.png)
Raiffeisenbank is one of the five largest banks on the Czech market, offering a wide range of services for both private individuals and companies since 1993. In 2021, the bank launched several acquisitions, resulting in the merger with Equa bank in 2022.
The integration of new clients brought a sharp increase in inquiries to the customer center. In order to maintain a high standard of care even with the exponentially increasing volume of communication, the bank decided to strengthen its capacities technologically. The cooperation with SentiSquare in 2022 began with a clear motive: to automate routine communication and handle the increased workload.
Once the bank achieved this goal, the focus of the cooperation shifted even further. From 2025, the cooperation will primarily focus on customer interaction analytics, which helps to manage the contact center more effectively.
By proactively leveraging the insights of customer interaction analytics provided by SentiSquare, Raiffeisenbank increased its first contact resolution (FCR) from 83% to 90%, an increase of 7 percentage points in 6 weeks.
.png)
The merger with Equa bank brought a dramatic increase in communication, with clients of the original Equa bank writing emails even 2.5 times more often than the previous clients of Raiffeisenbank.
The customer center processes 90 thousand incoming calls, 45 thousand outgoing calls and 7 thousand emails per month. With such volume, it was crucial for the bank to manage the merger without losing clients, maintain satisfaction with customer care and at the same time be sure of compliance with strict banking regulations.
It was precisely in email communication, where the bank felt the growth the most, and where the cooperation with SentiSquare began. The main motivation was to automate the processing of emails: to sort them, assess their relevance and priority and, using smart routing, direct them directly to agents who can resolve the given request.
Raiffeisenbank decided to deploy the No-Code NLP platform SentiSquare, which uses a unique connection of AI models to analyze a huge amount of customer communication.
Today, the SentiSquare platform addresses key areas on several fronts in the bank: the development of bankers' soft skills, auditing, routing and automation of emails, including the creation of response suggestions.
Resolving a client's request the first time is a key indicator of satisfaction and efficiency for the bank and is one of the most important metrics in the bank's call center. SentiSquare measures FCR using specialized models that directly analyze the language of bankers (e.g. whether they say "you need to visit our branch" or "I'll enter your request right away and solve it"). Thanks to analytics and targeted work by supervisors, Raiffeisenbank managed to significantly increase FCR from 83% to a stable 90% within an incredible 6 weeks. This is a truly fundamental increase in the quality of handling.
One area that has contributed to the growth of FCR is targeted coaching of bankers based on detailed soft skills assessments.
Call center supervisors used to randomly select and listen to samples of calls to evaluate the work of telephone bankers. However, this manual method had very low relevance. Supervisors spent a lot of time finding errors in routine calls without having a chance to discover real problems. The bank therefore sought a tool that would show supervisors exactly which calls to listen to.
.png)
The bank has therefore changed its approach. To assess soft skills, the bank now uses small language models that assess:
Supervisors have immediate overviews for each call and can thus develop the banker in the right direction.
Thanks to the evaluation of 100% of communication, supervisors obtain much more detailed information about how the agent behaves in various situations and can give him more relevant feedback, thus improving the quality of customer care.
In addition to assessing bankers' soft skills, SentiSquare also ensures that emails reach the right people as quickly as possible. As soon as an email arrives at the bank, SentiSquare immediately evaluates its content, categorizes it, assesses its relevance and priority, and places the email in the correct queue. It immediately diverts approximately 15-20% of emails (spam, automatable queries) from the operator queue, identifies the urgency of the remaining ones, and directs them directly to a competent agent who can resolve the request without the need for lengthy forwarding.
Intelligent routing saves agents' time, and the bank has also managed to reduce handling time by 1 hour every year in the long term.
We are also currently working on expanding it to include dynamic suggestions and automatic responses. Instead of LLM generating a response "from scratch", it will search for the best answers that the bank has historically sent to a given problem. The tool will prepare a finished draft for agents, freeing them from concerns about inaccurate wording.
The SentiSquare platform is an absolutely key use for the bank in the area of audit control. The bank is subject to strict regulations on investment calls according to the European MiFID 2 directive. The No-Code NLP platform automatically checks 100% of investment calls and monitors key compliance rules: client authentication, recording warnings and compliance with mandatory texts for investment calls.
The audit of investment calls works on a hybrid model that combines small specialized models with LLM. First, relevance to MiFID 2 is detected - small language models evaluate whether it is a communication related to investments. Then specific topics, investment products and phases of the investment call are detected. The audit itself is then carried out, which looks for potential problems in the communication.
The resulting summary of investment calls using LLM then helps the compliance team to perform a faster manual check.
To gain control over 100% of interactions while keeping costs in check, the bank uses a hybrid approach that combines small and large AI language models. There are currently about 30 small specialized language models (SLMs) running on the SentiSquare platform.
The small language model can process and evaluate conversations more efficiently and quickly than the LLM. In addition, it is trained on the bank's data, which is why it achieves high success rates. The SLM thus processes 100% of all communication and at the same time determines which places the LLM should focus on for in-depth analysis.
It works as a perfect filter - it sorts out topics or process errors. Only for deeper understanding of the text, complex summaries or specific questions from supervisors does the large language model (LLM) get involved, which has a significantly lower risk of hallucinations thanks to pre-filtering using the SLM.
A dedicated team is responsible for the operation, development and continuous training of AI models in the bank, and this is a great advantage – people who sit directly in the bank, know the reality of the line and have absolute control and overview of the entire solution. They see exactly what AI is doing and what data it is working with, which, together with the operation on the bank's infrastructure, means that the bank has AI fully under control and does not have to worry about where the data is going.
Thanks to the deployment of the SentiSquare platform, Raiffeisenbank has gained a robust background that solves typical barriers when introducing artificial intelligence into a corporate environment.
AI Models are managed directly by business specialists from operations at the bank, because the platform works on the No-Code principle and does not require IT expertise to set up and manage it. The deployed small language models (SLMs) work lightning fast and allow for the analysis of 100% of communication, while their use also brings control over costs that could be unpredictable in the case of deploying large LLMs.
.png)
The platform does not require the purchase of expensive graphics cards and it runs within the bank's infrastructure, keeping sensitive data safe and under the full control of the bank.
The bank managed to increase the success rate of resolving requests at first contact (FCR) from 83% to 90% in just 6 weeks. Email processing is accelerated by one hour per year and supervisors have gained an exact overview of 100% of interactions.
The cooperation with Raiffeisenbank shows the enormous power of combining human management with advanced analytics. Supervisors have accelerated their work and now have comprehensive control over 100% of interactions. They know exactly which calls to listen to and focus their attention on the right places.
Thanks to daily supervision based on real data, the entire team has a huge motivation to continue working and improving.
Do you also spend hours in the customer service center listening to calls without being able to draw relevant conclusions from them? Or do you lack a comprehensive overview of the quality of your operators' work and would you like to give your people objective feedback instead of relying on random samples? With our solution, you get an overview of 100% of calls and data on the basis of which you can truly motivate and develop your people.
.png)
.png)
.png)