How Investment Banking Analysts Can Use AI Coding Agents

Investment banking analyst using AI coding tools at workstation

THE NEW FRONTIER IN INVESTMENT BANKING

Why AI Coding Agents Are Revolutionizing Analyst Work

Investment banking analysts face a paradox: they're hired for their analytical prowess and financial acumen, yet they spend 60-80% of their time on mechanical tasks—copying data between systems, reformatting pitch books, debugging Excel formulas at 2 AM, and manually updating comps tables. This is where AI coding agents like Claude Code, GitHub Copilot, and similar tools are fundamentally reshaping the analyst experience.

Unlike generic AI chatbots, coding agents understand the technical infrastructure of finance: they can read and write Excel VBA, generate Python scripts for data manipulation, automate PowerPoint formatting, parse regulatory filings, and build complex financial models from scratch. They're not replacing analysts—they're eliminating the tedious execution work that prevents analysts from doing what they were actually hired to do: think strategically, identify opportunities, and provide insights.

The transformation is profound but often misunderstood. AI coding agents excel at what computers have always been good at—repetitive, rule-based tasks executed with perfect consistency—but now with natural language interfaces that don't require you to be a software engineer. An analyst can describe what they need in plain English, and the AI generates working code, complete with error handling and documentation.

What makes this particularly powerful in investment banking is the domain-specific knowledge these models possess. They understand concepts like DCF modeling, comparable company analysis, accretion/dilution calculations, and Bloomberg terminal data structures. They can read a 10-K filing and extract relevant financial metrics, build a merger model with synergies and pro forma adjustments, or create a VBA macro that reformats transaction comparables exactly how your MD prefers them.

The analysts who thrive in the next decade won't be those who can execute tasks fastest manually—they'll be those who can effectively orchestrate AI tools to amplify their capabilities. This guide explores exactly how to do that, with practical workflows, real examples, and honest assessments of where AI helps and where human judgment remains irreplaceable.

PRACTICAL APPLICATIONS

Core Use Cases: Where AI Coding Agents Add Immediate Value

The power of AI coding agents lies in their versatility across the entire analyst workflow. Let's examine the specific areas where they deliver measurable time savings and quality improvements.

Excel Automation & VBA

Generate complex macros, automate formatting, create dynamic formulas, and build custom functions without needing to master VBA syntax. Transform 2-hour manual tasks into 30-second scripts.

Financial Model Construction

Build DCF models, LBO models, merger models, and operating models with proper structure, error checks, and documentation. Get templates customized to specific deal parameters instantly.

Data Processing & Analysis

Extract data from PDFs, clean messy datasets, merge multiple sources, perform statistical analysis, and generate insights—all through Python scripts you don't need to write yourself.

Document Parsing & Extraction

Pull specific data from 10-Ks, earnings transcripts, credit agreements, and other documents. Convert unstructured text into structured datasets ready for analysis.

Visualization & Reporting

Create publication-ready charts, automate PowerPoint slide generation, build interactive dashboards, and format presentations to match bank standards.

Custom Tool Development

Build specialized calculators, comp screening tools, valuation multiples databases, and workflow automation utilities tailored to your team's specific needs.

The Excel Automation Revolution

Excel remains the primary tool in investment banking, but it's also the biggest time sink. AI coding agents transform how analysts interact with Excel, moving from manual cell-by-cell work to programmatic automation.

Real Example: Automating Comparable Company Analysis

Traditional approach: An analyst spends 3-4 hours manually updating a comp set—pulling data from CapIQ, reformatting numbers, calculating trading multiples, sorting by metrics, color-coding cells, and checking for errors.

With AI coding agents: The analyst asks Claude to "create a VBA macro that pulls data from cells A2:E50, calculates EV/EBITDA and P/E ratios, sorts by market cap descending, applies conditional formatting with green for top quartile and red for bottom quartile, and generates a summary statistics table." The entire process takes 5 minutes, and the macro can be reused indefinitely.

