Technology has become a basic input that drives everything around us today. From computers, multiple activities ranging from research to solving increasingly complex problems with robots can be accomplished. It's logical that finance is adapting to this new context; thus, in recent years, governments, companies, and programmers have developed technological projects, leveraging these new resources and the vast opportunities that exist in this field.

Financial technology or "Fintech," an acronym for "Financial Technology," may seem like a relatively distant and highly sophisticated term, often associated directly with specific applications such as cryptocurrencies like bitcoin or the well-known yet poorly understood "Blockchain." While it's true that these innovations have been extremely important, the reality is that the goal of this field is to improve and automate financial services through the use of technology. Thus, commonplace devices such as ATMs or point-of-sale terminals for card payments are part of this concept.

Initially, commercial banking played a fundamental role in technological developments. Since then, for decades, we've seen waves of new fintech products with different focuses covering a wide range of areas such as business management, digital banking, payments and remittances, financing, among many others. Recently, a user-oriented vision has been adopted for these developments. Thus, the financial ecosystem has allowed the emergence of startups competing to gain a place in the preference of the general population and overthrow the almost hegemonic power that banks have had until now in the financial system.

E-commerce is one of the branches that has benefited the most from these implementations. Currently, due to the pandemic, it has achieved even greater growth than it was already experiencing, consolidating itself in the market and playing an enormously important role in the current scenario. Alongside its rise, various risks such as fraud, information theft, and identity theft are becoming increasingly relevant. Therefore, companies like Mercado Pago and PayPal are gaining strength as they offer alternatives to address these issues.

Fraud, as mentioned earlier, is one of the vital problems being constantly addressed. As new technological developments are made, innovative ways to carry out these negative practices are also being discovered. Fintech offers multiple tools to combat it, including more secure payment methods with more controls, implementation of dynamic indicators or "KPIs," and programs that detect cases with the highest probability of occurrence practically at the time of purchase. But how are these programs being created and becoming increasingly accessible to companies?

Mostly, it's thanks to "machine learning," which has provided the infrastructure on which the necessary code can be built to develop these products. Making computers "learn" may seem highly abstract, but conceptually it isn't. In very general terms, the process is as follows: it starts with a statistical model created by the programmer that gives the probability of something happening or not, for example, whether an operation is fraudulent or not. Historical information must be used to build the model that is closest to reality, leaving part of that information for testing. This way, it's tested if the system is accurate, and if it's not optimal, the weight of the variables initially assigned is adjusted and modified to a more suitable one. As more information is included, this process is repeated, and the original model is constantly transformed, causing the computer to "learn" and become more effective over time.

As can be seen, the concept of machine learning isn't so complex; however, certain points must be taken into account when working on this topic. The first is the quality and quantity of information. The more data there is, the more precise the system tends to be, as it will have a larger statistical strength. Therefore, constructing a good, well-designed database will allow obtaining truthful, timely, and quick information to work with. The second is a solid knowledge of statistics, as there are various concepts that must be understood and managed to create a suitable model. Although the computer improves constantly, we show it how to go through that learning process, and if not properly instructed, it will give results that don't correspond to what we are looking for.

Although only the example of fraud detection has been presented, the implementation of machine learning has other applications in the field of finance, usually aimed at solving problems related to result prediction. For example, it could answer questions such as: Will the sale being made have a return? Will the job applicant with certain characteristics work with a certain efficiency? Will any analyzed stock rise by at least 3% the next day? Solving these types of questions is used to take actions and make decisions, creating programs with more sophisticated uses that undoubtedly facilitate life for individuals and organizations.

There are multiple programming languages that allow this process. Python is one of the most popular ones in the last decade because it's open-source, and the computing community has already created packages that include codes to perform certain functions. Therefore, everything doesn't need to be programmed from scratch, but what they have developed can be taken and used on the functionality we want to give to our program. This is making it easier to create these types of applications every day. There are already multiple companies exclusively dedicated to this, competition is increasing, therefore costs are decreasing, and they are becoming much more accessible for medium and even some small businesses.

Now, another important point on this topic is that simply hiring a person who knows programming isn't enough to develop these systems. One of the most important things to consider is knowledge about how the business and the industry itself work. The logic on which it is developed must be consistent with qualitative issues. We mustn't forget that, in the end, a model is nothing more than a representation of reality; the data is sometimes misleading and doesn't represent what is really happening. Therefore, we must never lose focus on what we are trying to symbolize. Sometimes, the best model isn't necessarily the one with the lowest error, as one might intuit at first, but the one that also respects the logic of the real functioning of what we are working on.

It is essential to understand that the information generated by machine learning programs is based purely on probabilities; there will always be exceptions to the rules, and it is precisely there where modifications will be made to predict more accurately. As decision-makers, this can then lead us to erroneous actions, such as considering a legitimate sale as fraud and thus losing business, or not hiring someone based on forecasted performance when they haven't even been given the opportunity to demonstrate their capacity. With the above, I want to emphasize that when we use these systems, we must understand how they work, at least the logic they use, and be aware of the associated risks, using them as just another tool and not as a panacea that tells us what to do and functions as a black box.

In summary, machine learning is a tool that allows the creation of models that adjust as new data is introduced. It has a large number of applications, such as fraud detection, aimed at decision-making. Every day, creating these systems is more accessible. In fact, only a computer with installed software is required to make basic software of this style. The professional who creates a machine learning system, whether an individual or a company, must have solid knowledge in statistics, model development, programming, and especially knowledge of the company and industry. The decision-maker, on the other hand, must understand how this program works, at least in its logic, as it is a predictive system based on probabilities, so it will not give 100% correct answers and could lead us to wrong resolutions.