AI-Driven Portfolio Optimisation for Indian Commercial Insurers 2026: PML Aggregation, Reinsurance Sourcing, and Profitability

Indian general insurers spent two decades building underwriting capability at the account level. The discipline of pricing an individual fire and special perils risk, a marine open cover, a contractors' all risks construction project, or a directors and officers liability programme matured through the 2000s and 2010s with growing sophistication in occupancy schedules, hazard categorisation, exposure assessment, and named perils analysis. The 2020s have shifted the operational frontier from the account level to the portfolio level. Indian commercial insurers in 2026 face questions that account-level underwriting cannot answer: where is our peak accumulation in Mumbai if a Mumbai floods event recurs, what is our concentration on chemicals exposure across Gujarat, what would happen to our combined ratio if our largest broker's portfolio reprices materially, how do we know whether our reinsurance treaty is appropriately structured for our actual portfolio composition rather than the composition the treaty was designed for three renewals ago.

These questions require portfolio-level analytical capability that combines exposure data, occurrence and loss history, treaty terms, and forward-looking scenarios. The traditional approach used spreadsheets and consultant studies, with portfolio analysis occurring perhaps annually around treaty renewal and remaining static between cycles. The 2026 approach uses automated pipelines that maintain a continuously updated portfolio view, support scenario analysis on demand, and feed both the underwriting authorisation process and the treaty placement strategy. AI is part of the technical stack supporting this transition, with applications across exposure data normalisation, accumulation computation, scenario simulation, and treaty optimisation.

The driver is partly regulatory and partly commercial. IRDAI's progression toward risk-based supervision through the IRDAI Risk Based Capital framework consultation and the broader supervisory direction signalled through circulars in 2024 and 2025 expects insurers to demonstrate portfolio-level risk management, with documented PML and accumulation controls, stress testing capability, and treaty structures that match the actual exposure profile. The commercial driver is the underwriting cycle: the Indian commercial insurance market has tightened through 2024 to 2026 with reinsurance pricing reflecting global hard-market conditions, and the insurers with the better portfolio analytics secure better treaty terms because they can document the risk and respond to reinsurer queries with credible data. The combination of regulatory direction and commercial pressure has produced material investment in portfolio analytics capability across the top 10 Indian general insurers and increasing investment at mid-sized carriers.

Portfolio analytics depends on quality exposure data, and exposure data quality is the single most underbuilt foundation in Indian commercial insurance. An insurer cannot compute an accurate PML for its Mumbai concentration if its bound portfolio data has unclear geocoding, inconsistent occupancy classification, or missing construction type detail.

The exposure data fields that a portfolio analytics pipeline requires include: precise geocoding at the property or risk level (typically latitude and longitude with sub-200 metre accuracy), occupancy classification using a controlled taxonomy aligned to industry catastrophe model schemas, construction type with sufficient granularity to differentiate vulnerability (steel-framed industrial structure versus RCC frame versus masonry versus light construction), year of construction, height or storey count for the building, total insured value broken down by building, plant and machinery, stock, and business interruption, applicable deductibles, business interruption indemnity period, named insured and parent organisation linkage for accumulation aggregation across legal entities, broker reference for portfolio-by-broker analysis, treaty allocation for facultative versus treaty risks, and the policy effective and expiry dates.

The data ingestion challenge is that the policy administration system frequently does not capture all these fields with required quality. Indian commercial insurers built their policy systems with a focus on policy issuance and premium collection, with the exposure data fields populated to a level sufficient for the underwriting discussion but not always to the level needed for portfolio analytics. The 2026 portfolio analytics initiatives typically include a data remediation work-stream that backfills the missing fields and corrects the inaccurate fields across the existing book.

