Alexandre Gleria, a Brazilian corporate attorney and CodeX affiliate, presented his Corporate Behavior Coding (CBC) project. The system uses behavioral data, primarily litigation records, board compensation figures, and qualified stakeholder perceptions, to rate and classify companies in ways that traditional financial statements cannot. The core insight is that financial data alone is biased and manipulation-prone, while behavioral signals like litigation patterns relative to industry peers and the ratio of executive pay to earnings reveal the human decision-making behind a business, which ultimately predicts its sustainability.

Watch video of 2.12.2026 CodeX Group Meeting with the CBC Project

Transcript

Roland Vogl:

We will now be turning it over to Alexandre Gleria, who will be updating us on his work on the CBC project, which is a project he’s been working on as a CodeX affiliate. He’s of course also a practicing corporate attorney in Brazil, and we’re excited to hear where the project stands at this point.

Alexandre Gleria:

Thank you very much, Roland, and thank you for the opportunity — for being so kind to us here from so far. I’ll try to speed up here because I have a lot of stuff to talk about today. Basically, what is the project about? We’re creating a method and a system that could indicate the human behavior behind business decisions. These could classify companies, sectors, and also jurisdictions. We’re finding that this could have a very broad application in terms of commercial and industrial perspective. 

Based on several research papers that we are writing, this could also be a game changer in several fields, especially because these sort of map human behavior. We could avoid the bias of the data that is provided by the companies, given that all the data that companies provide comes from their own sources. Of course, we have big four companies auditing these numbers, we have accounting looking into these numbers, but to this day there’s a lot of room for fraud, for loopholes, and so on.

These ratings and these classifications are also showing very promising results in terms of having anti-fraud properties, especially because we are observing that the ratings and the rankings of the company only evolve if the human behavior behind the business also evolves. This is a very cool feature that we are discovering in our research. I also have a lot of tests that I can show you of how this could be a game changer in several areas such as credit analysis, stock market analysis, and so on.

Basically, I have here just a cartoon of how this project started. In 2018 — much before the AI generation and products that we have today — we started a project in our firm to try to project, in a mathematical perspective, ways to measure financial impacts and probability of loss in contentious litigation. We also observed that we saw patterns in some of our clients — clients that were not aligned with our culture, clients that were very problematic. They also had a very peculiar pattern of litigation, a particular profile of litigation that this kind of company carried.

Based on both of these observations, from 2018 until today, we hired big four companies, interviewed C-level executives, hired mathematicians, data scientists, and economists. We have some state judges as partners of our project. One of these judges is helping a lot on the inside, and he’s also very good with equations and so on. We have economists on our team who are also here listening to us.

The first starting points of the Corporate Behavior Coding — the trigger points of the project, as we know today — is that we found that the litigation data was very rich in telling us all the stuff, all the information, that is not very clear in financial information. If a company has a bad product, this ends up in litigation. If a company is saying that it treats employees very nicely but behind the scenes does a lot of wrongdoing, this could also end up in courts and so on. If companies have corporate struggles, issues, and fights among partners — all kinds of stuff — in the end, we are confident in the judiciary branch. Basically, based on these, we classified this as a very nice source for understanding behavior, for understanding corporate behavior, or the behavior that is behind business decisions, because all business decisions are made — even though we are in the generation of machines and robots — by humans in the end.

This is a subsequent development of this project. We started to research other sources that could also indicate behavior, such as board compensation numbers and figures, in a sense that we saw that business perpetuity is very strongly connected with how the partners and shareholders of this business are being remunerated and paid over time. We also looked at the perception of qualified stakeholders.

Currently, we have a lot of data being collected from the markets — from, for example, X, or the image, or the press. But what matters here, we saw, is that we need to have quality in terms of data. So we started to realize that, for example, consumers, employees, or previous employees of the company could also provide a lot of good insights about the future, about the perpetuity of the business, and other cool features of the company.

