Стас Юрасов AI Eng 15 June 2026, 12:12

AI lost to Surzhyk and dialects: why SK UNIVERSALNA could not replace the medical call center with artificial intelligence

During the Tech Week in Odesa conference, Sergey Sabarin, Executive Director of Health Insurance at UNIVERSALNA, shared a candid case study of automation. The company dreamed of implementing an AI assistant to solve the problem of high staff turnover in its medical call center, but the project had to be put on hold.

"Hell of a job" and 100% staff turnover

Medical insurance is one of the main tools for retaining staff in Ukrainian companies today. However, medical assistance (call center) itself suffers from a severe shortage of personnel.

"We have very specific requirements for employees: they must be people with medical education, able to make decisions quickly and have a huge level of empathy. It's hellish work. So that you understand the scale: we hire 5 new people every week and 5 people are fired every week. The staff is stable at 150 people, although with such a rotation there should be 1500," shares Serhiy Sabarin.

The implementation of AI was supposed to be a "blue dream" - to close the processes of search, long training, and adaptation of operators.

Why AI “broke” in practice: the main reasons

The company collected a huge database of scripts, diagnoses, and invested huge resources into training the model. However, critical problems arose during the testing phase:

Transcription and language barrier. Since customers communicate by voice, the audio had to be translated into text. The company has over 60,000 insureds throughout Ukraine.
Dialects and surzhyk. AI turned out to be unprepared for the linguistic diversity of Ukraine. According to the director, when clients from Transcarpathia or Poltava region call, their specific pronunciation and surzhyk “break any model.”
The price of a mistake is life. Medicine is a highly sensitive field. Any mistake by AI in recognizing a diagnosis or prescribing treatment can be fatal. The company cannot take such reputational and human risks.
Economical Infeasibility. The resources spent on manually checking the remaining AI work exceeded the costs of people. It turned out that hiring and training 5 new employees every week is now cheaper and easier than maintaining the current AI model.

UNIVERSALNA is looking for partners: a challenge for Ukrainian AI developers

Despite its failure in medical assistance, UNIVERSALNA is not giving up on technology. The company has already successfully implemented AI in other, less sensitive and safer departments.

Moreover, Serhiy Sabarin is open to suggestions from the Ukrainian IT community:

"My goal is to show that we are ready to work further. If there are companies or developers among you who know how to solve the problem of recognizing complex Ukrainian language (surzhyk and dialects) in medicine, we are open to cooperation. Let's join forces to present a cool working case at the next conference."