The AI applications in exposure data work focus on three areas. First, address parsing and geocoding uses NLP models to parse free-text address fields (which dominate the Indian policy administration data) into structured address components, then resolves to a precise geocode using a combination of Indian postal data, OpenStreetMap, Google Maps Platform, MapMyIndia, and Bharat Maps. The geocoding accuracy on Indian commercial addresses with AI-assisted parsing reaches 88 to 94 percent at sub-200-metre accuracy, compared with 55 to 68 percent for traditional rule-based geocoding. Second, occupancy classification uses LLMs to map the free-text business description in the policy to the controlled occupancy taxonomy, with the model trained on labelled examples and supervised by an underwriter for the higher-risk classifications. Third, construction type inference uses a combination of policy data (where construction is partially recorded), site survey reports (where available), and external building footprint and image data to classify construction type at the property level. Where direct data is insufficient, the model produces a probabilistic classification with confidence scores that downstream analytics can use.

The data quality scoring is a productive output of this layer. The pipeline produces a per-risk data quality score that the underwriting and analytics functions use to interpret the downstream PML and accumulation numbers. A risk with poor data quality should not be excluded from the portfolio view but should be flagged as carrying higher uncertainty, with the underwriting and treaty discussion accommodating that uncertainty.

With acceptable exposure data, the analytics pipeline computes the portfolio-level risk metrics that the underwriting and treaty processes need. The 2026 pipeline produces several metric families running continuously rather than annually.

The geographic accumulation metric aggregates total insured value, total premium, and modelled loss at multiple geographic granularities. The granular level uses 1-km, 2-km, and 5-km grid squares that align to common natural catastrophe model footprints for flood, earthquake, and wind perils. The mid level aggregates to administrative units (district, state) for regulatory reporting and broader portfolio review. The macro level aggregates to broader hazard zones (coastal flood zones, seismic zones per the IS 1893 zonation, cyclone exposure corridors).

The peril-specific PML computation combines the exposure data with peril vulnerability functions to produce loss estimates at various return periods. The methodology in common use computes PML at 50-year, 100-year, 250-year, 500-year, and 1,000-year return periods for the major natural catastrophe perils relevant to Indian portfolios: monsoon flood (with sub-models for riverine flood and pluvial urban flood), cyclone wind (with the east coast Bay of Bengal exposure dominant), earthquake (with the high-hazard zones IV and V driving the upper return periods), and storm surge (with the coastal exposures relevant for selected portfolios). For each peril at each return period, the model produces the aggregate gross loss and the per-event maximum loss.

The catastrophe model providers active in the Indian market include the global firms RMS (now part of Moody's), AIR Worldwide (now Verisk), Karen Clark and Company, JBA Risk Management (specialised in flood), Munich Re's NATHAN system, Swiss Re's CatNet, and Hannover Re's Cassandra. Indian providers and insurance-specialist consultancies including CAT Risk Solutions, Aon Risk Analytics India, Marsh JLT India catastrophe team, WTW Risk and Analytics, and emerging Indian specialist firms offer model calibration and bespoke modelling for Indian peril specifics. Top-tier Indian insurers increasingly use multiple model providers with the outputs reconciled through a model-blending methodology, producing a more reliable central estimate than any single model. The model blending requires methodology discipline; the outputs from different providers can differ materially due to underlying assumptions on event frequency, hazard footprint, and vulnerability.

The line concentration metric aggregates exposure and risk metrics across lines of business to identify portfolio-level concentrations that may not be visible within a single line. For example, an insurer's combined exposure to a major chemical manufacturer through fire and special perils, public liability, marine cargo, and directors and officers insurance may exceed prudent concentration limits on the named-counterparty basis even if each line in isolation is well-controlled. The metric typically computes the largest single named-counterparty aggregate exposure across all lines, ranked, with the top 25 or 50 reviewed at portfolio governance.

The occupancy and sector concentration captures the portfolio's exposure to broader sectoral risks. A portfolio heavily concentrated in cotton textile manufacturing carries a different sectoral risk profile from a portfolio diversified across textile, food processing, pharmaceuticals, and IT services occupancy. The 2026 analytics typically reports the top 10 or 20 occupancy concentrations by line, the top sector concentrations across lines, and the year-over-year change in the concentration profile that the underwriting actions are driving.

