The Russian neural network has been taught to determine the customer's willingness to buy a car

Avito Auto has refined the A-Vibe corporate neural network, training it to analyze car dealers' dialogues with potential buyers and determine customers' willingness to purchase a car.
This is reported in the company's press release.

It is noted that the neural network determines the level of customer interest and tells dealers how to improve the quality of service. "The model generates a detailed report in just 30 seconds, which is 10 times faster than the average time it takes to manually analyze a call. At the same time, the accuracy of such neuro-analysis reaches 95%, and people often make several times more mistakes," said Artur Shcheglov, technical director of Avito Auto.

According to him, dealers can check the quality of the model: some of the calls are selected for manual verification, which allows the system to be trained on new examples. "The first features are already available to all dealers on the site, and by the end of the half—year they will receive additional comprehensive analysis tools, including general statistics on the dealership and detailed indicators for each manager," said Shcheglov.

Avito Auto estimates that dealers lose up to 34% of potential deals due to managers' mistakes in communicating with customers.

A-Vibe was trained on a specialized dataset that includes impersonal transcriptions of real conversations with clients, which allows her to understand natural speech. The model performs an in-depth analysis of the dialogue, assessing, in particular, the intonation and markers of customer interest.