4 August, 2021 Cognitive optimization to take EVO Assistant to the next level Several years ago, EVO Banco decided to offer a virtual assistant to its customers. But this company is much more than a product. It is about relationships and people. For this reason, they concluded to look for a partner to offer better assistance to its users through its conversational agent. EVO Banco, which is currently part of the Bankinter group, was born as a commercial brand in 2012 to become a quality alternative to traditional banking. The bank had the idea of questioning the established and offering intelligent answers to the needs of its customers. Under the Smart Banking model, EVO Banco has developed a unique proposal in its sector that leads it as one of the most valued banking references by consumers. This banking institution is well-known for its ambitious and innovative concept, as well as for its maturity in technologies based on Artificial Intelligence and Big Data. Several years ago, they decided to offer a virtual assistant to their customers. But this company is much more than a product. It is about relationships and people. For this reason, they concluded to look for a partner to offer better assistance to its users through its conversational agent. Challenge More than two years after the launch of its virtual assistant on its website, app and call center, EVO was looking to keep enhancing its conversational system. In order to do so, they were seeking for a partner that had accumulated experience in virtual agents and that would support their team of experts in the improvement process. In this case, Emergya's proposal and the scope associated with the different points was focused on responding to two main objectives: Improve the quality of the wizard and its performance ratios. One of the challenges was: helping the EVO Banco team to move towards the evolution and roadmap of the assistant. It also sought to guarantee, as autonomously as possible, the quality of the responses for all channels - current and future - in which the virtual assistant is present. Accelerate the industrialization of the assistant. On the other hand, another challenge was for EVO Banco to speed up the roadmap and Time to Market of the assistant in channels, integrations and functionalities where Emergya had a previous successful experience and to eliminate research, development and start-up times. In addition to these two main objectives, there was an added challenge: to achieve the purpose of the project successfully through the joint work and communication and common professional interest of the EVO Banco and Emergya team.+ Solution Emergya proposed an action plan that was defined in the following steps: Review and organization of the different categories and overview of the conversational agent, to start by identifying where the most value could be added. Cognitive optimization of the virtual agent and covering an in-depth analysis of its conversational design. This was intended to achieve the healing and reduction of intents, training sentences and refactoring of entities. Measuring the impact of improvements by comparing the initial state of the conversational assistant in relation to the measures implemented in the process of its evolution. With all these steps, the aim was to achieve curation and optimization of the virtual assistant to take it to the next level: improving its interaction with customers and its conversion rate. Results After the changes, an analysis process was carried out on the agent in web and app, to measure the improvement achieved in the level of assistant's success. In order to do this, the hit rates were analysed for a subset of 450 test sentences, consisting of about 3 training sentences per attempt covering 33.5% of first level attempts and different categories. The analysis' result showed an improvement of 29% over previous hit rates for the subset examined. On the other hand, a similar measurement process was carried out on the call center agent, using a subset of 136 training sentences representing first level attempts and different categories. In this case, the improvement achieved for this other subset in the level of accuracy reached 10%. These are the results obtained from the first optimizations implemented. However, the work with EVO Assistant does not stop here, the objective is to continue to make further improvements in both current and future operations that EVO Banco wants to implement. Emergya has reinforced our teams with expert knowledge and experience in the field of cognitive optimization of virtual assistants which is helping us to further evolve our conversational agent. Pedro Tomé, Head of Disruptive Innovation, Big Data & Advanced Analytics at EVO Banco