30:00

AI Foundations for CFOs

From Technical Understanding to Daily Productivity
30-Minute Foundation Session
⏱️ 10 Minutes

The AI Journey: From Theory to Your Desktop

Why This Matters to You

Understanding AI's evolution helps you recognize why today's tools are fundamentally different from previous attempts. This isn't just another tech trend—it's the culmination of 70 years of research reaching a critical inflection point.

1950s
The Birth of AI
Alan Turing & Early Concepts
Click for details

Alan Turing & Early Concepts: The "Turing Test" - can a machine convince a human it's human?

Business Impact: Pure theory, no practical applications

1980s
Expert Systems Era
Rule-Based AI Systems
Click for details

Rule-Based AI: If-then logic systems for specific domains

Business Impact: Limited success in narrow fields like medical diagnosis

1990s
Machine Learning Emergence
Statistical Learning Algorithms
Click for details

Statistical Learning: Algorithms that could learn from data

Business Impact: Early recommendation systems, basic pattern recognition

2000s
Big Data & Computing Power
Infrastructure Revolution
Click for details

Infrastructure Revolution: Massive datasets + cloud computing

Business Impact: Google search, early personalization

2010s
Deep Learning Breakthrough
Neural Networks Renaissance
Click for details

Neural Networks Renaissance: Image recognition surpasses humans

Business Impact: Smartphone cameras, voice assistants, fraud detection

2017
The Transformer Revolution
"Attention Is All You Need"
Click for details

"Attention Is All You Need": Google's paper that changed everything

Business Impact: Foundation for all modern language AI

🚀 The 2020s: The Generative AI Explosion

What Changed Everything:

  • Scale: Models trained on internet-scale data (trillions of words)
  • Compute: Advanced GPU clusters capable of massive parallel processing
  • Architecture: Transformer models that understand context and relationships
  • Training Techniques: Methods to align AI with human preferences

Result: AI that can understand, reason, and create like never before

🔍 Key Technical Breakthrough: The Transformer Architecture

Before Transformers: AI processed text sequentially (word by word)

After Transformers: AI processes entire documents simultaneously, understanding relationships between all words

Old AI (RNNs)

Sequential processing

Limited memory

Struggled with long text

New AI (Transformers)

Parallel processing

Global attention

Handles entire documents

This is why ChatGPT can write coherent emails while older AI could barely complete sentences.

Knowledge Check #1:

What was the key breakthrough that enabled modern AI capabilities?

A) Faster computers and more data
B) Better algorithms for rule-based systems
C) The Transformer architecture enabling parallel processing
D) Larger neural networks with more layers
Section 1 of 4
⏱️ 12 Minutes

How Large Language Models Actually Work

The Executive Understanding

LLMs are sophisticated pattern recognition systems that learned to predict what comes next by studying virtually all human text. This simple concept enables remarkably complex capabilities.

🧠 What Happens Inside an LLM

Click on each step to understand how LLMs process your requests:

1
Tokenization
Breaking Down Language
Click for details

What Happens:

• Text is converted into "tokens" (pieces of words or concepts)

• Example: "CFO report" becomes ["CFO", " report"] tokens

• This allows the AI to work with standardized units of meaning

2
Neural Processing
Understanding Context
Click for details

What Happens:

• Each token gets analyzed in relation to every other token

• The model builds understanding of relationships and meaning

• This is where the "attention" mechanism works its magic

3
Pattern Matching
Drawing from Training
Click for details

What Happens:

• Compares current context to patterns seen in training data

• Identifies the most likely and appropriate next words

• Uses statistical probability to guide decisions

4
Output Generation
Creating Response
Click for details

What Happens:

• Produces text token by token, each informed by all previous context

• Can generate anything from emails to financial analysis

• Maintains coherence throughout the entire response

🔢 The Numbers Behind the Magic

GPT-4

~1.8 Trillion Parameters

Trained on ~13 trillion tokens

Context: 128k tokens (~100 pages)

Claude 3.5 Sonnet

~200 Billion Parameters

Optimized for reasoning

Context: 200k tokens (~150 pages)

Gemini Pro

~137 Billion Parameters

Multimodal capabilities

Context: 1M tokens (~750 pages)

For Context: The human brain has ~86 billion neurons. These models have trillions of connections.

🔍 Advanced Concept: Retrieval-Augmented Generation (RAG)

The Challenge: LLMs are trained on general knowledge, but you need AI that knows your specific company data, policies, and context.

Your Question
Search Company Data
Retrieve Relevant Info
Combine with LLM
Contextual Answer

Example: "What was our Q3 EBITDA margin?"

  • Without RAG: "I don't have access to your specific financial data"
  • With RAG: "Your Q3 EBITDA margin was 23.4%, up 1.2% from Q2, driven primarily by cost optimization in manufacturing operations..."

RAG makes AI truly useful for internal business applications by grounding it in your actual data.

Knowledge Check #2:

What makes RAG (Retrieval-Augmented Generation) valuable for business applications?

