step by step, how the exchange ratio, pro forma share count, and combined net income interact to produce either accretion or dilution to the acquirer's earnings per share, and tell me which input I should stress-test first if the deal looks only marginally accretive.
Replace the deal structure details with your actual merger assumptions.
Structure an LBO debt schedule and covenant checks
I'm building a leveraged buyout model with a term loan, a revolving credit facility, and subordinated notes, each with different amortization schedules and interest rates. Lay out how the debt schedule should flow from the cash flow available for debt paydown, including the correct order of mandatory amortization, cash sweep, and revolver draws, and list the covenant ratios I should be tracking each period so I build the checks into the same tab.
Swap in your actual tranche structure, rates, and amortization terms.
Audit a finished model for structural errors
I've finished a three-statement model with linked income statement, balance sheet, and cash flow statement, and the balance sheet isn't balancing in years three through five. Based on the most common causes of a balance sheet break in a linked model, walk me through the specific checks I should run in order, starting with the cash flow statement's financing section and the balance sheet's retained earnings roll-forward, before I start tracing individual cells.
Describe which statement is actually off-balance so the checklist targets the right place.
Explain a sensitivity table before a client call
I have a two-way sensitivity table showing implied enterprise value across a range of WACC assumptions on one axis and terminal growth rate assumptions on the other, and I need to explain the pattern to a client who isn't a modeling person. Summarize in plain language why enterprise value moves the way it does across the table, call out which corner of the table represents the most aggressive assumptions, and keep the explanation under four sentences.
From Anthropic
What Anthropic offers for Investment Banking
Advancing Claude for Financial Services
An October 2025 update adding Agent Skills for comparable company analysis, discounted cash flow models, due diligence data packs, company teasers and profiles for pitch books and buyer lists, earnings analyses, and initiating coverage reports, alongside a beta of Claude for Excel and connectors including LSEG and Moody's.
· Source
Agents for financial services
Ten agent templates released in May 2026, including a pitch builder, meeting preparer, model builder, and valuation reviewer, shipped as plugins in Claude Cowork and Claude Code, with Anthropic's Excel, PowerPoint, and Word add-ins made generally available.
· Source
In Practice
How Investment Banking Teams Use Claude
According to Anthropic's published customer story
Crunched
According to Anthropic's customer story, Crunched builds Excel tooling on Claude aimed at investment banking, private equity, and consulting users, and its customers report large time savings on modeling, financial spreads, and company write-ups.
Illustrative walkthrough
Building the Base-Case DCF
- Describe the business, the forecast horizon, and the terminal value method out loud to Claude before opening a spreadsheet.
- Get a structured outline of the model: the tabs, the order they depend on, and which line items feed which outputs.
- Build the formulas in Excel using that outline, asking for specific formulas only when the row and column layout is already in hand.
- Ask Claude to sanity-check the terminal value logic for any obvious circularity before the discounting step.
- Trace every link and reconcile each output cell by hand before the file goes anywhere near a review.
Illustrative walkthrough
Prepping a Comps Set Before a Pitch
- Describe the target's sector, scale, and customer base in plain language.
- Ask for a candidate peer list grouped by end market and margin profile rather than just industry classification.
- Request the one-sentence sourcing rationale for each proposed peer.
- Pull the actual trading multiples from the data terminal and match them against the proposed set.
- Drop any peer that doesn't survive scrutiny once the real data is in front of you.
Illustrative walkthrough
Walking Through an LBO Debt Schedule
- Describe the capital structure, tranches, rates, and amortization terms in plain language.
- Ask for the correct cascade order through mandatory amortization, cash sweep, and revolver logic.
- Build the schedule in Excel using that structure as the layout guide.
- Ask Claude to explain how a change in leverage or exit multiple would move the final equity returns.
- Confirm that explanation against the live model's own output before using it in a discussion.
Investment Banking and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| financial spreads 8+ hours to 1 hour | Reported by customers of Crunched, an Excel tool built on Claude, per Anthropic's customer story | Source |
| company write-ups 8 hours to 20 minutes | Reported by customers of Crunched, per Anthropic's customer story | Source |
| >50% time savings on Excel modeling | Reported by customers of Crunched, per Anthropic's customer story | Source |
The Core Workflow
What Financial Modeling With Claude Actually Looks Like
Four tasks cover most of where Claude earns its place in a modeling workflow.
Structuring the model before you build it
Describe the deal and the model type, and Claude can lay out the tabs, the order dependencies run in, and which assumptions feed which outputs before you open Excel.
Writing the formulas, not just describing them
Paste your actual row and column layout and ask for the formula, not a description of one. Specificity about your layout is what turns a vague answer into a usable one.
Auditing a model that's already built
A balance sheet that won't balance, a circular reference that's blowing up, a sign error buried three tabs deep: Claude can walk a checklist of likely causes faster than tracing cell by cell.
Explaining outputs to people who didn't build the model
Sensitivity tables, accretion/dilution bridges, and scenario outputs often need translating for a client or a committee. Claude is good at that translation when you give it the real numbers.
