Choosing the Right AI Model

Welcome to This Week’s Issue!

With the rapid evolution of Large Language Models (LLMs), companies now have a wide range of AI tools to choose from. Picking the right model impacts costs, security, compliance, and long-term strategy.

Should you go with a closed-source model like GPT-4 for convenience, or an open-source model like FinLLaMA for flexibility? Should your AI run in the cloud, or does on-premises hosting offer better control?

In this issue, we’ll break down:
✅ The key differences between open-source and closed-source LLMs
✅ How specialized finance AI models can improve accuracy
✅ Cost and infrastructure considerations when selecting an LLM

By the end, you’ll have a clearer framework for choosing the best AI model for your finance operations—and the right questions to ask before making an investment.

Quick Favor: Help Shape the Future of AI in Finance! 🚀

We’re building AI Copilots to streamline finance workflows—and I’d love your input!

✅ Which finance tasks take up the most time in your day?
✅ Are you already using AI in your workflow?
✅ Want an AI agent built for you—free?

➡️ Take 2 minutes to answer the poll

If you can, please share it with your peers—the more insights, the better! Thanks for being part of this journey. 🚀

The Balanced View: Understanding the Foundations of AI Models

Artificial Intelligence (AI) is evolving rapidly, with companies competing to develop increasingly powerful models. Major players like OpenAI, Google DeepMind, Anthropic, Mistral, Meta, and others are continuously pushing the limits of AI capabilities.

For finance professionals, understanding which AI model powers an automation tool or custom automation is essential. Different models possess unique strengths, biases, and limitations.

The Basics: What is an AI Model?

An AI model is a trained algorithm that processes data to recognize patterns, make decisions, or generate predictions. Unlike traditional software, which follows predefined rules, AI models learn from data and improve over time. 

Understanding Large Language Models (LLMs) is essential for finance leaders, as they will likely either implement or use AI-powered tools in the near future. While deep AI engineering knowledge isn’t required, finance professionals should:

• Know the differences between open-source and closed-source models.

• Understand how LLMs can be customized (uptrained) on company or industry-specific data.

• Assess different LLM capabilities for various financial tasks.

• Evaluate the trade-offs between cloud-based and on-premises AI hosting.

LLMs are AI models designed to understand and generate text. They work by predicting the next word in a sequence based on extensive training data. Not all LLMs are the same—some excel at complex reasoning (e.g., legal and compliance tasks), while others are optimized for speed, cost, or specific industry applications.

Some LLMs are trained on specific data (e.g., health records, images, credit decisions), while others are more general-purpose and can handle a wide variety of tasks.

Below, we will examine different LLM categorizations and discuss how to choose between them when building AI automation or purchasing AI-powered solutions.

Open-Source vs. Closed-Source LLMs:

• Closed-Source Models: Proprietary AI models developed by companies like OpenAI, Google, and Anthropic.

• Open-Source Models: Publicly available AI models that companies can modify, customize, and deploy independently.

When to Choose Closed-Source Models

• Quick deployment for plug-and-play AI assistants.

• Minimal IT resources required to manage AI infrastructure.

• High accuracy needed for general business tasks (e.g., chatbots, text summarization).

When to Choose Open-Source Models

• Data security is critical – Models can be hosted on-premises.

• Tailored AI – Models can be customized for company-specific workflows and policies.

• Cost efficiency – Higher upfront investment but lower ongoing expenses.

Fine-Tuning Open-Source Models on Company Data

One major advantage of open-source LLMs is the ability to fine-tune them on your company’s own data.

What is Fine-Tuning?

Fine-tuning enhances an AI model’s performance by training it on company or industry-specific data, making it more accurate for specialized tasks.

📌 Example: AI-Powered Invoice Processing
A finance team fine-tunes an LLM on historical invoice data to classify transactions more accurately, flag unusual spending, and automate expense reconciliation.

Cloud vs. On-Premises AI Deployment

LLMs can be hosted in multiple ways depending on business needs, security requirements, and available infrastructure. Cloud-based hosting allows for easy scalability and accessibility, leveraging third-party providers like OpenAI, Google, or AWS. On-premises hosting gives companies full control over data security, compliance, and model customization but requires significant IT resources.