A) It makes AI responses faster
B) It allows AI to access and use company-specific data
C) It reduces the cost of using AI
D) It makes AI more creative

💡 Click to See: How LLMs Handle a CFO Request

🎯 What This Means for Your Daily Work

LLMs can handle any task that involves:

  • Analysis: Breaking down complex financial data and identifying patterns
  • Writing: Creating reports, emails, presentations in your voice and style
  • Reasoning: Working through multi-step problems and trade-offs
  • Translation: Converting between formats (data to insights, technical to business language)
  • Research: Synthesizing information from multiple sources

They excel at tasks that require understanding context, following instructions, and generating human-like output.

🔑 The Key Insight for CFOs

LLMs don't just follow rules—they understand intent. When you ask for a "financial summary for the board," the model understands:

  • Board-level means strategic, high-level perspective
  • Financial summary implies key metrics and trends
  • The format should be executive-ready
  • The tone should be professional and confident

And with RAG technology, it can access your actual financial data to provide real, actionable insights.

Section 2 of 4
⏱️ 8 Minutes

Your Internal AI Ecosystem

From Theory to Your Reality

Now that you understand how LLMs and RAG work, let's explore the AI applications available to you. This isn't theoretical—these are the tools you can use today for immediate productivity gains.

🏗️ Your AI-Powered Applications

Your organization uses Cortex as the secure AI foundation that powers user-friendly applications designed specifically for business users like you.

Your Request
AI Application
Cortex Platform
Best LLM
Response

How This Architecture Benefits You:

Easy-to-Use Applications: No technical knowledge required

Automatic Optimization: Apps choose the best AI model behind the scenes

Enterprise Security: Cortex ensures your data never leaves the organization

Compliance Ready: Built-in audit trails and governance

ARTIE

Your specialized AI application for Finance operations

Click for finance-specific features

Chat-in-a-Box

Create custom RAG applications for your team

Click for capabilities

Microsoft Copilot

Integrated across Office suite

Click for capabilities

Cortex Platform

The secure AI foundation (API only)

Click for technical details

Knowledge Check #3:

You need to analyze budget variances and identify cost optimization opportunities. Which tool should you use first?

A) Microsoft Copilot
B) ARTIE
C) Chat-in-a-Box
D) Cortex Platform directly

📈 Click to See: Daily CFO Scenarios with Your Tools

🎯 Choosing the Right Application for Your Tasks

Your Decision Framework:
  • Complex Financial Analysis: Use ARTIE for finance-specific workflows and insights
  • Team Knowledge Sharing: Create custom AI assistants in Chat-in-a-Box
  • Daily Office Tasks: Use Microsoft Copilot for emails, documents, and meetings
  • Document Analysis: Use ARTIE or Chat-in-a-Box depending on the complexity
  • Custom Business Solutions: Chat-in-a-Box for department-specific AI assistants

🚀 Your Realistic 30-60 Minutes Daily Time Savings

With your specific application ecosystem, here's the time you can reclaim:

  • Email & Communication (Copilot): 15 minutes daily
  • Financial Analysis (ARTIE): 25 minutes daily
  • Information Lookup (Chat-in-a-Box): 10 minutes daily
  • Strategic Research (ARTIE): 15 minutes daily

Total: 65 minutes of high-value time returned to strategic leadership and decision-making.

🎯 Your 4-Week Implementation Plan

Progressive adoption using your actual applications:

  1. Week 1: Start with Copilot for email management and document creation
  2. Week 2: Try ARTIE for one financial analysis task you do regularly
  3. Week 3: Create your first Chat-in-a-Box AI assistant with finance policies or historical data
  4. Week 4: Use ARTIE for more complex strategic analysis tasks

By month's end, you'll have experienced your complete AI application ecosystem and identified the highest-impact use cases for your role.

Section 3 of 4

AI Glossary for Finance Leaders

Your Quick Reference Guide

Every AI term explained in plain business language. Click on any term to see the definition, why it matters to you as a CFO, and real-world examples.

Artificial Intelligence (AI)
Software that mimics human intelligence

Simple Definition: Computer systems that can perform tasks that typically require human intelligence.

Why CFOs Care: AI can automate financial processes, provide insights from data, and support decision-making.

Example: Software that can read invoices, categorize expenses, and flag unusual transactions.

Machine Learning (ML)
AI that learns from data without programming

Simple Definition: A type of AI that improves its performance by learning from examples in data.

Why CFOs Care: ML systems get better over time, making more accurate predictions and classifications.

Example: Expense categorization that becomes more accurate as it processes more receipts.

Large Language Model (LLM)
AI trained on vast amounts of text

Simple Definition: AI systems trained on massive amounts of text that can understand and generate human-like language.

Why CFOs Care: LLMs can write reports, answer questions, analyze documents, and communicate in natural language.

Example: ChatGPT, Claude, or GPT-4 that can help draft board presentations or analyze financial documents.

Generative AI
AI that creates new content

Simple Definition: AI that can create new content like text, images, or code based on prompts.

