Resolving 82% of Banking Customer Queries Instantly: How DraconX Built a Generative AI Customer Service Assistant

A Small Finance bank in Malaysia with strong customers base needed AI that felt human not robotic. DraconX designed, built, and deployed a responsible generative AI assistant that now handles the majority of customer interactions faster, more accurately, and with measurably higher satisfaction than the previous model.

52%
Reduction in support centre operating costs
82%
Queries resolved without human escalation

The Opportunity

MID Bank operates in Malaysia which manages over 2 million active customer relationships. As digital banking volumes grew and in branch transactions declined, the bank’s contact centre faced escalating demand for support driven primarily by repetitive, high-volume queries around account balances, card management, transfer status, and product information.
Agent attrition was rising due to the repetitive nature of the work. Wait times were growing. And the bank’s leadership, committed to a people first philosophy, was unwilling to deploy a solution that felt impersonal or that left vulnerable customers without a path to human support. They needed generative AI done responsibly knowledgeable, empathetic, and safe.
Following a competitive evaluation, MID selected DraconX for its proven responsible AI framework, its experience in regulated financial environments, and its track record of deploying conversational AI that consistently scores well on both resolution quality and customer warmth.

We didn't want a chatbot that deflects. We wanted an AI that genuinely helps and knows when a human needs to step in. DraconX built exactly that.” — Dr. Fatima Al-Rashid, MID Bank

The Solution

A Responsible, Multilingual Generative AI Customer Assistant

DraconX led a discovery process that mapped MID’s top 200 customer query types by volume, complexity, and emotional sensitivity. This informed a deployment strategy that distinguished between queries suitable for full automation, those requiring AI-assisted human support, and scenarios such as financial hardship or bereavement, where human handling was non-negotiable.
The assistant was built on a large language model fine-tuned on MID’s own product documentation, policy library, and three years of anonymized interaction transcripts. DraconX’s safety and grounding layer prevents hallucinations, ensures all responses are anchored to verified policy content, and maintains regulatory compliance.
The development process followed DraconX’s responsible AI methodology: phased rollout with live customer testing, bias auditing across age, language, and demographic segments, and independent review of sensitive interaction flows including complaint handling and account closure.
Critically, the assistant was designed with sophisticated intent and sentiment detection identifying when a customer’s emotional state or query complexity warranted an immediate handoff to a human agent, complete with full context transfer so the customer never had to repeat a word
 

The Impact

The Support Centre Reimagined: Faster, Cheaper, and Genuinely Customer-Centric

Within six months of full deployment, 82% of all customer support interactions were being resolved entirely by the AI assistant exceeding the 65% project target by a significant margin. Average resolution time fell from 10 minutes for contact centre calls to under 2 minutes for automated interactions.

Support centre operating costs fell by 52%, enabling MID to redeploy agents to complex relationship management, financial hardship support, and proactive outreach roles reducing attrition markedly and improving agent job satisfaction scores. The bank reinvested a portion of the savings into expanding its financial wellness advisory service.

Digital support rose by 28 points which is a remarkable achievement in a channel that customers have historically rated poorly. Post-interaction surveys cited speed, accuracy, and clear communication as the primary drivers of satisfaction, with multiple customers specifically noting that the assistant felt like talking to a knowledgeable, attentive person rather than navigating a decision tree.

Call Us Today to Schedule a Free Consultation

Case Studies

Related Case Studies