The company and the task
Client: One of the ten largest banks in Belarus; winner of the Bank of the Year in Belarus grand prix in 2020, 2022 and 2024.
Industry: Banking and fintech · Year: 2025
They needed a complete analysis of 100% of contacts through better processes rather than more headcount: a deeper understanding of what customers need, freeing staff from routine, finding hidden flaws in processes and fixing them quickly.
Before
Sample-based checks did not give a representative picture, scoring one conversation took up to 20 minutes, complaints surfaced after the event, agents tagged the reasons for contact by hand, and a huge archive of recordings and chats went unused.
What we did
- A vendor comparison and a pilot; a phased rollout across every part of the contact centre, then into the chats; a pilot and full rollout across every branch in the network.
- Selection criteria: pre-built dictionaries, working with single-track recordings, emotion recognition, the Belarusian language, logging of repeat contacts (non-FCR), a personal dashboard with notifications for employees, and an API with no limits.
- Further tuning of the speech models: recognition quality of 95% in remote channels and 82% in branches (recordings from audio badges).
- Analysis of repeat contacts and of customers prone to complaining; of irritating situations — phrases that are not customer-focused, unjustified transfers; and of the link between chats and the calls that follow.
Timeline: 2022–2025: from choosing a vendor to a rollout across every channel
Results in numbers
What the client says
«The bank’s specialists are surfacing a great many interesting and important insights that customers never state outright. Contacts are moving into digital channels, which takes a substantial load off staff in every channel. At the same time the length of conversations has fallen by an average of 10%.»