DCF and Three-Statement Models
Structuring an AI DCF Model and a Three-Statement Model
The highest-leverage use of an AI DCF model from Claude is at the structuring stage, before a single formula exists. Describe the business, the forecast horizon, and the terminal value method you're using, and ask Claude to lay out the full build in order: revenue drivers, margin assumptions, the unlevered free cash flow bridge, the WACC calculation, discounting, and terminal value. That sequencing decision, done wrong, is the single most common source of a DCF that looks fine but has a hidden circularity or a double-counted line item.
The same logic applies to a three-statement model. Claude is genuinely useful for mapping how the income statement, balance sheet, and cash flow statement should link, which cells are hardcoded assumptions versus formulas, and where the model is most likely to break when a forecast period changes. It is not useful as a substitute for building and testing the links yourself.
Where this breaks down is specificity. A vague request produces a vague model outline. The more detail you give about the actual business, the capital structure, and the forecast assumptions, the more the output looks like something you can actually build from rather than a generic template.
Anthropic's model builder template, part of the agent templates released for Claude Cowork and Claude Code, is built around exactly this workflow: structuring a model from a plain-language description of the deal before the formula-by-formula work begins.
Merger Models
Building an AI Merger Model and Walking Through Accretion/Dilution
An AI merger model from Claude is most reliable for the parts of the model that are mechanical but easy to get subtly wrong: the exchange ratio calculation in a stock deal, the pro forma share count, the combined income statement before and after synergies, and the accretion/dilution bridge that ties it all together. Ask Claude to walk through how each input moves the final EPS accretion or dilution figure, and you get a clearer sense of which assumption is actually driving the result before you present it.
The accretion/dilution AI explanation is where Claude earns its keep in a merger model review. A managing director asking "why is this only marginally accretive" wants a one or two sentence answer pointing at the driving assumption, not a re-derivation of the whole bridge. That's a translation task, and it's one Claude handles well when given the real pro forma numbers rather than a hypothetical.
What Claude should never be the last word on is the actual math inside the model. Exchange ratios, synergy phase-in schedules, and financing assumptions are deal-specific and often negotiated; an AI-drafted bridge is a starting point for the analyst's own build, checked line by line against the signed term sheet.
Comps and Precedent Transactions
Claude Comps Analysis in Practice
Comps work splits into two tasks, and Claude is stronger at one than the other.
Proposing and defending a peer set
Describe the target's sector, scale, and customer base, and Claude can propose a peer list with a one-line rationale per name, which is useful ammunition when a peer gets challenged in review.
Grouping peers by what actually matters
Claude is good at clustering comps by end market, margin profile, and growth rate rather than just by headline industry classification, which usually produces a more defensible set.
Drafting the write-up around the set
Turning a comps table into the short written rationale that goes in a pitch book is a drafting task Claude handles quickly, provided you supply the actual multiples rather than asking it to guess them.
What still needs the terminal
The multiples themselves, current trading data, and precedent transaction values have to come from your data provider. Claude comps analysis is a structuring and write-up layer, not a data source.
LBO Models
LBO Debt Schedules and Covenant Checks
Leveraged buyout models have more moving parts in the debt schedule than almost any other model type: multiple tranches, different amortization profiles, a cash sweep, and covenant ratios that need to be checked every period. Claude is useful here for laying out the order those pieces should flow in, specifically how cash available for debt paydown should cascade through mandatory amortization, then the cash sweep, before any revolver draw logic kicks in.
It's also useful for a sanity check on covenant logic: given a leverage ratio covenant and a fixed charge coverage covenant, what should trigger a breach, and where in the model should that check actually live so it updates automatically as assumptions change. That's a structuring question, and getting it wrong early means rebuilding the whole schedule later.
The returns math at the end of an LBO, the IRR and multiple of invested capital an exit scenario produces, is where an analyst's own model has to take over entirely. Claude can explain how a leverage change or an exit multiple assumption moves those returns, but the live model is the only source of truth for the actual number.
Where It Breaks and Who's Responsible
Model Audits, Confidentiality, and What the Analyst Still Owns
Every output described above is a draft, and the professional responsibility for the final number never moves off the analyst or associate who signs their name to the model. A formula Claude writes still has to be traced back to the source cells it references. A comps set Claude proposes still has to be checked against live trading data. An accretion/dilution explanation still has to match what the actual pro forma model produces, not just sound plausible.
Confidentiality is the other hard line. Non-public financial information, anything covered by an NDA, and deal specifics that haven't been publicly announced should never be pasted into a general-purpose chat interface without your firm's explicit data-handling policy, approved tooling, and IT sign-off. Where firms have adopted Claude for this kind of work, it's typically through controlled deployments, like the connectors and Excel integration described in Anthropic's update on Claude for Financial Services, not through an individual analyst's personal account.
