A primary-year funding banking analyst at a bulge bracket financial institution within the US prices $170k–$190k all-in. They spend most of their first yr formatting pitch books, constructing the identical DCF they constructed final month, reconciling accounts that may want reconciling once more in 30 days, and writing variance commentary that explains the previous to individuals who already lived by means of it.
The ratio of judgment to repetition skews closely towards repetition, and that ratio has not modified in a long time. In 2026 it’s beginning to change.
AI shouldn’t be good sufficient to exchange monetary judgement (but). However for the repetitive half of the job: the formatting, the primary drafts, the matching, the narrating, AI is now quick, correct and built-in sufficient to be genuinely helpful on the identical afternoon you set it up.
On this article, we’ll have a look at 4 sensible finance workflows the place Claude already exhibits sturdy promise as we speak: funding banking supplies, monetary modeling assist, month-end reconciliation, and variance evaluation. We’ll additionally have a look at the place it nonetheless wants human evaluate earlier than anybody ought to belief it in a critical workflow.
How Finance Groups Use Claude for Funding Banking Work
Funding banking runs on paperwork. CIMs, teasers, course of letters, purchaser lists, merger fashions, pitch decks. The work is actual and repetitive: an analyst constructing a one-pager for a deal teaser spends hours formatting, sourcing information, and structuring the identical 4 quadrants they constructed final week for a special firm.
Anthropic launched a devoted Funding Banking plugin for Claude Cowork on February 24, 2026. It’s open supply, free to put in, and offers Claude 7 slash instructions backed by 9 underlying expertise throughout three workflow classes: deal supplies, displays, and transaction assist. Fast terminology word because it comes up all through this information: expertise are the area information modules that activate robotically when related; instructions are the slash instructions you invoke explicitly. Every command calls a number of underlying expertise.
What it accommodates
Deal supplies: CIM drafting, teaser technology, course of letters, purchaser lists, and information pack extraction from current paperwork. Displays: strip profiles and pitch deck inhabitants utilizing your agency’s branded PowerPoint templates. Transaction assist: merger mannequin building and a deal tracker for reside milestones and motion objects.
Putting in it
The plugin requires Claude Cowork (desktop app, Enterprise plan or above) or Claude Code (Professional Plan or above.) Set up the financial-analysis core plugin first, it supplies the shared modeling instruments and all MCP information connectors that the IB plugin is dependent upon. Then add investment-banking on high.
By way of Claude Code:
claude plugin market add anthropics/financial-services-plugins
claude plugin set up financial-analysis@financial-services-plugins
claude plugin set up investment-banking@financial-services-plugins
By way of Cowork desktop: Settings → Plugins → Add market from GitHub → enter https://github.com/anthropics/financial-services-plugins → set up financial-analysis, then investment-banking.
/one-pager [Company Name] Generates a single PowerPoint slide with 4 quadrants: Overview, Enterprise, Financials, and Possession. Respects your current template’s margins and branding. That is the strip profile that populates pitch books and purchaser lists.
/cim [Company Name] Produces a full Confidential Data Memorandum: government abstract, enterprise overview, monetary evaluation, and market positioning sections. Claude drafts the construction and content material; your staff fills in proprietary information and tightens the narrative.
Remainder of the instructions so that you can attempt your self:
/teaser [Company Name] Generates an nameless one-page firm teaser for early-stage deal advertising. Identical core construction because the CIM however stripped of figuring out data.
/buyer-list [Company Name] Assembles a strategic and monetary purchaser universe. Claude categorizes potential acquirers by sort, sizes the match, and constructions the output for straightforward evaluate and prioritization.
/merger-model [Acquirer acquiring Target] Builds an accretion/dilution M&A evaluation. Output contains sources and makes use of schedule, professional forma financials, and sensitivity evaluation on buy worth and synergies.
/process-letter [Deal Description] Produces bid directions and course of correspondence for a reside transaction.
/deal-tracker Tracks energetic offers, milestones, and motion objects. A structured challenge administration view for reside mandates.
How you can get probably the most out of it
The plugin ships with generic methodology. The true worth comes while you customise the talent recordsdata to your agency: drop in your terminology, reference your branded PowerPoint template within the talent recordsdata, modify the CIM construction to your own home format. After that, each CIM draft, each one-pager, each purchaser record comes out in your voice.
Claude carries full context between Excel and PowerPoint in a single session. An analyst can run /merger-model, replace assumptions in Excel, then ask Claude to construct the abstract slide in PowerPoint with out switching instruments or dropping context. This cross-app workflow is in analysis preview for paid plans as of February 2026.
Sincere caveat
These instructions produce first drafts, not remaining deliverables. The CIM wants your agency’s proprietary market intelligence. The customer record wants your banker’s community information. The merger mannequin wants human verification of each assumption earlier than it goes to a shopper. Use these as the place to begin, not the completed product.