Beyond comps, common Excel automation tasks include:

Dynamic formula generation: Create nested IF statements, INDEX-MATCH combinations, or array formulas without memorizing syntax
Data consolidation: Merge data from multiple worksheets or workbooks with automatic reconciliation
Report formatting: Apply consistent styling, number formatting, and layout to match team templates
Error checking: Scan large models for circular references, broken links, or inconsistent formulas
Sensitivity tables: Build comprehensive data tables showing how outputs vary across input assumptions
Template customization: Modify standard templates for deal-specific requirements without manual rework

The key insight is that AI coding agents don't just speed up existing workflows—they enable workflows that were previously impractical. Creating a custom VBA macro used to require knowing programming and syntax; now it requires clearly articulating what you want to accomplish.

BUILDING SOPHISTICATED MODELS

Financial Modeling with AI: From Template to Custom Analysis

Financial modeling code and automation scripts

Financial modeling is where AI coding agents demonstrate their most sophisticated capabilities. They understand the structure, logic, and conventions of various model types, and can generate complete, working models or assist with specific components.

DCF Models: Beyond the Standard Template

Every analyst learns to build DCF models using standard templates, but real-world valuations require extensive customization. AI coding agents excel at adapting base templates to specific situations—seasonal businesses, companies with NOLs, international operations with multiple currencies, or complex capital structures.

Rather than spending hours manually adjusting formulas and layout, an analyst can describe the specific requirements: "I need a DCF model for a subscription software business with 90% revenue visibility, 15% churn, expanding gross margins from 70% to 75% over five years, and ESOP dilution of 2% annually. Include scenario analysis for churn rates between 10-20%."

The AI generates a complete model with:

Revenue build with recurring and new customer components
Churn impact calculations with cohort retention curves
Margin expansion trajectory with intermediate period logic
ESOP dilution flowing through to per-share values
Scenario tables showing sensitivity to key assumptions
Proper formatting, labeling, and documentation

LBO Models: Structuring Complex Transactions

LBO models involve intricate debt schedules, equity waterfalls, and returns calculations. AI coding agents can build these structures rapidly and accurately, handling details like:

Multi-tranche debt structures: Senior, subordinated, mezzanine, and PIK notes with different terms, covenants, and prepayment rules
Cash sweep mechanisms: Excess cash flow calculations with multiple use priorities (mandatory amortization, optional prepayments, distribution to equity)
Management rollover: Complex equity structures where existing management reinvests proceeds at different valuations
Transaction expenses: Proper treatment of financing fees, OID, and other upfront costs
Exit scenarios: Multiple exit routes (strategic sale, secondary buyout, IPO) with different valuation approaches
IRR waterfalls: Preferred returns, catch-up provisions, and carried interest calculations

Practical Workflow: Building a Custom Merger Model

An M&A analyst needs to model a stock-and-cash acquisition with earnout provisions, integration costs, and revenue synergies phasing in over three years. Instead of starting from scratch:

Step 1: Ask the AI to generate a base merger model template with standard accretion/dilution analysis.

Step 2: Request specific additions: "Add a section for earnout calculations based on achieving EBITDA targets of $50M year 1, $60M year 2, with maximum earnout of $100M paid 40% in cash and 60% in stock."

Step 3: "Include integration costs of $20M in year 1 and $10M in year 2, front-loaded in Q1 and Q2."

Step 4: "Model revenue synergies ramping from 0% in year 1 to 5% run-rate by end of year 3, with 40% EBITDA margin on synergies."

The AI builds each component with proper linkages, updates all dependent calculations, and maintains model integrity throughout. Total time: 20-30 minutes versus 4-6 hours manually.

Model Auditing and Error Detection

One underappreciated capability of AI coding agents is model auditing. They can review complex Excel models and identify issues that are easy to miss:

Circular references that may cause calculation errors
Hardcoded numbers that should be linked to assumptions
Inconsistent formulas across rows or columns
Missing error checks (division by zero, negative values where impossible)
Broken links to external files or worksheets
Formatting inconsistencies that could confuse users

An analyst can upload a model file and ask: "Review this LBO model and identify any structural issues, formula errors, or best practice violations." The AI scans the entire workbook and produces a detailed report with specific cell references and suggested fixes.

