How We Reduced Customer Support Costs and Admin Work by 60% Using Agentic AI

In 2025, we partnered with a commercial banking credit lending team to explore how Agentic AI could fundamentally transform the way lending operations function. Rather than approaching AI as a standalone chatbot or productivity layer, the ambition was to create an intelligent operational platform embedded directly into the credit origination and underwriting process. The vision centred on reducing manual effort, accelerating decision-making, and consolidating fragmented workflows into a single AI-powered workspace that could support lending teams end-to-end.

Our discovery phase focused heavily on understanding how lending teams actually worked in practice. Through a series of workshops with credit underwriters, relationship managers, and subject matter experts, we mapped the full lifecycle of a commercial lending deal. What emerged was a highly fragmented ecosystem: information spread across disconnected internal systems, repetitive manual document retrieval, time-intensive credit memo creation, and a heavy reliance on institutional knowledge to navigate risk assessment. These operational inefficiencies created significant cognitive load for teams managing complex lending decisions under pressure.

From these insights, we identified key decision-making moments where AI could provide meaningful operational support. The solution was designed as a multi-agent AI system capable of interacting with core banking infrastructure, retrieving documentation, and assisting with underwriting workflows in real time. We designed and prototyped several core capabilities, including a Deal Origination Agent that analysed incoming applications and surfaced risk signals, a Document Intelligence Layer that aggregated deal documentation via API integrations, and a Credit Memo Generator capable of producing structured, editable lending summaries. A conversational interface also allowed users to query deals using natural language, while a dedicated audit and traceability layer ensured transparency and compliance across all AI-generated outputs.

The experience itself was intentionally designed beyond the conventions of traditional enterprise software. Instead of a static dashboard-led interface, we created an intelligent workspace built around human-AI collaboration. Rapid prototyping sessions were conducted directly with lending teams, enabling concepts to be validated and iterated in real time. As the platform evolved toward production, we extended the existing banking design system into a dedicated AI product framework focused on clarity, trust, and explainability. Conversational UI patterns were introduced for structured financial data, alongside multimodal interactions including text, voice, and experimental AI avatar interfaces. The platform was integrated into the organisation's existing infrastructure using secure SSO access and deployed directly into the daily workflow of the lending teams.

The outcome was a shift from AI as an assistant to AI as an operational engine embedded within the credit lending process. Lending teams were able to generate and refine credit memos significantly faster, retrieve deal documentation instantly using natural language, and reduce their reliance on navigating multiple disconnected systems. The platform also improved consistency across underwriting outputs while creating a single centralised environment for managing complex lending activities. By combining Agentic AI with deep workflow discovery and experience-led product design, the project demonstrated how AI can move beyond automation and become an active participant in enterprise decision-making.