How to Find Undervalued Stocks (Undervalued.ai)

After a decade of investing, I've consistently faced two challenges: finding time for research and cutting through the noise to get the most useful information. This led me to build a stock analysis system that combines quantitative filtering with artificial intelligence to identify value opportunities in the stock market. In short, we use AI to identify great companies trading at reasonable prices compared to their intrinsic value.

Finding undervalued stocks requires a comprehensive approach that considers multiple business dimensions. Below is the full break-down of the methodology I use to power undervalued.ai

Initial Universe Filtering

Narrowing Down a Massive Market

Analyzing every publicly traded company in the world is impossible. In particular, there's the complexity of different currencies, the different reporting formats, and, more importantly, the questionable reliability of data for some markets. As such, I decided to focus on the NYSE and NASDAQ for now. Those two stock exchanges host most major global companies and provide standardized, reliable data.

The methodology to find undervalued stocks starts with two filters. The first one:

• Excludes sectors requiring specialized analytical frameworks (Financial Services, Real Estate, Utilities, Healthcare)
• Applies minimum thresholds for market capitalization and trading volume

Comment: This initial filter is similar to Warren Buffett's "Circle of Competence" concept. In short, it's better to focus on subjects we understand rather than spread ourselves thin across unfamiliar topics.

Then, we apply a second, sector-specific, filter. For example:

• Growth sectors (Technology, Communication Services) are evaluated with different profitability expectations than mature industries
• Capital-intensive sectors (Energy, Basic Materials, Industrials) are assessed with appropriate interest coverage ratios

And this leaves us with a smaller universe that is more qualitative. We can then proceed with the in-depth AI-analysis.

Peer Grouping and Assessment

Before proceeding with our analysis pipeline, we group the selected stocks by:

• Market cap size bracket (Small, Mid, Large)
• Industry classification

This is very important because we need to compare apples with apples. Even within the same sector, comparing a $2 billion software company to a $200 billion tech giant would distort the analysis.

Then, within each bracket, we rank companies on the two most important dimensions when looking at fundamentals:

1. Their Growth. For example, their revenue growth, operating cash flow growth, EBIT growth, etc.
2. Their Profitability. For example, their ROE, net profit margin, operating margin, etc.

The key concept is that nothing is "absolutely" good or bad. A 15% profit margin might be exceptional in retail but mediocre in software – context matters.

Companies scoring in the bottom quartile across multiple metrics relative to their peer group are eliminated. This helps us eliminate 'value traps' - companies that appear cheap but are, in reality, declining businesses with bad fundamentals.

Multi-Dimensional Analysis

Once a stock passes our filtering methodology, it enters our AI Analysis Pipeline. In short, it is like having a team of world-class financial analysts looking at several dimensions and providing an assessment under the supervision of a manager.

Here are the dimensions as of the 26th of February 2025

1. Financial Statement Analysis

Three specialized AI agents work in parallel to analzye the financial statements over several years, with a focus on the latest quarters. Each agent looks at a different part of the financial statements:

• One agent for the Balance Sheet structure and strength
• One agent for the Income Statement quality and growth patterns
• One agent for Cash Flow sustainability and capital allocation efficiency

2. Fundamental Synthesis

Then, another dedicated AI agent integrates the three findings and look deeper into:

• Real earnings power (removing accounting distortions)
• Cash generation capability
• Quality of growth (organic vs. inorganic)
• Financial flexibility and operating leverage

We stress fundamentals because numbers tell a story, AI can hallucinate if provided with too much data (hence the multi-layer), and unfortunately, many companies use accounting techniques to inflate reported earnings while their actual business performance deteriorates.

3. Peer Comparison Analysis

In parallel, an agent compares each stock with its direct industry competitors across dimensions and look into:

• Relative valuation using sector-appropriate multiples
• Competitive positioning assessment
• Profitability chain analysis

Definition: Profitability Chain Analysis examines how efficiently a company converts revenues into profits (gross margin → operating margin → net margin). For example, Apple's extraordinary gross margins highlight its pricing power. Costco's thin margins but high inventory turnover demonstrate its operational efficiency, etc.

4. Technical Analysis

Another AI agent looks at the technical component. In particular:

• Price momentum and reversal patterns
• Moving average relationships
• Volume patterns and confirmation signals

We integrate those technical signals because they help identify entry points.

5. Macro-Environmental Analysis

Then, another set of agents examine how the company fits within the broader economic landscape. Among other elements like the labor market, consumer sentiments etc, we look at:

• The interest rate sensitivity and inflation impact potential
• The global market trends and sector rotation patterns
• And the large flows to understand institutional movements

The objective here is to get a pulse of the market and how would the company likely perform in this global context.

6. Insider Trading Assessment

When multiple C-level executives or major shareholders start buying (or selling) their company's stock, it's definitely an interesting signal. Here again, another AI agent look at several points like:

• The buy/sell patterns and timing
• The role/position of insiders involved. For example, the CEO or a cluster (coordinated action) would be weighted more than a director.
• The size of transactions relative to holdings

Final assessment

As a final step, another AI agent takes as input all previous AI-generated analysis, gives context, and then gives a final assessment with recommendations. In particular, this agent provides:

• A clear investment thesis explaining why a stock is undervalued, priced correctly, or overvalued
• A one-liner summary of key findings
• And specific price levels that would trigger reevaluation

We can trust AI

In our system, we leverage the most advanced large language models from Anthropic and OpenAI, creating specialized AI agents that function like a team of world-class analysts:

• Each "analyst" agent specializes in a different aspect of stock evaluation
• A "manager" agent integrates and weighs these various perspectives
• The entire system functions without conflicts of interest or emotional biases

AI is a fantastic tool to ensure no conflict of interest. Indeed, when reading the report provided by an investment bank, there is always the question of loyalty. Investment banks must serve both the companies they rate (who pay for banking services) and the investors who rely on those ratings. And this can lead to human bias with expensive consequences. Most notably, during the 2008 subprime crisis, toxic mortgage-backed securities received the same pristine AAA ratings as U.S. Treasury bonds. Meanwhile, firms desperately tried to dump these worthless assets into the retail market.

Quoting Margin Call:

Sam Rogers: And you're selling something that you know has no value.
John Tuld: We are selling to willing buyers at the current fair market price.

As final note, while no system is perfect, this methodology has proven effective for me and is the engine behind the reports you can find on undervalued.ai