Explore how artificial intelligence is transforming the traditionally manual world of mergers and acquisitions financial analysis.

Derek shares how Socratic AI is solving a massive pain point for investment bankers and M&A advisors who spend countless hours cleaning up messy financial data from private companies. From Excel spreadsheets to PDF bank statements, Derek explains how his team uses a sophisticated combination of LLMs, pattern matching, and custom algorithms to normalize chaotic financial documents into professional-grade models.

This conversation dives deep into the technical challenges of parsing tabular financial data, the strategic decisions around when to use different AI models, and how the latest reasoning models are being applied to spot financial anomalies that could impact multi-million dollar deals.

Takeaways

• The M&A Data Problem: Private company financials are often messy and unstructured, requiring hours of manual cleanup before analysis can begin
• Smart Model Selection: Success comes from using the right AI model for each specific task – not just throwing everything at the most powerful LLM
• OCR vs. LLM Trade-offs: Even with advanced models, extracting tabular data from PDFs remains challenging and requires hybrid approaches
• Reasoning Models in Action: New reasoning capabilities are being used to hunt for financial anomalies and errors that could cost millions
• The Ferrero Rocher Effect: Foundation models are just the “peanut in the center” – the real value comes from all the layers around it (workflow orchestration, domain expertise, user experience) that create the full delicious experience
• The Vertical SaaS Advantage: The real value isn’t in the AI models themselves, but in orchestrating multiple models into domain-specific workflows
• Productivity Multiplier: Small AI-native teams can now accomplish what would have required 10x more people just a few years ago

Sound Bites

• “We use a combination of pattern matching, rules, and large language models to interpret and standardize financial data – you can’t just throw it into ChatGPT and get an output.”
• “If a column gets off by one in financial data, you’ve screwed up the entire thing – the integrity of that table needs to be maintained.”
• “I feel like I’m ten people now and I’m doing the job of what would have been 10 people.”
• “Investment banking analysts work 80-100 hour weeks because they’re going cell by cell, formula by formula – we can set an AI that doesn’t get tired to do that type of deep thinking.”
• “The foundation model is just the peanut in the center – everything around it is the deliciousness that adds to the whole Ferrero Rocher.”

Chapters

00:00 Introduction and Socratic’s AI Overview 
01:46 The M&A Analyst Workflow Problem 
04:18 Types of Financial Documents and Data Sources 
05:21 AI Techniques for Data Normalization 
07:02 Choosing Between LLMs and Algorithms 
08:32 PDF Processing and OCR Challenges 
11:55 Post-Normalization Analysis and Features 
14:45 Rule-Based vs AI-Driven Analysis 
17:16 Reasoning Models and Parallel Processing 
21:20 Visual Reasoning Capabilities 
23:45 The “Wrapper” Debate and Value Creation 
26:39 AI Tools the Team Uses Daily 
29:47 Prototyping Tools and Workflow Evolution 
34:29 Future Roadmap for Socratic’s AI 
37:14 Personal Values and Work-Life Balance 
38:43 How to Connect and Get Involved

Connect with us
Where to find Derek
Website: Socratic.ai
LinkedIn: https://www.linkedin.com/in/bomanderek/
Company LinkedIn: Socratic.ai

Where to find Sani: 
LinkedIn: https://www.linkedin.com/in/sani-djaya/
Get in touch: [email protected]

#AI #MachineLearning #MergersAndAcquisitions #FinTech #StartupTech #LLM #ReasoningModels #VerticalSaaS #FinancialAnalysis #InvestmentBanking