Citadele Banka answers customer queries in under 5 seconds on average with Azure AI assistant Adele

Citadele Banka wanted to improve efficiency by reducing manual workload, automating repetitive operational tasks, and letting its people access standardized answers based on internal knowledge. Customers with questions about products and services had to wait for a response from the call center. Every day the bank also processes a large variety of documents across operational, insurance, financial, legal, identity and compliance domains, at around 3-5 minutes per document before the new platform.

After an initial pilot, Citadele Banka used Microsoft Copilot and Azure AI services to build Adele, a customer-facing multi-agent virtual assistant, along with 10+ internal AI agents and an AI-powered Document Intelligence Platform. Adele retrieves information from over 3,500 data sources and typically answers queries about the bank's products and services. The internal agents, available through Microsoft Teams, Microsoft 365 Copilot and internal Citadele Chat, cover IT support, HR questions, internal procedures, learning materials, and support for SkyBranch and call center employees. The Document Intelligence Platform automates the end-to-end process of converting unstructured data into structured data, registering documents in business systems and electronic archives, and integrating with digital channels.

Adele handles around 40% of the bank's daily customer conversations and responds in under five seconds on average, with answer quality at 87%; since December 2024 its usage has increased by 221%, and by 459% through the mobile app. The Document Intelligence Platform cut processing from around 3-5 minutes per document to 5-15 seconds, now handles 10,000–12,000 documents per month, a 4x increase over the past year, and has a 99.6% accuracy rate. The bank says each new document type usually reaches break-even in 2–3 months, and the IT Support Copilot achieves satisfaction rates close to 95%.

• Run pilots first to confirm the models deliver practical value, then scale the successful ones with in-house builds on the cloud platform.
• Give employees the skills and confidence to use AI through training events and communities, since adoption depends on people as much as technology.
• Let staff build their own agents with no-code and low-code tools alongside centrally delivered ones, so improvements come from the teams doing the work.