The global MLOps (Machine Learning Operations) market, solutions that help data scientists simplify and optimize machine learning deployment processes, will grow at an average annual rate of nearly 45% until 2030. The projection was made by the research company Valuates Reports, which estimates a jump in valuation from US$ 186.4 million, reached in 2023, to US$ 3.6 billion. One of the main reasons for the heating of this market may be the reduction in the time frame for developing predictive models. The assessment is by Carlos Relvas, Chief Data Scientist at Datarisk, a company specialized in the use of artificial intelligence to generate value in the concept.decision as a service”.
According to him, to develop similar systems with traditional methods, organizations take an average of two to three weeks, depending on the complexity of the sector.
“On the other hand, when using MLOps, the data scientist can automate the entire creation process. First, he does all the model training using automatic machine learning that tests algorithms to see which one works best. At this point, the scientist can also, if he wants, upload his own code and save all the documents and codes, thus ensuring the protection of the documentation of all the databases. The success of MLOps is due to the fact that it eliminates all these steps, with the model creator himself being responsible and having everything he needs to go from the beginning to the end of the project,” he says.
In 2024, Datarisk launched a market-ready MLOps solution focused on serving leading companies in activities such as credit granting, fraud risk, propensity for job change, agricultural productivity, among others. Only during the first half of this year, the tool was used to perform over 10 million queries, and among the benefits gained by users of this technology, one of the biggest highlights was precisely the reduction in time. With the startup's MLOps, the average lead time of three weeks has been reduced to a matter of hours.
Carlos Relvas also explains that after this first training is built, a second stage within Datarisk's MLOps platform begins, which is the part where the scientist can automatically, by himself, create an API for the model to be used in external environments. The third stage, according to him, is solution management. At this stage, the goal is to ensure that this model, which has been developed, trained, and is being used, continues to perform well over time. "The tool can monitor both the usage of your applications and the functioning of the APIs to ensure not only that everything is operating as programmed but also to allow the assessment of the model's quality. The solution enables verification, for example, if any variable has changed over time, and issues alerts to the end user if the model is losing performance," he states.
The market's receptiveness and the prospects that Datarisk has made allow the company to project growth of more than five times the volume of use of this solution by the end of 2025.
The Co-founder and CEO of Datarisk, Jhonata Emerick, explains that by becoming a pioneer in offering solutions in the MLOps concept in Brazil, the startup is putting into practice the strategy of maturing and refining its main business theses. "We understand the market's needs more deeply, and now we are prepared to offer solutions capable of transforming the reality of data science in the country in an absolutely relevant way," he says.
According to Emerick, in the specific case of developing predictive models, MLOps solutions emerge as a response to slow internal processes designed for a time when companies did not need to manage a data area with the agility that is currently required.
“IT queue systems are generally adopted in which the data science area finishes creating a model and passes it on to the engineering area to create an API. The engineering area, in turn, will take a significant amount of time to do its part, when it will then pass the project on to the credit engine team, for example, so that it can finally implement this API, which will lead to other deadlines. The result is that, when the model is implemented, the situation is already different. This is why the MLOps solution becomes so effective in terms of optimization,” he concludes.