Utilizing Claude for Comparable Firm Evaluation, DCF Fashions and Valuation Outputs
Uncooked prompting whereas constructing monetary fashions produces output that appears right and isn’t. An analyst at a monetary modeling consultancy ran this check in January 2026: similar immediate to Claude for Excel and Excel’s Agent Mode. Claude’s mannequin had a cleaner structure and higher styling. It additionally discounted money flows utilizing a debt-to-equity ratio as a substitute of WACC, set the fairness threat premium at 120% as a substitute of 5-6%, and used a special discounting methodology for the terminal worth. It appeared investment-committee-ready and was arithmetically damaged.
That failure mode has a repair, and it’s the financial-analysis plugin.
Putting in it
The financial-analysis plugin can also be the muse for the IB plugin from part. Should you put in that already, you have got this too. If not:
claude plugin market add anthropics/financial-services-plugins
claude plugin set up financial-analysis@financial-services-plugins
As soon as energetic, you get two instructions plus MCP connectors to each main monetary information supplier.
/comps [Company Name]
Runs a comparable firm evaluation. Claude selects the peer group, pulls present buying and selling multiples from related information sources, builds the comps desk, and outputs a formatted Excel workbook with industry-standard construction. The peer choice is the one factor you evaluate and modify – that judgment can’t be automated. All the pieces else: pulled, calculated, formatted.
/dcf [Company Name]
Builds a full DCF. The plugin’s methodology layer is what makes this completely different from a uncooked immediate: it pulls the present authorities yield curve from LSEG to set the risk-free charge, retrieves historic fairness costs and beta to anchor the price of fairness, and checks for inner consistency earlier than outputting. The inputs are market-driven and traceable, not assumed.
What you continue to confirm each time: WACC inputs (fairness threat premium, beta, price of debt), that the discounting is constant throughout projected money flows and terminal worth, and that FCF is pulling from the correct line objects. The plugin prevents the plain failures. It doesn’t eradicate the necessity for a human to learn the mannequin. Wall Road Prep’s 2026 testing discovered that Claude hallucinated historic monetary information and each AI device scored zero on circularity dealing with: each dangers that persist no matter plugin.
Utilizing Claude in Excel with out slash instructions
The plugin instructions produce new fashions. Claude in Excel additionally works on fashions you have already got, and that is the place it earns time each day.
An analyst inheriting a 47-tab mannequin constructed by somebody who left the agency asks: “Clarify this complete spreadsheet to somebody seeing it for the primary time.” Claude traces each dependency chain and cites the precise cells. What used to take days of reverse-engineering takes an hour.
Situation evaluation runs conversationally. “What occurs if we delay all Q2 hires by one quarter?” Claude updates each affected cell, preserves the formulation, and exhibits the precise runway affect. You discover with out touching the mannequin construction. System debugging works the identical means: as a substitute of searching by means of cells, you get a direct rationalization of which cell is feeding the error, what format it expects, and the place the mismatch originates.
MCP connectors
When you have energetic information entitlements with S&P International, LSEG, Daloopa, PitchBook, Moody’s, or FactSet and have configured them in your Claude settings, they’re reside in Excel robotically. “Pull [Company]’s LTM income, EBITDA, capex, and web debt from Daloopa” populates the cells instantly. “Get the present 10-year authorities yield from LSEG” updates the risk-free charge reside. The guide export-format-paste step disappears.
The place to begin
Mannequin audit first. Add an current mannequin and ask Claude to clarify its construction, map the important thing assumptions, and flag components errors. That works as we speak with no plugin required and no threat of dangerous mannequin output. As soon as you’re comfy with how Claude reads your fashions, transfer to situation evaluation. Use /comps and /dcf final, and plan to confirm the monetary logic earlier than something goes to a shopper.
Utilizing Claude for Month-Finish Reconciliation
Account reconciliation sounds easy and destroys days. Each shut cycle, an accountant exports the GL steadiness, pulls the financial institution assertion or subledger element, manually matches transactions, investigates exceptions, paperwork the reconciling objects, and builds a workpaper for audit. Then AR. Then AP. Then intercompany. Then prepaids. By the point the working account is completed, it’s day three of shut.
Anthropic’s finance plugin (completely different from monetary evaluation plugin) ships with a structured reconciliation talent that understands the methodology and applies it constantly. It’s a separate plugin from the financial-analysis plugin utilized in sections 1 and a couple of, and lives in a special repository.
Putting in the finance plugin
claude plugin market add anthropics/knowledge-work-plugins
claude plugin set up finance@knowledge-work-plugins
Or through Cowork desktop: Settings → Plugins → Add market → https://github.com/anthropics/knowledge-work-plugins → set up finance.
As soon as put in, Claude has entry to 6 expertise: journal-entry-prep, reconciliation, close-management, financial-statements, variance-analysis, and audit-support. Every has a corresponding slash command.