When to Use Cloud-Based AI

• No need to manage AI infrastructure – The cloud provider handles updates and maintenance.

• Scalable AI with lower initial costs – Easily scale up AI usage without significant upfront investment.

• Best for closed-source models – These require cloud access for optimal performance.

When to Use On-Premises AI

• Sensitive financial or client data must remain in-house – Ensures maximum data security and compliance.

• High AI usage makes cloud API costs unsustainable – Reduces long-term operational expenses by avoiding recurring API fees.

• Full control over an internally customized AI model – Allows tailored optimization and adherence to company-specific policies.

📌 Emerging Trend: As AI models become smaller and computing power more efficient, it is now possible to run a locally hosted LLM on a personal computer. This makes on-premises AI accessible even to smaller companies.

Key Takeaways for Finance Leaders

• AI model selection matters – Different models offer varying levels of accuracy, security, and cost efficiency.

• Customization is key – Open-source models can be tailored to business needs, while closed-source models offer plug-and-play convenience.

• Hosting considerations – Cloud-based models are easier to scale, while on-premises deployment provides maximum security.

Real-Life Scenarios: How I Choose an LLM for AI Implementation Projects

When selecting an LLM for AI implementation, I prioritize flexibility, cost efficiency, and finance-specific capabilities. Over time, I’ve leaned toward open-source models due to their customization potential, data security, and long-term cost savings.

Why I Prefer Open-Source Models

While closed-source LLMs offer strong performance, they lock companies into pay-per-use pricing and lack transparency. Open-source models allow:

✅ Full control over data privacy—no third-party servers involved.
✅ Fine-tuning on finance-specific data for greater accuracy.
✅ Long-term cost reduction, avoiding API fees.

Platforms like Meta’s LLaMA and Hugging Face provide pre-trained models optimized for financial workflows, making open-source a better long-term investment for many projects.

Finding Finance-Specific LLMs

Not all LLMs are built for finance. I focus on pre-trained models fine-tuned on financial data, including:

• FinBERT – A pre-trained open-source NLP model built by further training the BERT language model on financial texts. It's specifically designed for financial sentiment analysis.

• FinLLaMA – Based on the Llama 27B foundational model, FinLlama is fine-tuned for financial sentiment classification.

Comparing Models for Intelligence, Speed, and Cost

To refine my choice, I use tools like Artificial Analysis to compare models based on:

✅ Intelligence – How well the model handles complex finance tasks.
✅ Speed – Can it process large reports quickly?
✅ Cost – Does it make sense to pay for API usage, or should we self-host an open-source model?

Infrastructure & Cost Considerations

Beyond model selection, I evaluate:

💰 Hosting – Cloud-based AI (closed-source) offers quick deployment, while on-premises AI (open-source) ensures data control and compliance.
💰 Total Cost of Ownership – Open-source models have higher upfront costs but lower ongoing expenses, making them ideal for high-usage cases.

Choosing the right LLM for your business is a complex decision—one that often requires consultants and heavy involvement from your IT team. AI implementation isn’t just about selecting a model; it involves infrastructure setup, integration with existing systems, compliance checks, and cost management.

As a CFO, you don’t need to be an AI expert, but understanding the key differences between open-source and closed-source models will allow you to:

✅ Ask the right questions when working with consultants and AI vendors.
✅ Ensure AI investments align with security, compliance, and business objectives.
✅ Manage implementation budgets effectively by weighing long-term costs vs. short-term convenience.

Closing Thoughts

Choosing the right AI model is just the starting point—the next step is making it work for your specific needs.

In next week’s issue, we’ll cover fine-tuning AI models for finance applications.

Fine-tuning allows a general AI model to adapt to your company’s data and workflows, improving accuracy and relevance. If you’re considering AI for finance, understanding this process will help you get the most value from your investment.

See you next week! 🚀

We Want Your Feedback!

This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!

Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation! 

Did you find this newsletter helpful? Forward it to a colleague who might benefit!

Until next Tuesday, keep balancing!

Anna Tiomina 
AI-Powered CFO