ProposalOverview
Proposal

AI at RBL Bank: lower cost on the AI already in use, RBL-owned models, and new use cases

RBL Bank has moved Generative AI from pilots to enterprise-wide use across 1.5 crore customers and 47 lakh credit cards. Dvar proposes to run that AI on models RBL owns, at lower cost, to validate every model, and to build new use cases with RBL's teams.

AI in use at RBL todayLower cost, same quality

  1. Generative AI across service, operations and risk
    • Enterprise-wide adoption in FY26
    • RBL Care 2026: 83% of tickets resolved in 24 hours
    • IBA award for Best AI & ML Adoption
  2. 6.3× lower cost at comparable quality
    • 324 card queries in 12 languages
    • 99.6% accuracy; Gemini 3.8 Flash 99.7%
    • ₹15 per 1,000 responses; Gemini ₹94

Smarter, RBL-owned modelsPost-trained and validated

  1. Models trained on RBL's own data
    • Post-trained on RBL's tickets, cases and rules
    • Hosted in India or in RBL's cloud
    • Model weights owned by RBL
  2. Evidence for RBI's model risk guidance
    • Inventory of every model in use
    • Independent validation, vendor models included
    • Every response evaluated; fallback on failure

New use cases with DvarDesigned and built with Dvar's team

  1. Card collections within fair-practice rules
    • Next step and message for each overdue account
    • In the cardholder's own language
    • Officer approval on every message
  2. Further use cases on RBL's priorities
    • NRI and FCNR onboarding with Emirates NBD
    • Document reading for business and secured loans
    • Account-opening risk checks

Public facts referenced: RBL Bank MD & CEO's FY26 AGM speech and Q1 FY27 earnings call; RBI draft guidance on model risk management (June 2026). Accessed 22 Nov 2026. Evaluation results are Dvar's, on sample data.