The clash and accumulation risk between lines is the more sophisticated metric that mature analytics pipelines maintain. Clash risk arises when a single event triggers losses across multiple lines on the same insurer. A major industrial fire can trigger fire and special perils loss, business interruption loss, public liability loss (if the fire injures third parties), motor loss (if vehicles in the premises are damaged), and marine cargo loss (if goods in transit at the time are affected). The pipeline identifies the named-counterparty or named-site exposures where clash potential is highest, supports the treaty design that accommodates clash, and informs the underwriting limits on such risks.

Stress testing scenarios

The pipeline supports scenario-based stress testing as a productive output. The common stress scenarios for Indian insurers include: monsoon flood recurrence (Mumbai 2005-equivalent, Chennai 2015-equivalent, Bengaluru 2022-equivalent, Kerala 2018-equivalent), east coast cyclone landfall (Fani-equivalent on Odisha, Mocha-equivalent on Andhra Pradesh, Cyclone Yaas-equivalent on West Bengal), Himalayan seismic event (Kashmir-equivalent magnitude 7.6 event, Sikkim-equivalent magnitude 6.9 event), and industrial concentration scenarios (single-site loss at named major insureds in chemicals, refining, power generation). Each scenario produces estimated gross loss to the portfolio, net of reinsurance recoveries, and the capital impact under the Indian solvency framework. The scenario outputs feed the board risk reporting, the treaty discussion, and the IRDAI risk-based supervision dialogue.

Beyond PML and accumulation, portfolio analytics in 2026 increasingly drives profitability segmentation: identifying which client segments, broker sources, geographic markets, and occupancy classes are profitable and which are not. The analytics supports the underwriting strategy and the broker relationship management with evidence rather than impression.

The portfolio profitability metric for each segment combines: gross written premium, paid losses, outstanding case reserves, IBNR allocation, broker commission, ceding commission earned, allocated operating cost, and reinsurance ceded premium net of recoveries. The output is a segment-level combined ratio with the underlying expense and loss components separately visible. Indian commercial insurers historically reported these metrics at line level and at major occupancy class level; the 2026 analytics extends the reporting to the client segment, broker source, geographic market, and combined-cut levels that support actionable underwriting decisions.

The AI application in the profitability analytics focuses on three areas. First, loss development and IBNR projection uses machine learning methods that extend traditional actuarial triangulation. Gradient-boosted models and survival analysis methods produce loss development factors that vary by segment and by claim characteristics, with the segment-level IBNR being more accurate than a portfolio-wide development factor applied uniformly. Several Indian insurers have implemented ML-enhanced IBNR through 2024 to 2026 with the appointed actuary's oversight, with the model outputs supporting rather than replacing the actuarial judgement. Second, broker source profitability analyses the loss experience by broker, identifying brokers whose books consistently outperform or underperform the portfolio average after adjusting for the mix of risks. The analysis informs the broker relationship management, the commission structure discussion, and the strategic broker selection. Third, client segment retention prediction uses ML models to predict which clients are likely to non-renew and which are likely to renew at acceptable terms, supporting the renewal workflow prioritisation and the price elasticity discussions.

The segment-level pricing relativity is the output that drives action. The analytics produces a relativity that the segment's risk warrants relative to the portfolio average, separating the relativity into the loss component (modelled loss cost relative to portfolio average) and the expense component (acquisition cost and operating cost for the segment). The relativity supports the underwriting decisions on new business pricing, renewal price increases, and segment-specific terms.

