as possible to the client, the answer must be 80% correct."

[tanjimajuha20]

Alexander Krushinsky reminded that for the incoming line, the standard (good) automation indicator is a value in the range of 30%-60%, but the corporate segment is one of the most difficult segments for automation: "Employees of corporate clients are "professional" clients of the bank, who, as a rule, know the bank's products and know how to use the tools that the bank provides. As a rule, such clients ask much more complex and diverse questions than "individual" bolivia whatsapp number database clients. Therefore, it seems to me that 23% is a very decent indicator, especially if you take into account the number of Sber employees and imagine how many hours of operator labor are saved due to these 23%. An increase in accuracy by 14 p.p. is also every seventh call. The result is at least tangible."

According to Evgeny Surkov, the role of AI in achieving the indicator remains questionable: "Optimization of the support service is always a complex process, ideally intersecting with all banking service processes. Therefore, there may be many explanations for the improved indicators. For example, the routes/scripts of the support itself may not have been optimally composed earlier, the user path in remote service and offline services of the bank may have been redundant or unclear. With optimization for AI, existing problems could be identified and eliminated, which could bring the main effect. Sber also notes the effectiveness of AI bots in routing and in helping operators. But 23% of customer requests resolved directly by AI, without involving an operator, say that full automation of support procedures is still a long way off and AI is more about helping operators than about independently resolving issues. Most likely, any issues resolved by AI can be resolved without contacting the support service - for example, by carefully reading the manuals. Where non-trivial solutions are required, AI will not be able to come up with anything on its own, without human participation, for now. This would require giving it unsafe powers. And the clearer the procedures and manuals, the less someone is needed for additional explanation - no matter whether it is a bot or an operator."

"Automating processes with business clients is usually more complex than with individuals," says Maxim Milkov, leader of the AI ​​direction at Softline Group. "The point is how many scenarios can be identified in communications with the contact center. The fewer scenarios and the more homogeneous they are, the greater the effect can be expected from the implementation of such AI assistants."

Sberbank also uses artificial intelligence to predict the customer satisfaction index (CSI). At the same time, a set of machine learning models predicts this indicator for all requests. And contact center operators now have a personal AI mentor who works in the background and is always ready to answer any question.

Sergey Lekhanov, Director of the Corporate Solutions Center Division of Sberbank, commented on the results of the project: "We see a huge positive effect from the AI ​​transformation of our contact center. And this is not only a saving of the bank's funds, but also, most importantly, a saving of our clients' time. Now entrepreneurs receive a qualified answer and an effective solution to their problem even faster, since the routing speed has increased several times. The qualifications of our specialists have also increased in the same way, since they use an AI mentor, which simplifies the work and increases the benefit for the client."
Maxim Milkov believes that the effect of implementing AI depends on how optimal the process was before its implementation: "For example, if routing to an operator took an average of 15 seconds, then it is most likely quite difficult to achieve an additional effect. On the other hand, if the process is initially very suboptimal, for example, the average routing time is 20 minutes, then the solution to the problem may consist of conducting business analytics, including based on data, to identify "bottlenecks" and make changes to processes without using AI."

Igor Afonin, Head of the Multimedia and Unified Communications Competence Center at T1 Integration LLC, noted that post-processing of customer calls can enhance the effect of using an AI assistant: "Using the call database as a dataset for training AI, it is possible to conduct analysis and replenish the knowledge base for specific areas of customer cases. The model can also be further trained, which should further provide an additional effect in both routing and service."