Why CFOs Care: Can generate financial reports, create presentations, and produce analysis documents.

Example: AI that writes a complete variance analysis report from your financial data.

Transformer Architecture
The breakthrough that enabled modern AI

Simple Definition: A method for AI to process all words in a document simultaneously, understanding relationships between them.

Why CFOs Care: This breakthrough enabled AI to understand context and generate coherent, relevant responses.

Example: Why ChatGPT can write a coherent email instead of just random words.

RAG (Retrieval-Augmented Generation)
AI that searches your documents before responding

Simple Definition: A method that allows AI to search through your company's documents and data before generating a response.

Why CFOs Care: Makes AI responses specific to your company's actual data, policies, and context.

Example: AI that can answer "What was our Q3 EBITDA?" by looking at your actual financial reports.

Tokens & Tokenization
How AI breaks down text for processing

Simple Definition: The process of breaking text into smaller pieces (tokens) that AI can understand and process.

Why CFOs Care: Understanding tokens helps you know AI limitations and costs (often charged per token).

Example: "CFO report" becomes separate tokens ["CFO", " report"] for the AI to process.

Parameters
The "brain connections" in AI models

Simple Definition: The internal connections in an AI model that store learned information (like synapses in a brain).

Why CFOs Care: More parameters generally mean more capable AI, but also higher costs and complexity.

Example: GPT-4 has ~1.8 trillion parameters, making it very capable but expensive to run.

Context Window
How much text AI can "remember" at once

Simple Definition: The maximum amount of text an AI model can consider at one time when generating responses.

Why CFOs Care: Determines whether AI can analyze your entire financial report or just small sections.

Example: Claude can handle ~150 pages at once, while GPT-4 handles ~100 pages.

Prompt / Prompting
Instructions you give to AI

Simple Definition: The instructions, questions, or requests you give to an AI system.

Why CFOs Care: Better prompts lead to better results. Learning to prompt effectively is a key skill.

Example: "Analyze our Q3 variance report and highlight the top 3 concerns" is a clear, effective prompt.

Neural Networks
AI inspired by how brains work

Simple Definition: AI systems loosely inspired by how the human brain processes information through connected nodes.

Why CFOs Care: The foundation technology behind most modern AI capabilities.

Example: The underlying technology that enables AI to recognize patterns in your financial data.

Training Data
Information used to teach AI

Simple Definition: The massive amounts of text, images, or other data used to teach AI systems.

Why CFOs Care: Quality of training data affects AI performance and potential biases.

Example: LLMs trained on books, websites, and documents to learn language and knowledge.

Algorithm
Step-by-step instructions for AI

Simple Definition: A set of rules or instructions that tells a computer how to solve a problem or complete a task.

Why CFOs Care: Different algorithms are suited for different business problems.

Example: An algorithm that determines which invoices need manual review based on risk factors.

Multimodal AI
AI that works with text, images, and more

Simple Definition: AI that can understand and work with multiple types of input like text, images, audio, and video.

Why CFOs Care: Can process financial documents with charts, graphs, and mixed content types.

Example: AI that can analyze both the text and charts in your quarterly earnings presentation.

Attention Mechanism
How AI focuses on relevant parts of information

Simple Definition: A technique that allows AI to focus on the most relevant parts of input when generating responses.

Why CFOs Care: This is what makes modern AI so much better at understanding context and generating relevant answers.

Example: When analyzing a financial report, AI can focus on the variance section when answering questions about performance gaps.

Cortex Platform
Your organization's unified AI platform

Simple Definition: Your company's centralized platform that provides secure access to multiple AI models and tools.

Why CFOs Care: Ensures data security, compliance, and governance while giving access to best-in-class AI capabilities.

Example: One secure platform where you can access GPT-4, Claude, and other AI models without data leaving your organization.

Cloud Computing & GPU Clusters
The infrastructure that powers modern AI

Simple Definition: Massive networks of specialized computers (GPUs) accessed over the internet that provide the computing power needed for AI.

Why CFOs Care: Understanding the infrastructure helps with cost planning and vendor negotiations for AI services.

Example: The cloud infrastructure that allows you to run complex AI analysis without buying expensive hardware.

Turing Test
Classic test of machine intelligence

Simple Definition: A test proposed by Alan Turing where a machine is considered intelligent if a human can't tell they're talking to a machine.

Why CFOs Care: Helps understand how we measure AI capability and why modern AI feels so human-like.

Example: Modern AI like ChatGPT can often pass informal Turing tests in business conversations.

RNN (Recurrent Neural Networks)
Old AI that processed text word-by-word

Simple Definition: An older type of AI that processed text sequentially (one word at a time) and had trouble remembering information from earlier in long documents.

Why CFOs Care: Understanding RNN limitations helps you appreciate why modern Transformer-based AI (like ChatGPT) is so much better at understanding context.

Example: RNNs might forget the beginning of a financial report by the time they reach the end, while Transformers can consider the entire document simultaneously.

Section 4 of 4 - Reference