The client segment definition that supports the analytics combines several dimensions: industry sector (using the controlled taxonomy aligned to NIC classifications and to the catastrophe model occupancy schema), client size (turnover band, sum insured band, employee count band), geography (state or region), policy structure (sum insured, deductible, indemnity period for BI), and relationship history (years with insurer, claims experience over the relationship, renewal rate). Combinations of these dimensions produce thousands of micro-segments; the analytics typically rolls up to 30 to 80 actionable segments that the underwriting strategy targets.

The broker performance metrics that support the broker relationship management include: gross written premium contributed, loss ratio achieved, new business count and conversion rate, renewal retention rate, average policy size, mix of risks (concentration in particular sectors or geographies), policy administration quality (data quality on submissions, completeness of supporting documents, claims documentation quality). Indian commercial insurers in 2026 increasingly maintain broker-level scorecards that consolidate these metrics, with the analytics output supporting the broker review meetings and the strategic broker decisions.

Reinsurance treaty design has historically been a periodic exercise driven by the renewal calendar, with deep analysis in the months before treaty placement and limited adjustment between cycles. The 2026 approach uses continuous portfolio analytics to inform treaty design throughout the year, supporting the renewal cycle with documented analysis and providing the insurer with the ability to respond rapidly to portfolio changes that affect treaty performance.

The treaty optimisation question asks: given the insurer's portfolio composition, risk appetite, and capital position, what reinsurance structure produces the best combination of risk transfer, ceding commission economics, and counterparty diversification. The question has multiple dimensions including the choice between proportional treaty (quota share with or without surplus), non-proportional treaty (excess of loss for property catastrophe, for property per risk, for casualty per occurrence, for casualty stop loss), facultative versus treaty for specific risks, and the layering structure within excess of loss programmes.

The AI and analytics applications in treaty optimisation focus on three areas. First, the portfolio loss simulation that produces the gross-loss distribution across the modelled exposure. The simulation uses the catastrophe model outputs, the historical loss experience, and the forward-looking exposure to produce a probability distribution of annual aggregate loss and a separate distribution of large single-event loss. The distributions feed the treaty design through sensitivity analysis on different attachment points, limits, and reinstatement structures. Second, reinsurer counterparty modelling evaluates the credit quality, capacity, and historical pricing behaviour of the reinsurer panel, supporting the cession allocation across reinsurers and the diversification of counterparty risk. Third, renewal cycle automation maintains the treaty submission packages, the responses to reinsurer queries, and the documentation throughout the year, with the renewal cycle execution becoming faster and more accurate.

The treaty layer analysis is the most analytically productive application. For each treaty layer (the per-risk excess of loss attaching at INR 50 crore with INR 100 crore limit, the catastrophe excess of loss attaching at INR 500 crore with INR 1,000 crore limit, the casualty excess of loss attaching at INR 25 crore with INR 50 crore limit, and so on), the analytics produces the expected ceded losses, the expected ceded premium, the modelled reinstatement utilisation, and the layer's contribution to the insurer's overall risk-adjusted return on capital. The analysis allows the insurer to evaluate alternative layer structures (different attachment, different limit, different reinstatement terms, different aggregate features) and select the structure that optimises the combined economic and risk objectives.

The proportional treaty optimisation for the major Indian insurers focuses on the quota share retention level, the surplus capacity, the cession to GIC Re under the obligatory cession regulations, and the optional cessions to international treaty markets. The 2026 analytics evaluates the trade-off between retained net income (favoured by higher retention) and capital efficiency plus ceding commission economics (favoured by lower retention with appropriate surplus structure). The trade-off varies by insurer's capital position, growth ambitions, and current return on equity targets.

The facultative versus treaty decision for specific risks uses the portfolio analytics to identify risks that should not pass through the treaty due to outsized size, unusual characteristics, or specific exclusion provisions. The 2026 decision pipeline automatically flags the facultative candidates at quotation stage, supporting the underwriting decision and the placement strategy.