ADVANCED ANALYTICS

Data Processing and Analysis: From Raw Data to Insights

Investment banking generates enormous amounts of data, but it's often scattered across systems, trapped in PDFs, or formatted inconsistently. AI coding agents equipped with Python capabilities can transform data chaos into analytical clarity.

The Data Extraction Challenge

Consider a common scenario: analyzing historical M&A transactions in a specific sector. The data exists across multiple sources—press releases, regulatory filings, industry databases, news articles—but extracting and standardizing it manually is prohibitively time-consuming.

With an AI coding agent, an analyst can describe the task at a high level: "I need to extract M&A transactions in the healthcare IT sector from 2020-2025, pulling deal value, EBITDA multiples, buyer and target names, and deal rationale. The data is in various PDF investor presentations."

The AI generates a Python script that:

Reads multiple PDF files from a specified directory
Uses pattern matching to identify transaction details
Extracts relevant data points with error handling for missing information
Standardizes formats (company names, dates, currency)
Outputs a clean CSV file ready for analysis in Excel
Flags anomalies or data quality issues for manual review

Advanced Statistical Analysis

Beyond data extraction, AI coding agents enable sophisticated analysis that most analysts would struggle to perform manually. These capabilities include:

1

Regression Analysis for Valuation

Build multivariate regressions to understand which factors drive valuation multiples in a peer group. The AI can test different variable combinations, check for multicollinearity, and produce summary statistics with R-squared and p-values.

2

Time Series Forecasting

Apply forecasting models to predict revenue, margins, or other metrics based on historical patterns. The AI can implement moving averages, exponential smoothing, or more sophisticated ARIMA models depending on data characteristics.

3

Cohort Analysis

For subscription businesses, analyze customer cohorts to understand retention, expansion, and lifetime value patterns. The AI generates cohort tables, survival curves, and visualizations automatically.

4

Benchmarking and Peer Comparison

Pull public company data from multiple sources, standardize metrics, calculate percentiles and quartiles, and identify outliers. The AI handles the entire data pipeline from extraction to visualization.

Working with Unstructured Data

Much of the valuable information in investment banking exists in unstructured formats—earnings call transcripts, management presentations, credit agreements, and regulatory filings. AI coding agents can parse these documents and extract structured insights.

For example, analyzing sentiment in earnings call transcripts across a peer group: the AI can process dozens of transcripts, identify mentions of key topics (supply chain, pricing power, M&A), quantify sentiment (positive, negative, neutral), and summarize the findings. This type of analysis—once the domain of specialized NLP researchers—is now accessible to any analyst who can describe what they're looking for.

Compliance Note: When working with non-public information or material non-public information (MNPI), ensure all AI interactions comply with your firm's information barriers and data handling policies. Most major banks have specific guidelines for AI tool usage with sensitive data.

Automated Report Generation

Once analysis is complete, AI coding agents can generate formatted reports, presentations, or data visualizations automatically. An analyst might request: "Create a PowerPoint slide deck summarizing this M&A analysis, with a title slide, executive summary, transaction list table, valuation multiples chart, and key findings. Use our standard template formatting."

The AI generates the entire deck with appropriate fonts, colors, layouts, and content organization—work that would traditionally take hours of manual PowerPoint formatting. While final polish and customization will still require human review, the bulk of execution work is automated.

IMPLEMENTATION STRATEGIES

Practical Workflows: Integrating AI into Daily Operations

Understanding what AI coding agents can do is different from effectively integrating them into your workflow. Success requires developing specific patterns and practices that maximize efficiency while maintaining quality control.

The Iterative Refinement Approach

The most effective way to work with AI coding agents is through iterative refinement rather than expecting perfection on the first attempt. Start with a basic request, review the output, then progressively add requirements and adjustments.

Example Workflow: Building a Custom Valuation Dashboard

Initial Request: "Create a Python script that reads an Excel file with company financials and calculates EV/EBITDA and P/E multiples."

Review: The AI generates working code but uses basic formatting. Test with sample data.

Refinement 1: "Add error handling for missing data and negative EBITDA. Flag companies with N/A multiples."

Refinement 2: "Create a summary statistics section showing median, mean, and quartiles for each multiple across the peer group."

