Claude becomes more useful in investment banking when the task stops looking like “build a DCF” and starts looking like a controlled work order: one deal, one purpose, approved sources, explicit definitions, a deliverable contract, tests, stop conditions, and a named human reviewer.
This Oria guide shows how to set up that operating model. It comes with a companion pack of 21 Markdown agents mapped to the six stages promoted in our original Investment Banking Guide: Originate, Value, Model, Diligence, Execute, and Communicate. Each agent has a narrow job, defined inputs, a repeatable method, controls, an output contract, and a handoff to the next reviewer.
The result is not an autonomous deal team. It is a practical way to give Claude bounded work, preserve evidence lineage, and move reviewed analysis toward decision materials and complex slides.
Start with a work order
A useful banking prompt should read like a staffing note. State the deal ID and audience; define the objective; name the approved sources and as-of date; specify currency, units, periods, and accounting conventions; list constraints; describe the required artifact; define the checks; and tell Claude when to stop.
For example: ask the Comparable Companies agent to screen candidates against five approved criteria, use one market-data cut, label every input reported, calculated, assumed, or not found, recalculate each multiple, and stop if source dates conflict. That is much safer and more reproducible than “find me good comps.”
The pack includes a reusable work-order template. It also shows when to save a stable procedure as a Claude Skill: after the team has run it manually, reviewed ordinary and edge cases, captured templates and definitions, and assigned an owner and version. Keep live deal data out of the Skill.
The six-stage agent system
Originate
Sector Screen Mapping · Buyer & Acquirer Lists · Thesis & Angle Development
Use Sector Screen Mapping to define and rank a market, Buyer & Acquirer Lists to build an evidence-backed counterparty universe, and Thesis & Angle Development to turn cleared research into a falsifiable deal thesis. These agents do not contact anyone or make a recommendation; they surface candidates, evidence, gaps, and counterarguments for banker review.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Value
DCF Valuation · Comparable Companies · Precedent Transactions · LBO Analysis
Use DCF Valuation, Comparable Companies, Precedent Transactions, and LBO Analysis to build transparent valuation views. Every input needs a source, as-of date, definition, and status. The agents produce versioned analytical support—not an approved valuation or lender view.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Model
Three-Statement Build · Sensitivity Tables · Scenario Toggle Design · Model Audit Review
Use Three-Statement Build for the forecasting backbone, Sensitivity Tables and Scenario Toggle Design for coherent ranges, and Model Audit Review as a read-only challenge. The reviewer must not silently repair the builder’s work. It returns exceptions to the builder and requires a fresh review after material changes.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Diligence
Data Room Synthesis · Earnings Quality Review · Risk Flag Register
Use Data Room Synthesis to create the source ledger, Earnings Quality Review to test the support for adjustments and cash conversion, and Risk Flag Register to consolidate reviewed issues. These are evidence and triage tools; they do not complete legal, tax, accounting, or quality-of-earnings diligence.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Execute
CIM Drafting · Management Presentation · Process Letter Drafting · Buyer Q&A Management
Use CIM Drafting, Management Presentation, Process Letter Drafting, and Buyer Q&A Management only after the controlling evidence and numbers are reviewed. Every claim should point to a source ID. Q&A remains draft until an authorized human approves and releases it.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Communicate
Deal Memo Drafting · Committee Recommendation · Committee Q&A Prep
Use Deal Memo Drafting, Committee Recommendation, and Committee Q&A Prep to structure the decision, alternatives, conditions, risks, and hardest questions. Claude can assemble the evidence-backed case; the committee and accountable bankers still decide.
Each agent is included in the downloadable 21-Agent Investment Banking Pack.
Where the research’s 13-agent architecture fits
Our safety research produced 13 broader control roles. The promotional materials are not internally consistent: the post text says 21 skills and names six stages; the image contains 20 child labels under five folders; and the current ZIP contains 21 files using the six-stage taxonomy. This pack follows the 21-file ZIP for continuity and maps it into the safer architecture: Sector Screen and Buyer Lists share the Buyer/Target Screener control model; the four valuation agents map to specialist builders plus an independent Valuation & Model Reviewer; Data Room, Earnings Quality, and Risk Register map to indexing, evidence extraction, and red-team review; the execution and communication agents map to Narrative, Deck QA, Process/Q&A, and Deal Red Team roles. Visual-only labels such as WACC Analysis, Merger Model, and Debt Capacity are either components of a governed workflow or require a separately approved specialist; they are not silently advertised as additional agents.
This is a taxonomy choice, not 21 autonomous people. The important separation is between evidence, analysis, review, narrative, and release.
From model or CIM to complex slides with Oria
Oria is the presentation layer. Once facts and model outputs are reviewed, an agent can prepare page briefs with an answer-first headline, supporting message, chart specification, source IDs, and caveat. Oria helps turn those approved briefs into beautiful, complex slides while preserving the banker’s control over content and judgment.
Run Deck QA after the first complete draft and after every material model update. Reconcile repeated numbers, dates, units, chart labels, source notes, confidentiality legends, and disclosures. No agent should publish, send, or release a client document.
Adjacent packs cover the neighbouring work: sixteen skills for pitch books, twenty-three finance agents for modelling, and the twenty-agent finance playbook.
The control rules that matter
Use only firm-approved environments and entitled sources. Treat nonpublic deal information as confidential and potentially MNPI. Keep each deal inside its permission boundary. Preserve an evidence ledger and exact versions. Separate reported facts, calculations, assumptions, and judgments. Never invent a missing value. Never let a builder be its only reviewer. Never parallelize agents that write to the same workbook or deck. Escalate conflicting sources, unstable models, unclear permissions, legal or accounting questions, and unsupported claims.
Download the guide for the full operating model, then install the companion pack for the 21 agents, routing rules, work-order template, and quick-start instructions.
Frequently asked questions
Are these Claude Skills or agents?
The post used “skills” as a broad capability label, its image listed 20 workstreams, and the existing ZIP contained 21 files. The companion replacement implements the 21 ZIP names as Claude Code subagents. In Claude’s product model, a Skill is a reusable procedure in a SKILL.md; an agent is a bounded worker in .claude/agents/ that can apply procedures and tools.
Can the agents edit my model or send client materials?
The distributed defaults are read-oriented and produce proposed analysis, logs, and drafts. They do not send communications, publish documents, contact counterparties, trade, or approve valuation, legal, tax, accounting, compliance, or release decisions.
How does Oria fit?
Oria turns reviewed evidence and page briefs into complex slides. It is not the source of valuation judgment and does not replace model, senior-banker, legal, compliance, tax, or accounting review.
Run the first controlled workflow
Start with one reversible internal task, a completed work order, frozen inputs, separate builder and reviewer roles, and a named human approval gate.
Claude helps the team find, test, and structure the answer. Oria helps turn the approved answer into complex slides.