Skip to main content

Banking

Quality assurance for underwriting with Agentic AI

Kyndryl ai icon mono-medium-light version

Business challenge and opportunity

For banks serving the micro, small and medium enterprises (MSME) sector, data quality is paramount. Underwriting decisions rely heavily on detailed interviews between bank executives and customers with a specific set of questions that require complete answers.

However, in the reality of daily operations conducted over audio calls, critical questions are often missed or responses are left incomplete. This leads to a cycle of rework where teams must re-contact customers to fill data gaps, delaying loan processing and frustrating clients. Quality Control (QC) teams must also manually listen to these recordings to verify completeness, a process that is time-consuming, error-prone and impossible to scale effectively.

Automating the review process with an AI agent can unlock new opportunities, especially for banks managing hundreds of applications a day. With appropriate governance and security controls, agentic AI can enable better, faster adherence to underwriting checklists, reduce the need for manual audio audits, and accelerate time-to-decision for MSME loans.

Technical challenges

While many banking leaders recognize the AI opportunity for underwriting, there are key operational and technical challenges to success at scale:

number 01

Accurate speech processing
Customer conversations can occur in multiple languages, making the process complex. In India, for example, it’s not uncommon for these conversations to require information gathering in Tamil, Oriya, Hindi or English. Conversations are often also affected by background noise, accents and inconsistent audio quality. If speech recognition is inaccurate, customer intent and compliance signals can be lost. This leads to unreliable insights, limits automation value, and increases the cost and risk of manual review.

number 02

Distinguishing speakers in real‑world conversations
The speech recognition system must perform precise speaker notations to reliably differentiate between the executive interviewer and the customer in live conversations and accurately capture information so that the backend process for ingesting and validating data is accurate.

number 03

Lack of contextual understanding
Simple transcription is insufficient. The right AI solution must be designed to interpret unstructured text and apply logic to determine whether specific questions from a multi-point checklist were both asked and adequately answered. Failing to validate conversations limits the organization’s ability to enforce standards, demonstrate compliance with their regulatory requirements, and improve sales or service outcomes.

number 04

Workflow disruption
Any agentic AI solution for conversation analysis and checklist validation must be integrated into the MSME application via Application Programming Interfaces (APIs), operating in real time without disrupting existing workflows. Poor integration slows frontline teams, reduces adoption, and creates operational friction.

Our solution

Using the Kyndryl Agentic AI Framework, Kyndryl helps banks achieve:

By automating quality assurance for MSME underwriting calls, Kyndryl agentic AI significantly reduces manual reviews across the workflow. This shortens underwriting cycles, enables teams to process higher application volumes, and supports business growth without a proportional increase in headcount.

Platforms like advanced AI algorithms, Automatic Speech Recognition (ASR) and Large Language Models (LLMs) are used to track multiple languages, accents, changes in vocal pitch, rhythm, and acoustic signatures to separate speakers and cross-reference interactions over the phone Accurate transcription across regional languages with different accents establishes a reliable foundation for underwriting decisions in linguistically diverse markets. This can improve data quality, reduce errors caused by misinterpretation, and can enable consistent underwriting standards at scale.

By validating conversations against a predefined multi‑point question checklist, the Kyndryl solution can support customers in consistently meeting their underwriting requirements.

Transforming unstructured audio into structured, actionable data gives underwriters clearer visibility into data completeness and customer responses. This can support enhanced risk evaluations and credit assessments.

Real‑time validation results are returned directly to the MSME application via API, enabling immediate feedback on missing or partially answered questions. This can prevent costly rework, reduce back‑and‑forth with customers, and improve operational efficiency and customer experience.

By integrating directly into existing MSME workflows, the solution can increase adoption and deliver value without slowing frontline teams. This preserves productivity while embedding intelligence exactly where decisions are made.

Kyndryl Consult applies deep banking industry expertise to define annotation frameworks and accuracy metrics aligned to financial underwriting standards. This enables explainability and defensibility to be core to our AI solution, so it can be trusted by risk, compliance and business stakeholders.

For banks at the beginning of their AI-native journey, we take a two‑phased approach that allows organizations to move from transcription to advanced, agent‑driven analysis over time. This helps futureproof investment by enabling additional automation, insights and decision support as AI maturity grows.

Benefits you could achieve

progress picto

By automating and applying agentic AI to underwriting call quality assurance, banks can achieve a number of high-impact outcomes:

  • Strengthen process compliance to minimize risk: Automatic validation that every required underwriting question is accurately asked and answered, significantly reduces operational risk.
  • Accelerate loan processing and improve customer experience: Eliminating the back-and-forth iterations caused by missing information enables faster credit decisions and disbursement while increasing customer satisfaction.
  • Scale Quality Control: Replacing manual, sample-based audio reviews with an automated AI system capable of auditing customer interactions speeds processing and saves time.
  • Support regional diversity: Seamlessly handling conversations in multiple languages, helps ensure consistent service quality across different geographies.
  • Enhance data integrity: Providing underwriters with a validated, structured status for every checklist item, reduces reliance on subjective human interpretation.

For one banking customer in India1, the Kyndryl Agentic AI implementation supports:

  • MSME underwriting in four languages – Tamil, Oria, Hindi and English
  • Significant analyst time saved in capturing underwriting questions
  • Improved quality of customer data tracked

Accelerate your journey to AI-native

Our AI Benchmark Assessment provides a maturity snapshot and personalized roadmap for accelerating the value of AI in your organization.Take the AI assessment.

Kyndryl AI Maturity Assessment

Why Kyndryl

With decades of mission-critical operations experience, we bring deep expertise in designing, building, and managing AI across complex IT estates.

The Kyndryl Agentic AI Framework brings together the right governance and security controls and human oversight to enable businesses to move beyond isolated pilots to integrated, intelligent systems. It combines secure, scalable agent orchestration with real-time observability and governance, powered by Kyndryl Bridge.

Through Kyndryl Consult, we align AI strategy with business outcomes. Kyndryl Vital provides proven user-centered design and rapid prototyping. And our trusted delivery experts support frictionless integration and help with reskilling and upskilling to build the foundation for success at scale.

We apply a forward engineering approach — using insights from existing systems to design and deploy future-ready AI agents and architectures that are adaptive, resilient, and scalable. Our method is grounded in open ecosystems, data sovereignty, and transparent decision-making.

And with a carefully selected alliance ecosystem — across cloud, data, and AI — we bring the right partners and technologies together to design fit for purpose solutions and orchestrate AI at scale without disrupting the business.

With Kyndryl, AI becomes a core capability — not just a tool — driving productivity, innovation, and growth.

Download digital version
Reshma agentic photo

What’s your enterprise vision for AI?

Let’s co-create it together.

Talk to an expert
  1. This document is current as of the initial date of publication and may be changed by Kyndryl at any time without notice. Not all offerings are available in every country in which Kyndryl operates. The performance data and client examples cited are presented for illustrative purposes only.  Actual performance results may vary depending on specific configurations and operating conditions.  Kyndryl products and services are warranted according to the terms and conditions of the agreements under which they are provided.