More companies are moving to AI-based chatbots to limit the need for humans. This new Israeli startup nixes the bots, but sticks with the AI; using NLP models and with a little human intervention, they are trying to help developers provide better services, so no one gets upset at an unsuccessful, useless bot.
Not a bot, and not just AI
Israeli startup Tymely is developing a text-based customer service system that uses NLP models to solve problems. "The end customer sends an email that reaches the retailer's Zendesk. We interface with ZD and pass the text through our NLP model, which analyzes the text and extracts relevant pieces of information to solve the inquiry. The model we developed is based on a combination of several models, some transformers, some pre-trained with fine tuning on our data and some trained on our data from scratch," Ohad Rozen, founder and CEO of Tymely, explained in a conversation with Geektime.
Rozen said that the data used by their models include more than half a million correspondences between customer service representatives of different retailers about service representatives. They say that the models they developed allow identifying products in the text and unusual events, identifying answers to questions – such as the reasons for cancelling a deal or exchanging products– and understanding the intent behind the information sent by email to the company.
Tymely’s product isn’t completely automated; employees also check to make sure the correct model is being used if a customer is down to a small nuance – which according to Rozen is the area "where the bots fall". They are therefore able to correct the mistakes in understanding the need. "From here the AI takes the lead and handles the inquiry automatically," he added. This is how Tymely closes the gap between today's technology – which usually does not understand the text correctly and makes a lot of mistakes – and provides optimal customer service.
Although the product is not autonomous, it's not that Tymely doesn't want it to be. Rozen said that "In the distant future, of course, we will be able to reduce the use of humans, but we are already 10 times more efficient than a human service representative, even if he is aided by the AI systems that exist on the market today."
Let's take for example a company that experienced a tsunami of customer service calls in a short time: Strauss with the recall for Elite Chocolate. How was your system work in such an event?
Rozen: "Strauss was flooded with emails to its customer service. If Strauss had used Tymely’s system, 80-90 percent of the inquiries would have been handled within a few minutes and updated within a few hours for this specific case. And only a small part of the cases would be transferred to Strauss' team. Thus, on the one hand, the vast majority of customers would receive the care they expect from excellent customer service, and on the other hand, the Strauss customer service team would not collapse from overload, but be able to deal with a reasonable number of inquiries."
Yesterday (Monday) the Israeli startup, founded by Rozen and Roy Penn (CTO), announced that it has completed a fundraising round of $7 million. The round was led by the Israeli fund Hetz Ventures, and DESCOvery, the investment arm of D. E. Shaw Group also participated. The Israeli startup employs 10 people in its development center in Tel Aviv.