--- title: "Group Credit Life Pricing Model for an Indian Life Insurer | Agility" url: https://agilitytech.ai/case-studies/group-credit-life-pricing description: "Case study: a leakage free claim rate model and rate card multiplier framework for a home loan group credit life portfolio at a leading Indian life insurer." publisher: Agility (agilitytech.ai) --- Insurance · Life insurance pricing # Group Credit Life Pricing Model A leading Indian life insurer priced group credit life cover for home loans from a flat rate card. We built a claim rate model and pricing framework that prices each microsegment on its own risk. 5 years Member level experience Claims history from FY22 to FY27 used to train and test the models Leakage free Walk forward training Each year trained only on earlier years, every feature checked for leakage 10 years Forward projection Hybrid engine projecting claim rates and retention a decade ahead 11 steps Handover runbook Written so another team can resume the work within a week **The short answer:** we replaced a flat rate card with a microsegment pricing model for a leading Indian life insurer’s home loan group credit life book. A leakage free walk forward Poisson GLM predicts claims per 1,000 lives, a hybrid engine projects them ten years ahead, and the result becomes rate card multipliers the actuaries apply on top of their own baseline, rerun each financial year. ## Project Overview ### Industry Life insurance ### Region India ### Portfolio Home loan and loan against property group credit life ### Time frame 2026 ### Technology Stack AWS SageMaker and S3 Python (pandas, scikit-learn, statsmodels) Poisson GLM and gradient boosting Dynamic time warping (fastdtw, tslearn) Streamlit and Plotly dashboards Amazon Bedrock for RAG summaries ## The Challenge ### Pricing a bank channel portfolio without microsegment data Group credit life insurance pays off a borrower’s loan if they die during the loan term. The insurer sells it through banks and housing finance partners alongside home loans and loans against property. Its rate card was flat and set by the actuarial team, so it could not show which borrower segments were over or under priced. The team wanted pricing grounded in five years of member level claims experience, and decisions they could explain to internal governance and to bank partners at renewal. ## What We Delivered ### From a flat rate card to microsegment pricing A validated, leakage free walk forward pipeline predicts claim rates for every microsegment, projects them ten years ahead, and turns the result into rate card multipliers the actuarial team applies on top of its own baseline. Because the pipeline is documented and repeatable, it can be rerun each financial year without rebuilding it. Even a modest reduction in mispricing at microsegment level matters on a portfolio of this size, both for margin on new business and for reserving on the existing book. #### Pricing and actuarial Rate card multipliers structured as an additive layer over the actuary’s baseline table, then a second, policy specific layer Microsegments that were consistently over predicted identified and handled separately Red, amber and green risk classification rebuilt from the team’s own manual labels #### Engineering and handover Member level data processed on AWS SageMaker and S3 inside the insurer’s own cloud environment Interactive dashboards for actuaries to explore predicted versus actual claims by segment Plain language policy summaries generated with Amazon Bedrock, plus an 11 step runbook for successors ## Challenges & Solutions ### A flat rate card for a varied book #### Problem Group credit life cover for home loans and loans against property was priced from a flat rate card set by the actuarial team. Borrowers differ widely by age, loan tenure, region and lending partner, so some segments were overpriced and others underpriced. #### Solution We built a model that predicts claim rates per 1,000 lives for each microsegment: policy, financial year, age band and tenure band. The output feeds rate card multipliers the actuaries can apply directly. #### Impact Pricing that reflects the risk of each microsegment ### Models that look good only because they peek #### Problem Claims data is easy to leak. Branch, region and state claim rates, and loan mix percentages, can quietly include information from the very year being predicted, which makes a model look far better on paper than it will be in use. #### Solution We used a walk forward method: each target year is trained only on data from earlier years, with policies split 75/25 and a compulsory held out set. Every rolling rate was rebuilt to use prior years only, and features lag by one financial year. #### Impact A leakage free pipeline that can be rerun every year without being rebuilt ### Policies of very different ages #### Problem A policy in its first years behaves differently from one that is seven or ten years old, and one model across all of them blurred those patterns. #### Solution We fit three window models by policy age (0 to 3, 4 to 6 and 7 to 10 years) for each target year, and grouped similar claim patterns with dynamic time warping clustering inside each window. #### Impact 15 model fits per full run, each matched to its policy age window ### Seeing ten years ahead #### Problem Pricing decisions need to hold for the life of the cover, not just the next year, and member numbers fall as loans are repaid or closed. #### Solution A hybrid projection engine uses data driven multipliers for years one to five and the published IALM 2006 to 08 mortality table for years six to ten, with member level retention decay. A five policy validation framework checks projections against real member survival tables. #### Impact Forward projections the actuarial team can test and explain More on our insurance work: [AI consulting for insurance](https://agilitytech.ai/ai-consulting-insurance) ## Frequently Asked Questions **What is group credit life insurance?** Group credit life insurance pays off the outstanding loan if a borrower dies during the loan term. In India it is usually sold through banks and housing finance companies alongside home loans and loans against property, under one master policy for the lender. **How do you price group credit life cover with machine learning?** We predict the claim rate per 1,000 lives for each microsegment, such as policy, financial year, age band and tenure band, using a walk forward Poisson GLM trained only on earlier years. The predictions are turned into rate card multipliers that sit on top of the actuary’s own baseline table. **What is data leakage in insurance pricing models, and how did you avoid it?** Leakage happens when a feature quietly includes information from the year being predicted, for example a branch claim rate that counts this year’s claims. We rebuilt every rolling rate to use prior years only, lagged mix features by one financial year and kept a compulsory held out set of policies. **How far ahead does the model project claims?** Ten years. Years one to five use data driven multipliers, and years six to ten use the published IALM 2006 to 08 mortality table, with member level retention decay as loans are repaid or closed. **Can the actuarial team explain the prices to regulators and partners?** Yes. Every multiplier traces back to a microsegment’s own claims experience, and the framework is documented in an 11 step runbook, which supports internal governance and renewal discussions with bank partners. ## Pricing a portfolio on a flat rate card? Tell us about your book and your data. We will show you what microsegment pricing could look like, with a model your actuaries can explain. [Book a free 30 minute call](https://calendly.com/nihar-t-agilitytech/30min)[See case studies](https://agilitytech.ai/case-studies)