Beginning in 2017, Yabble created a platform where organizations could easily analyze thousands of customer data points shared through surveys or customer feedback forms, and pull together clear, data-backed insights. In 2021, they added Yabble Query, a tool that allows users to ask Yabble a myriad of questions to better understand their own data, and uses AI-powered algorithms to provide relevant insights into the questions most important to users so they can spend less time analyzing data and more time driving the business. This year, they introduced Yabble Count, an AI tool that analyzes thousands of comments and other unstructured data sets, categorizing them by sentiment, and organizing data into themes and subthemes to understand the key topics and feedback resonating with customers.

While Yabble was seeing success with helping some of the biggest brands better understand their customers, they were getting bogged down in the manual labor required to develop actionable insights from their customers’ data. They would spend days or even weeks translating data sets—until they worked with OpenAI.

“Users were loving how easy it was to finally understand mountains of data and feedback forms and have that information presented in a digestible way but it took weeks in some cases to analyze data sets for some clients,” says Ben Roe, Head of Product at Yabble. “We knew that if we wanted to expand our existing offers, we needed artificial intelligence to do a lot of the heavy lifting so we could spend our time and creative energy elsewhere—OpenAI fit the bill perfectly.”

Using OpenAI’s GPT‑3 natural language comprehension capabilities, Yabble was able to rapidly transform complex, unstructured data into relevant themes and subthemes. Using GPT‑3, data sets that would usually take Yabble teams days to code and develop insights for, were now being translated into meaningful themes in minutes. GPT‑3 also allowed Yabble Query to understand and process more complex questions from users and respond with more germane insights that were sure to be based on the relevant data set.