Model audits deserve the same caution in reverse: a checklist of likely error sources is a useful starting point, but the only way to confirm a balance sheet actually balances, or a circularity actually resolves, is to trace the cells yourself. Treat every Claude-assisted audit as a faster way to find where to look, not as the finding itself.
Reading about financial modeling with Claude and being fast at it are different things
Working through a DCF structure, a merger model bridge, or an LBO debt schedule with an AI tool is a skill, and it's one you build by doing it, not by reading about it. Nightschool AI's free curriculum starts with hands-on exercises in exactly this kind of structured, formula-level prompting.
Frequently Asked Questions
Does Claude or ChatGPT do better financial modeling?
Both handle the common modeling tasks, structuring a DCF, writing Excel formulas, explaining sensitivity tables, reasonably well, and the differences are modest rather than dramatic. Claude tends to be stronger with long source documents like CIMs or credit agreements and more explicit about the reasoning behind a formula or a comp selection. ChatGPT performs comparably on shorter, well-scoped tasks. Neither tool replaces building and checking the model yourself, so the practical choice often comes down to which one your firm has approved and which interface fits your workflow.
Does Claude or ChatGPT do better financial modeling when building a three-statement model or DCF?
Both can lay out the structure of a three-statement model or a DCF reasonably well, and neither should build the live file for you. Claude tends to hold a long, detailed description of the deal more consistently across a multi-step structuring conversation, which helps when you're walking through the revenue build, margins, and free cash flow bridge in one sitting. ChatGPT performs comparably on shorter, well-scoped requests. The real differentiator is what you do next: build the formulas yourself, test every link, and reconcile the statements before the model goes anywhere near a review.
How do I get Claude to write actual Excel formulas instead of just describing them?
Give it your real row and column references, not a generic description of the model. Tell it exactly which cells contain which inputs and ask for the formula that goes in a specific cell. Vague prompts produce vague, non-pasteable answers; specific prompts with your actual layout produce formulas you can drop straight into the sheet and then verify.
Does Claude or ChatGPT do better financial modeling on an investment banking merger model?
Both tools can walk through a merger model's mechanics: the exchange ratio calculation, the pro forma share count, and the accretion/dilution bridge. Claude and ChatGPT are close enough on this task that the better choice often comes down to which one your firm has approved and which interface you already use day to day. Ask either one which input is driving the accretion or dilution result before you present the deal. The deal-specific inputs, synergy assumptions, financing terms, negotiated exchange ratios, still only your own model and the signed term sheet can verify.
Is it safe to paste confidential deal data into Claude?
Not into a general-purpose consumer chat interface, no. Non-public financials, NDA-covered information, and unannounced deal terms should only go through tooling your firm has specifically approved, with the data-handling and access controls that come with an enterprise deployment. Check with your firm's policy and IT before putting any real deal information into any AI tool.
Does Claude or ChatGPT do better financial modeling on comps analysis?
Both can propose a candidate peer set from a plain-language description of the target and draft the one-line rationale for each name. Claude tends to give a slightly clearer sourcing rationale per peer by default, which helps when you need to defend a single name in review, while ChatGPT produces a similar set with rationale that's sometimes shorter unless you ask for more detail. Neither tool is a data source: the actual trading multiples still have to come from your firm's data terminal before any comp goes in a pitch book.
Does Claude or ChatGPT do better financial modeling when explaining a sensitivity table to a client?
Either works well here, and this is one of the areas where the two tools are genuinely comparable. Give either Claude or ChatGPT the actual values from your sensitivity table and ask for a short, plain-language summary of the pattern and which corner represents the most aggressive assumptions. The output quality depends more on how much real data you supply than on which tool you use. Pick whichever one your firm has approved for this kind of client-facing language task.
Can Claude check my model for errors?
It can work through a checklist of likely causes for a specific symptom, a balance sheet that won't balance, a circular reference, a sign error, faster than tracing cell by cell from scratch. It cannot confirm the model is actually correct. Use it to narrow down where to look, then trace the actual cells yourself before trusting the fix.
Does Claude or ChatGPT do better financial modeling on LBO mechanics, like debt schedules and covenant checks?
Both can lay out the correct cascade for a debt schedule, mandatory amortization, then cash sweep, then revolver logic, and both can explain how leverage or coverage covenants should be checked each period. The difference between Claude and ChatGPT on this task is modest; neither replaces your own live model for the returns math at exit. Ask either tool to walk through the covenant logic before you build it into the schedule, then verify the actual IRR and multiple of invested capital in your own file.
Does Claude or ChatGPT do better financial modeling on accretion/dilution math, and should I trust it without checking?
No, not without checking, regardless of which tool produced it. Claude and ChatGPT are both reliable at explaining why a merger model is accretive or dilutive and which assumption is driving that result, but the actual figures need to match what your pro forma model produces. Treat any accretion/dilution explanation as a guide to where to look in your own build, not as a substitute for the calculation itself.
Claude in Investment Banking, by role
The model still has to be yours
Claude drafts faster. The tie-out is still the job.