SpareBank 1 SR-Bank

SpareBank 1 SR-Bank, a Norwegian regional bank, faced mounting pressure to scale customer service operations without proportional headcount growth. Inquiry volumes across chat, phone, and email were climbing, and a significant share of contacts were repetitive — login errors, account access issues, and routine banking questions that consumed skilled agent time but required little judgment. Regional banks like SR-Bank operate with tighter cost structures than global institutions, making staff expansion an unsustainable response to rising digital demand. The bank also lacked a mechanism to absorb sudden volume spikes — such as system outages — without overwhelming its service center and degrading the customer experience.

In 2016, SR-Bank partnered with Scandinavian software developer boost.ai to deploy Banki, a virtual agent built on boost.ai's conversational AI platform combining machine learning and natural language understanding (NLU). Banki was trained to recognize over 1,900 customer intents, enabling it to handle a broad range of retail and business banking queries with precision. The deployment prioritized a 'chat-first' model reflecting actual customer channel preferences. A standout integration was proactive error handling: when customers encountered error codes during e-banking login — for instance during BankID outages — those codes were automatically routed to Banki, which surfaced context-specific guidance in real time. In 2017, SR-Bank extended Banki's capabilities to execute banking transactions directly, deepening its utility beyond informational responses.

Banki became the operational equivalent of 20 full-time employees, processing over 23,000 conversations per month and driving a 149.3% increase in support capacity — with no additional hires. The impact on incident response was especially measurable: during a major BankID outage in March 2018, Banki handled over 4,000 conversations in a single day, a volume that would have overwhelmed a human-only service center. Key outcomes include:

• 49.5% of total B2C and B2B customer support traffic fully automated
• 24% reduction in error-handling inquiries reaching human agents
• High self-service resolution rate — only 22% of conversations escalated to a human advisor during the October 2018 test phase
• Staff transitioned from repetitive query handling to higher-complexity advisory roles

• Proactive routing is a force multiplier: automatically redirecting system error codes to a virtual agent at the moment of failure eliminates a large class of inbound contacts before they form.
• Intent breadth drives containment: training a virtual agent on 1,900+ intents is what enables a high self-service resolution rate — shallow intent coverage produces shallow results.
• Channel preference should shape deployment strategy: SR-Bank's chat-first pivot aligned the AI investment with how customers already wanted to communicate, accelerating adoption.
• Workforce transition is part of the ROI story: repurposing support staff as AI Trainers preserves institutional knowledge while building the operational capability to sustain and improve the system over time.