Operating your first reconciliation
Drop your GL export and financial institution assertion into the Cowork challenge. Then run:
/reconciliation money 2026-02
Claude compares either side, calculates the distinction, and builds the workpaper. It categorizes every reconciling merchandise: timing variations that may clear subsequent interval, objects that want a journal entry, and exceptions that want investigation. It assigns getting older buckets and flags something over your materiality threshold.
Observe: For AR subledger reconciliation, use:
/reconciliation accounts-receivable 2026-02
The compounding curve
Month 1: Claude applies the generic methodology. Roughly 60% of things match robotically. You resolve the exceptions in the identical Cowork session: sort out the sample in plain language: “this vendor at all times settles two days after bill date,” “this intercompany cost posts to price middle 402 however needs to be 408,” “this financial institution price has no GL equal and needs to be flagged as a brand new journal entry.” Claude incorporates these explanations into the workpaper and carries the patterns into the following session.
Month 2: Claude applies what it realized. It handles 85% or extra of matches by itself. The exception record shrinks, and the objects it flags are genuinely uncommon.
Month 3: The reconciliation takes half the time it did in Month 1.
These numbers come from a single practitioner’s account (David Dors, Constructing Revenue, February 2026), not a managed benchmark. Deal with them as directional. The compounding sample is actual no matter precise percentages, each sample you educate Claude in Month 1 carries ahead.
With ERP connectors
In case your group has related NetSuite, SAP, or one other ERP through MCP, Claude pulls GL balances and subledger element robotically. With out connectors, you paste information or add recordsdata. The reconciliation works both means.
The sincere limitation
The finance plugin runs inside Cowork, which requires Claude Desktop to be open in your machine. In a single day batch reconciliations, high-volume AP matching, and ERP-native reconciliation throughout tons of of accounts want server-side infrastructure, not a desktop app. For that scale, purpose-built platforms are the correct instruments. They encode three-way matching logic, pay as you go amortization guidelines, and intercompany netting at a depth a general-purpose agent doesn’t.
What Claude’s plugin handles properly is the analyst-driven shut workflow: one accountant, a handful of key accounts, a month-to-month cadence the place the time financial savings compound. That’s most finance groups.
Each FP&A staff spends hours every shut cycle writing variance commentary. The true problem shouldn’t be quantity, it’s coherence throughout aggregation ranges. A vendor-level change flows right into a GL account, rolls into a value middle, and surfaces on the P&L line. The commentary at every stage must be constant and inform the identical story upward. Sustaining that consistency manually, throughout 4 enterprise items and two product strains, is the place time really goes.
AI helps with the drafting layer of that drawback, not the reason layer. Claude can generate structured first-draft commentary from a verified information desk, labeling variances, flagging materials actions, sustaining constant tone throughout sections. What it can’t do is clarify why a quantity moved with out being instructed.
The explanation behind a variance lives in your ERP, your CRM, your headcount system, and the judgment of the analyst who lived by means of the quarter. Claude produces coherent narrative from the information you feed it. The richer the context you present: prior commentary, GL element, price middle breakdowns, identified one-time objects, the extra helpful the draft.
Variance commentary continues to be price doing with AI. The drafting step is the one which consumes disproportionate time relative to its analytical worth, and that’s precisely the place Claude delivers.
Utilizing the finance plugin
When you have the finance plugin put in, run:
/variance-analysis opex 2026-02 vs price range
The plugin decomposes the variance into drivers, builds a waterfall chart, and produces commentary structured by class. For income variances, it breaks out worth and quantity results. For OPEX, it disaggregates by division and account. The waterfall goes instantly into your reporting package deal.
What the analyst really evaluations
AI-generated variance commentary has one particular failure mode: it narrates what the information says with out understanding what the information means. A 12% income miss within the West area may be a single account that closed late, a structural pipeline drawback, or a pricing choice that may reverse in Q2. Claude doesn’t know which one. The analyst does. That judgment is the one factor that can not be automated on this workflow.
The place purpose-built instruments have an edge
For groups with enterprise FP&A platforms, goal constructed instruments do variance detection plus narrative technology as a related workflow pulling actuals out of your ERP, working the calculation, and drafting commentary in a single step. If you’re already paying for considered one of these platforms, use them for this. They’re designed for it.
Claude’s benefit is for groups not able to undertake a full FP&A platform: the finance staff that runs on Excel, has entry to Claude by means of a broader enterprise settlement, and needs to chop commentary time this shut cycle with no new software program implementation.
The place to Begin
Choose one workflow. Not all.
In case your staff does deal work, set up the IB plugin and run /one-pager on a reside firm this week. If you’re in FP&A, take final month’s variance commentary, paste it into Claude with the present numbers, and see what comes again. If you’re in accounting, run one financial institution reconciliation by means of Cowork this shut cycle and examine the time.
Cheers.