Refinement 3: "Generate a scatter plot showing EV/EBITDA vs. EBITDA growth rate, with each company labeled."

Refinement 4: "Export results to a formatted Excel file with separate tabs for raw data, statistics, and charts."

Each iteration builds on the previous version, gradually adding sophistication until the tool meets all requirements. This approach is faster and more reliable than trying to specify everything upfront.

Common Workflow Patterns

Morning Data Updates

Many analysts start their day updating pitch materials with latest market data, comp multiples, or transaction activity. AI coding agents can automate these repetitive refreshes:

Pull latest stock prices and calculate updated trading multiples
Refresh transaction comparables with new deals announced overnight
Update market context slides with overnight index movements and sector performance
Regenerate sensitivity tables with revised assumptions

Ad Hoc Analysis Requests

When a senior banker asks for quick analysis ("Can you show me how leverage ratios have trended in our target's peer group over the last five years?"), AI coding agents enable same-day turnaround:

Describe the analysis needed in natural language
Generate code to pull relevant data from your databases or public sources
Create visualizations showing the trends clearly
Export to PowerPoint with appropriate formatting

Quality Control and Verification

Before presenting any model or analysis, use AI coding agents for systematic quality checks:

Verify that all formulas are consistent across time periods
Check that balance sheet balances and cash flow statement reconciles
Confirm that sensitivities behave logically (higher discount rate = lower value)
Scan for common errors like circular references or #REF! errors

Effective Prompt Engineering for Finance

The quality of AI-generated code depends heavily on how you phrase requests. Effective prompts in investment banking context include:

Less Effective Prompt	More Effective Prompt
"Make a DCF model"	"Create a 3-statement DCF model for a SaaS company with revenue growing 25% annually, expanding EBITDA margins from 15% to 25%, CapEx of 3% of revenue, and NWC changes of 10% of revenue growth. Use 10% discount rate and 8x terminal EBITDA multiple."
"Fix this Excel formula"	"This SUMIFS formula in cell D15 is returning #VALUE! error. It should sum revenue from table A2:B50 where the date in column A is in Q1 2024. The date format is MM/DD/YYYY."
"Analyze this data"	"I have quarterly revenue data for 15 companies from 2020-2024. Calculate year-over-year growth rates, identify the fastest and slowest growers, and create a chart showing growth trends. Flag any quarters with negative growth."
"Create a chart"	"Generate a bar chart comparing EV/EBITDA multiples for 10 companies, sorted from highest to lowest. Color bars above median in blue and below median in gray. Include a horizontal line showing the median value."

Key principles for effective prompts:

Be specific about numerical parameters, time periods, and formats
Provide context about the business, industry, or transaction
Specify expected output format (Excel file, Python script, chart type)
Mention edge cases or error conditions to handle
Reference standard methodologies or conventions when applicable
Include sample data or desired output structure when possible

CRITICAL CONSIDERATIONS

Limitations, Risks, and Best Practices

While AI coding agents are powerful tools, they're not infallible. Understanding their limitations and implementing proper safeguards is essential for professional-grade work.

Where AI Coding Agents Fall Short

AI coding agents excel at execution but struggle with several aspects of investment banking work that require human judgment:

Strategic Judgment

AI can build a model, but it can't tell you whether a 6.5x EBITDA valuation makes strategic sense for this specific buyer given their market position, integration capabilities, and growth objectives. That requires understanding business context AI doesn't possess.

Qualitative Assessment

Evaluating management team quality, cultural fit in mergers, competitive moat sustainability, or strategic alternatives requires human insight. AI can summarize information but can't make nuanced judgments about people and organizational dynamics.

Novel Situations

When dealing with truly unique circumstances—an unprecedented capital structure, a first-of-its-kind transaction, or novel regulatory environment—AI coding agents may apply standard approaches that don't fit. Human creativity is essential for developing new frameworks.

Relationship Dynamics

Understanding client preferences, navigating sensitive negotiations, reading body language in meetings, or managing stakeholder politics are inherently human domains. AI is a tool for analysis, not for relationship management.