We have also seen, as you can take a look at in the diagrams, that these traces of human behavior are very dense in the litigation and board compensation documents and in the perception of qualified stakeholders. While we have a lot of biases and a lot of limited scope in financial statements of a company, you have a lot of conflicts of interests. Of course the company wants to show the best numbers possible. It’s very unlikely that you could see a raw version of the corporate culture behind those numbers.

Based on that, we built the classification — the rationale of what kind of data we could consider as behavioral data. I will not explain this in detail because we don’t have time. The rationale of the behavioral data is essentially what I just mentioned: avoiding bias, quality of the information, the depth of this information. Most of this data is coming from external sources, and the ratings evolve if the behavior of the business people also evolves.

Basically, we are in the final stage of the paper. As you know, we are also researching a lot of mind-blowing facts from the numbers and figures that we extracted from this study. We are finding that there could exist a lot of symmetries that we also find in nature, within these numbers and these classifications. That is a very cool feature as well of CBC, but we are digging on that and also researching with other specialists.

One thing that is also cool about this project is — why are the results so good? We started to dig a lot on that. Basically, we find that the combination of two sorts of different data provides a lot of images that the numbers alone are not telling us. In an analogy that we make here: if I assume a computer system — as you know, a computer system runs on a binary data scheme — and I assume in this scheme that I am just looking at financial data, that is the zero. I’m not getting a lot of depth in the view of what I’m analyzing. When we add the behavioral data, then we can see images and assumptions that could help decision-makers understand a lot of what is going on in the business. As you can see here, zero doesn’t mean nothing, but when I add the one in the binary code, I can see vision, if you look from outside of it.

We are also seeing that these methods have a lot of advantages over other very recent and promising research. I will not talk about that in depth because of time. Let me just show some results in two or three minutes, and then we can open for questions.

Basically, we ran a lot of tests on 180 publicly traded companies in Brazil. We created some ratings from each of them. Here are some very good-rated companies from big four companies and credit agencies. As you can see, these companies all presented a very bad rating according to our methodology, like ten years before a Chapter 11 event, for example.

Roland Vogl: 

Can you give an example of what behaviors you measure? Like, do you give a rating to companies that display certain behaviors? You mentioned fraudulent behavior, but what other specific behaviors are you tracking? Do you find they have an impact and a dependent variable that goes to the performance of the company? You’re looking at measuring human behavior in a company and then seeing if it has an impact on stock performance, litigation, bankruptcy, or certain outcomes. What exactly are you measuring?

Alexandre Gleria:

We are measuring, for example, the amounts — combining pairs of information and measuring the asymmetry of this information. As an example, if I have a next level of litigation on a certain matter and the company presents very good earnings, this could be an indication that this business, in a small fraction of its DNA, is sustainable. Or, for example, if this company’s litigation level is very acceptable due to market standards — but measuring the occurrences of litigations of a particular company and comparing it to how common litigation is in that particular industry. If one company has more litigation than another, that really is the variable you’re measuring: a high amount of litigations, and then seeing if that has some predictive power over the company’s financial performance.

It does, but isolating litigation data alone is worthless. You need to combine it, and you see the asymmetry between litigation, for example, and other data and the financial robustness of the business. Or, for example, if you see the directors’ payment — the salary of the board of directors — and compare it with the earnings of the company, you can also see levels of whether this business is sustainable, or whether there’s a pattern that indicates shareholders were cashing out the company, and in two or three years this company will go through Chapter 11.

Roland Vogl: 

Got it. Could you share the main conclusions and takeaways?

Alexandre Gleria:

The main conclusions are that our results, as I said, were very robust. We created stock portfolios just based on this methodology, and the returns — not only in Brazil but also in the U.S. — are equally robust. In 2022, we also ran a real test for a hedge fund in Brazil. We analyzed a 20-stock portfolio to see what was going on with it. We realized that one particular stock had some issues when mapping these images based on these contrasts. We identified the problem, told the fund managers what was going on, and since then the stock is down almost 90%. We have a lot of tests here that will be in our paper.

