This is how brands use AI to decipher our emotions

One of the most effective methodologies to decipher all this data comes with sentiment analysis. An artificial intelligence technique based on NLP (natural language processing), ‘machine learning’ techniques (for supervised classification tasks), ‘deep learning’ (which provide this additional understanding of nuances and contexts in the analysis) and ‘text-to-speech’ (to analyze real-time conversations and audios through transcription) that allow you to identify, extract and analyze data published in emails, comments on social networks, forums, reviews, conversations, etc. And they are used by companies to nourish their commercial strategies.

It is not just about identifying patterns or signs in the text that indicate the presence of emotions, attitudes and opinions. It is about “knowing the customer, their expectations, needs and knowing how they feel during an interaction at any time of the day.” ‘customer journey’ and on any channel,” says Laia Mercadal, director of Consulting and Digital Transformation at E-voluciona by Intelcia.

“Until very recently, NLP algorithms were trained to recognize certain words in texts and compute them to give a simple ‘positive or negative’ rating, and this was already a considerable achievement,” notes Bruno Gerlic, Chief Revenue Officer at PredictLand AI. . Little by little, they began to improve this binary classification until they were now able to combine feelings with other types of valuable information in the same system, sometimes called ‘text mining’ or ‘text mining’. In this sense, “they now provide much more valuable information, as it improves decision-making in companies with more granularity and less latency.”

In simple terms, when a text is fed into a machine learning algorithm, it returns a score between 0 and 1 indicating how positive the text is. Through lexical resources and natural language processing, the emotional connotations of words can be evaluated and a text classified as positive, negative or neutral. In addition to lexical resources, there are numerous machine learning and deep learning models that are specifically trained. Satisfied, dissatisfied, satisfied, irritated, upset, grateful, surprised… each feeling leaves a mark on the way we express ourselves, and the system is capable of deciphering this emotional code.

Thanks to the advances of generative AI, whose models, called LLMs (such as ChatGPT 4), have been trained primarily on the Internet, the algorithms detect these fine nuances of communication in feelings, natively understand all types of languages ​​and styles, and They know how to interpret the context. «If the system detects, for example, repeated customer complaints about a functionality of a product and depending on the intensity of the associated feelings, it will label it as ‘priority’ or ‘low criticality’ and generate a report addressed to the product manager. », explains Gerlic.

Improvement path

But how can this technique really help create a better customer experience? From E-voluciona they evaluate the quality of the service with parameters and metrics to know how you feel and the degree of satisfaction through analyzers such as FCR (First Call Resolution) to know if the query has been resolved in the call or through the NPS (Net Promoter Score) that indicate the brand’s level of recommendation. “All this allows us to know the real customer satisfaction rate and continually improve the quality of service,” says Mario García Láinez, director of Soluciones IA. «In the contact center sector, the use of these technological tools and skills serves to improve the quality of service and reduce callbacks. The objective is to ensure that the Average Operating Time, that is, the time for managing and resolving a query, is as low as possible with high productivity and customer satisfaction.

Combined with other ‘text mining’ techniques, sentiment analysis also contributes to improving multiple experiences, PredictLand AI recognizes. From the redesign or personalization of products and promotional messages for each type of customer, proposing communication according to the emotional state, to anticipating cases of abandonment risk to activate personalized retention strategies, especially after analyzing support queries and complaints.

However, there are many aspects of our human interactions and behaviors that cannot be quantified by algorithms. In this context, neuromarketing emerges as a complementary tool to reveal those unconscious processes that influence our decisions and behaviors.

Salima Sánchez, psychologist specializing in Neuromarketing and Consumer Behavior, points out how psychology studies and explains very well the psychological processes and biases to which we are subjected, but “through neuromarketing lor what we achieve is to go a little deeper and say: okay, I’m not going to ask the person because the person can unconsciously lie to me, I’m going to ask their brain, which I know is not going to do it. With techniques such as electroencephalography, which records the electrical activity of the brain, allowing the identification of patterns of attention, emotion and memory, eye tracking, which monitors eye movements to understand where consumers focus their attention and for how long. or with physiological responses such as heart rate, skin conductance and galvanic skin response, the user experience can be evaluated in a very precise way and with data.

Analyzers

Depending on the type of data available, its quality and the objectives pursued, companies must carefully choose between multiple formulas, each with its pros and cons. According to Gerlic, “general platforms such as Microsoft, Google or Amazon offer sentiment analysis tools within their cloud services. Then there are multiple specialized softwares, which usually provide another layer of intelligence to a specific process. For example, social media analysis software, online brand reputation monitoring software, additional modules in customer management platforms (CRM), project management platforms, etc. But there are also open source models for IT departments to adjust and integrate to a specific need.

More specifically, in the contact center sector, speech analytics makes it possible to analyze more than 25 emotions at different moments of the conversation in both voice and text, converting unstructured data into consumable information. and structured for analysis. Mario García explains that “generative AI is applied to all this information, taking advantage of its ability to understand, to detect the root causes and be able to improve the service.”