Data Privacy and Compliance Considerations

Investment banks handle highly sensitive information subject to strict regulatory and confidentiality requirements. When using AI coding agents:

Information barriers: Ensure AI tool usage doesn't inadvertently breach information barriers between deal teams or between public and private side operations
MNPI handling: Never input material non-public information into external AI services without explicit approval from compliance
Client confidentiality: Be cautious about inputting client-specific data, deal terms, or proprietary information into cloud-based AI services
Data retention: Understand where your inputs are stored, for how long, and who has access to them
Regulatory scrutiny: Document your AI-assisted work appropriately for regulatory reviews and internal audits

Best Practice: Many banks are deploying internal AI tools that operate within the firm's security perimeter. Use these when available for sensitive work, and reserve external AI services for general-purpose tasks with no confidential information.

Quality Control Framework

AI-generated code and models require systematic verification before use in client-facing work or investment decisions:

Verification Checklist

Test with known inputs: Run the model with scenarios where you know the correct answer to verify logic
Sense check outputs: Do the results align with your intuition and domain knowledge?
Review formulas manually: Spot check key calculations to ensure they match intended methodology
Check edge cases: Test with extreme values, negative numbers, zero values, and missing data
Compare to benchmarks: How do the results compare to similar analyses you've done previously?
Document assumptions: Ensure all key assumptions are clearly stated and justified
Peer review: Have another analyst review AI-generated work, especially for high-stakes analyses

Skill Development in the AI Era

A common concern is whether relying on AI coding agents will atrophy analysts' technical skills. The answer depends on how you use these tools:

The wrong approach: Blindly accepting AI outputs without understanding the underlying logic, becoming dependent on automation without developing fundamental skills.

The right approach: Using AI to accelerate learning by seeing well-structured examples, freeing time to understand concepts more deeply, and focusing human effort on strategic thinking rather than mechanical execution.

Early-career analysts should still learn to build models manually to understand proper structure, error checking, and best practices. But once those fundamentals are solid, AI coding agents become force multipliers rather than crutches.

Managing Expectations

AI coding agents are improving rapidly, but they still require human oversight and iteration. Set realistic expectations:

First attempts often need refinement—budget time for iteration
Complex tasks may require breaking into smaller steps
You still need to understand finance and modeling principles
Code may need debugging or optimization for specific use cases
Final presentation and communication remain human responsibilities

The analysts who succeed with AI coding agents are those who view them as collaborative tools requiring guidance and judgment, not autonomous systems that work unsupervised.

LOOKING AHEAD

The Future of AI in Investment Banking

AI coding agents are still in early stages, and their capabilities are expanding rapidly. Understanding emerging trends helps analysts prepare for the evolving landscape.

From Coding Assistance to End-to-End Automation

Current AI coding agents primarily assist with specific tasks—writing a VBA macro, building a model component, analyzing a dataset. The next generation will handle entire workflows from start to finish.

Imagine describing a deal scenario and having AI generate a complete pitch book: pulling comparable transactions, building valuation models, creating financing structures, drafting management presentation materials, and producing formatted slides ready for review. The technology is moving in this direction, though human oversight and strategic guidance will remain essential.

Multimodal Capabilities

Future AI systems will seamlessly work across modalities—reading documents, interpreting charts, understanding spoken instructions, and generating deliverables in whatever format is needed. An analyst might verbally describe an analysis during a meeting, and the AI produces a model reflecting that discussion by the time the meeting ends.

Real-Time Market Integration

As AI systems gain API access to market data, terminal systems, and internal databases, they'll provide real-time analysis that's always current. Rather than manually updating decks with overnight market movements, your materials will refresh automatically with context about what changed and why it matters.

Personalization and Learning

AI coding agents will increasingly learn individual and team preferences—how you format models, which assumptions you typically use, what level of detail your MDs prefer. They'll adapt their outputs to match your specific style and requirements without explicit instruction.

Implications for Analyst Roles

These advances will continue shifting analyst work from execution to strategy and judgment. The value-added components of the role—industry expertise, relationship building, creative deal structuring, persuasive communication—will become even more central.

