Annually 1,000,000 people die victims of suicide in the world, one person every 40 seconds! The Chilean Fundación Para La Confianza through its Línea Libre is the first containment barrier for this scourge.
After an intense campaign in various media, the number of calls they receive increased from 22 to 67 calls a day. Faced with this brutal increase in demand and with the same staff of psychologists, they approached the Deep Talk team so that we could adjust the models and that with our Machine Learning algorithms we could help them to classify and prioritize the calls.
After assembling a dataset of 3300 pairs of messages labeled by psychologists, we trained our Custom Detection product, a model that allowed them to detect “suicide intention” with 80% accuracy. After little more than a month of work by a multidisciplinary team that included three psychologists, the project was rigorously tested for 2 months before moving on to production and detection in real time. Last week, for example, there were 58 true cases that could be properly detected and prioritized, with the importance of accurate and timely detection in cases like this.
One of the main banks in LATAM, needed to unveil the internal structure of its digital service process, to know what its main service flows were and what its clients talked about most in conversations with bank agents. They came to Deep Talk and asked us if we had any tools to help them in that process. Of course we have! was our answer, we launched the Flow Visualizer and went through the entire chat conversation database. Thus, managers and decision makers supported the exploratory analysis of said conversational data with a graphical tool that allows visualizing the existing conversation flows, to understand, for example, the main components of the conversations and to be able to reach a finer segmentation of the reasons for such conversations in chats.
Then with all this data, the Bank created new bots and re trained all their bots with higer precision.
One of the main retailers in LATAM approached with a very precise requirement, they wanted to order all the internal queries that were made to the internal technical support help desk. They did not know what topics these consultations were about, nor did they know which of them were more important and which were not as high priority. In Deep Talk we launched our Topic Detection product that preprocessed removing noise and accommodating the data and thus grouped, categorized and ranked 4740 unstructured queries in Spanish and Portuguese, yielding results of great value for the people of the company.The entire dataset, no matter how many conversations I had, talked about 30 concrete and specific topics on which they could start working. 51% of their queries were concentrated on the first 5 topics and 33% of the queries on only the first 2 of these 30 topics. Based on this information, they redesigned the menus of the consultation system to respond to all these issues that they did not take into consideration, focusing of course on the 5 most important ones. The latter allowed the company to obtain results that quickly added value to their internal technical support consultation system.
Get value from your unstructured data
We believe conversational data it's an asset for companies