Just to conclude — if possible in two minutes — I want to show you the platform that we are building based on what we are creating.

So basically, here we can compare a lot of companies, not only in Brazil but also outside Brazil. The data is not real because, as you know, we filed the patents a few days ago, but here you can compare a lot of information on litigation, the index of the company, the litigation asymmetries, and the litigation map comparison between two companies. There’s a lot of things that we can measure here regarding the methodology, but this is just a flavor. There’s a lot of theory behind it.

Roland Vogl:

How can people learn more? Is the paper available? Can you share a link?

Alexandre Gleria:

The paper will probably be available within two months. We are finishing it.

Roland Vogl:

Awesome. Thank you so much, Alexandre. I think it’s a cool idea to be able to understand legal indicators and signals — litigation, but also other things that are legally relevant — and then map that to a company’s performance. That’s what you’re trying to do in essence. So what’s your long-term goal? This started as an interest from practicing corporate law with large public companies in Brazil. You said at the beginning you saw some clients with different corporate cultures and behaviors than others. You could almost anticipate that one company would do better than another, and there is a connection between the culture and the way people conduct themselves in a business, and the outcome. That was your thesis, and that was the research project. But there’s also a patent now and a company that will presumably try to make those insights available, because it is potentially tradable information. Can you talk a little bit about how you see the future unfolding for this project?

Alexandre Gleria:

Well, basically, we realized that a company with a specific company profile and culture — they don’t leak targets, they avoid litigation, they have less litigation outstanding in comparison with peers, the quality of their products or services or intangibles are superior in a way that they don’t have a lot of claims in a general sense — not only in litigation but also in behavioral data, such as opinions issued by qualified stakeholders like a consumer, an employee, or an employer.

Based on that, we saw that, for example, if we compare and translate this into real numbers — the amount of litigation in terms of volume, the amount of litigation in U.S. dollars — and compare it with the economic capacity of the company, this gives you an idea of how sustainable the company is. The idea here is to scale that, because there are a lot of people working on unstructured data arrangements and platforms in order to use, for example, transformers to process a lot of material from companies that have a lot of information. But this approach, in our opinion, is not the best one. First, because it consumes a lot of compute resources, and the quality of the raw material and inputs used is not good either.

The idea is that with the right inputs — as in the behavioral data that I just gave examples of — we can scale this to provide, globally, more reliable corporate ratings and corporate classifications in all jurisdictions, regardless of whether that jurisdiction is common law, civil law, or whatever. Especially because we are running tests not only in Brazil but also in the U.S., the U.K., and other jurisdictions. The key point is that once this data is structured, we can scale very fast. Of course, this permits analysis of public companies more easily, but it could also be used by private health companies.

Roland Vogl:

Do you have a name for the paper already?

Alexandre Gleria:

The name — we are thinking about it. CBC was a very corporate name that we chose. The original informal name, the nickname of the project in the first months, was the Corporate Genomics Project. That’s why we named it CBC — because we built a whole theory on that and are just consolidating it to give an official name to the paper. In the paper, we have a lot of mathematicians involved because — I’m an attorney, even though I’m a specialist in tax law and corporate law — we need to have qualified people in this paper. We have a very strong background on math and financial markets in order to prove our theory.

To conclude on the tests: when we compare these ratings with rating agencies or any other kind of data available in financial markets, for some specifics these prove to be much more reliable than anything else. We also prepared slides showing that in the last 50 years we have a lot of innovation, for example in cancer diagnosis and medicines in general terms. Yet when you look through the corporate outlook over those fifty years, we still have fraud scandals, accounting mistakes, and so on. Basically, the idea here is to provide not only a new product and idea to the market, but also a culture of corporate diagnostics.

Roland Vogl:

That’s an exciting future. I think you could apply it to the corporate world, the crypto world, and other areas too. Thank you again, Alexandre. Thanks for joining us.