The complaint is an action guaranteed by consumer protection entities, and with the increasing digitization of society and the popularization of digital complaint platforms, there is a growing increase in customer complaints in these platforms. The escalation and the incorrectly handled complaints in the digital platforms impact the image of companies and generate an increasingly latent need to capture, process, analyze and generate knowledge using automations. With this scenario, this dissertation proposes the use of Text Mining and Machine Learning techniques in a complaints triage flow in order to minimize the probability of non-resolution of complaints from financial services customers. The proposed approach is evaluated through 3 different architectures of complexity, with the Naive-Bayes SVM that was used as a baseline model, the FastText that was used as the intermediate model and the DistilBERT that was used as a challenging technique. The use of only the narratives of the consumer's complaints presented little relevant result in the discrimination of the resoluteness, however when added to the narrative of the company's service, the models performed considerably well and can be applied in a service strategy. The results showed that the methods can generate value in decision-making support systems in the operational area of companies and can help them in serving their customers, raising the level of consumer satisfaction and reducing the risk of the company's image. The choice of a good prediction model depends on the performance strategy of the customer service, and the ability to sort the risk becomes preponderant in this choice.