Analysts who embrace AI tools and develop skills in effectively directing and supervising AI-generated work will have significant advantages. Those who resist the transition or fail to maintain critical thinking skills will struggle.

The most successful investment banking professionals in the next decade will be those who combine deep financial expertise with proficiency in orchestrating AI tools to amplify their capabilities. The technical barrier to entry is rising—but for the right reasons. Rather than manual Excel proficiency, success requires understanding how to leverage AI to solve complex problems at scale.

IMPLEMENTATION GUIDE

Getting Started: Your First Steps with AI Coding Agents

Understanding the potential of AI coding agents is one thing; actually integrating them into your workflow requires a deliberate approach. Here's a practical roadmap for investment banking analysts to get started effectively.

Phase 1: Identify High-Impact Use Cases (Week 1)

Don't try to revolutionize everything at once. Start by identifying 2-3 specific tasks that are:

Highly repetitive (you do them weekly or more frequently)
Time-consuming (take more than 30 minutes each time)
Rule-based (follow consistent logic rather than requiring heavy judgment)
Low-risk (not client-facing in final form, or easy to verify)

Examples of good starting points: reformatting data exports from CapIQ, updating weekly comps tables, generating standard charts, or calculating trading statistics.

Phase 2: Experiment and Learn (Weeks 2-4)

1

Choose Your Tool

If your bank provides internal AI coding tools, start there for compliance reasons. Otherwise, explore Claude, GitHub Copilot, or other coding-focused AI assistants. Many offer free trials.

2

Start with Simple Requests

Begin with straightforward tasks like "Write a VBA macro to format this table" or "Create a Python script to merge these two CSV files." Build familiarity with how to phrase requests and interpret outputs.

3

Test Thoroughly

Run AI-generated code on sample data first. Verify outputs match your expectations. Debug any issues—this helps you understand how the code works and builds confidence.

4

Document What Works

Keep a record of successful prompts and workflows. Building a personal library of proven approaches accelerates future work and helps teammates get started.

Phase 3: Scale and Integrate (Months 2-3)

Once comfortable with basics, expand to more sophisticated applications:

Build reusable tools: Create VBA macros or Python scripts you can apply repeatedly across different deals
Automate workflows: Chain together multiple steps—data extraction, analysis, visualization, and reporting
Customize for your team: Develop tools that match your group's specific templates, standards, and preferences
Share knowledge: Teach teammates what you've learned, creating multiplier effects across your team

Common Pitfalls to Avoid

Over-Reliance Without Understanding

Never use AI-generated code you don't understand, especially in client work. Take time to review the logic and ensure it matches your intent.

Skipping Verification

Always test with known scenarios before trusting outputs. AI can make subtle errors that aren't immediately obvious but produce incorrect results.

Ignoring Compliance

Don't input confidential information into external AI services without approval. The time saved isn't worth regulatory violations or client trust issues.

Perfectionism Paralysis

Don't wait for AI to generate perfect code on the first try. Embrace iterative refinement—start with something functional and improve it progressively.

Resources for Continued Learning

As AI coding agents evolve rapidly, staying current is important:

Follow your AI tool's release notes and new feature announcements
Join internal knowledge-sharing forums if your bank has them
Experiment with new capabilities as they're released
Connect with other analysts using AI tools to share techniques
Maintain foundational skills in Excel, Python, and financial modeling

Measuring Your Impact

Track the time savings and quality improvements from AI adoption to demonstrate value and refine your approach:

Time saved: How many hours per week are you saving on previously manual tasks?
Quality improvements: Are you catching more errors? Performing more thorough analysis?
Capacity expansion: Can you take on more complex projects or additional deal coverage?
Work-life balance: Are you leaving the office earlier while maintaining output quality?

Most analysts report saving 10-20 hours per week once they've fully integrated AI coding agents into their workflows—time that can be reinvested in higher-value activities like strategic thinking, client interaction, or personal development.

Final Thought: AI coding agents are not replacing investment banking analysts—they're eliminating the parts of the job that shouldn't require human intelligence in the first place. Embrace these tools to focus your energy on the work that actually matters: understanding businesses, structuring deals, and serving clients with insight and judgment that only humans can provide.