The reinsurance market dynamics in 2026 reflect ongoing hard-market conditions on international property catastrophe, easing conditions on some specialty lines, and selective capacity on cyber and casualty. The major Indian treaty placements at the April 2025, October 2025, and April 2026 renewals saw price increases of 4 to 12 percent on property catastrophe layers, 0 to 6 percent on per-risk property layers, 8 to 18 percent on casualty programmes, and 12 to 25 percent on cyber programmes, with significant variation by cedent based on portfolio experience and analytical sophistication. Cedents with strong portfolio analytics and documented exposure quality consistently secured better terms than the portfolio averages.

The reinsurer panel for major Indian cedents includes the international treaty leaders (Munich Re, Swiss Re, Hannover Re, SCOR, Lloyd's syndicates including Lancashire, Beazley, Catlin, Hiscox, the Berkshire Hathaway specialty group, Everest Re, RenaissanceRe, AXIS Capital, Arch Re, PartnerRe), the regional Asia-Pacific players (Korean Re, China Re, Toa Re, General Insurance Corporation of Singapore, Asia Capital Re), and the domestic Indian reinsurer GIC Re that retains the obligatory cession role under the IRDAI regulations. The 2026 placement increasingly uses GIFT City IFSCA-licensed reinsurance branches and IFSC subsidiaries, supporting onshore Indian access to international capacity with the documentation and settlement efficiencies of the domestic regulatory framework.

The vendor base supporting Indian commercial insurer portfolio analytics has matured through 2022 to 2026 across distinct layers. The implementation choices that an insurer makes across these layers affect the capability ceiling, the operating cost, and the pace of value realisation.

The catastrophe modelling layer is dominated by the global firms. RMS (Moody's), AIR (Verisk), Karen Clark and Company, and JBA Risk Management for flood specialty lead the Indian deployments, with the model licences typically running into single-digit crore rupees annually for top-tier insurers depending on the model coverage and version. The model providers offer Indian peril coverage with varying depth: monsoon flood is well covered by multiple providers, east coast cyclone by all major providers, earthquake by all providers though with debate on the calibration in zones IV and V, and the lesser perils (landslide, hailstorm, drought-linked perils for agriculture and BI) with more variable coverage. Indian specialist firms and academic partnerships provide complementary coverage on specific perils.

The portfolio analytics platform layer has more competition. International platforms including Moody's Analytics RAQ (RMS's portfolio platform), Verisk Touchstone, Aon ImpactOnDemand, Marsh Risk Analytics, WTW Igloo, and Milliman Arius for actuarial analytics are used at Indian insurers with varying depth. Indian providers, actuarial consultancy platforms, and proprietary insurer-built platforms provide alternative or complementary capability at the local end of the market. The platform choice affects the ability to combine catastrophe model outputs with the insurer's specific underwriting data, treaty structure, and operational workflow.

The AI and ML layer for the segmentation, IBNR, and prediction applications uses general-purpose platforms (cloud ML on AWS SageMaker, Azure Machine Learning, Google Vertex AI, Databricks ML), with the model development typically conducted by internal data science teams or by Indian analytics consultancies (Mu Sigma, Tiger Analytics, Fractal Analytics, Tredence, Bridgei2i, LatentView Analytics). The model serving and the integration with the underwriting and reserving workflows occur through the insurer's IT integration layer.

The build versus buy decision plays out differently across the layers. Catastrophe models are universally bought because the underlying scientific work is beyond any individual insurer's capability to replicate. Portfolio analytics platforms are sometimes built and sometimes bought; large insurers with strong technology capability build their own portfolio platforms to maintain control over the integration with internal systems, while smaller insurers find the buy decision more economic. AI and ML applications are typically built with internal teams or with consultancy support because the models embed the insurer's specific operational data and the productive use depends on tight integration with the underwriting and claims workflows.

The implementation economics at a representative Indian top-five insurer covers: catastrophe model licences at INR 3 to 8 crore annually, portfolio analytics platform at INR 2 to 6 crore annually if bought, AI/ML platform infrastructure at INR 1 to 3 crore annually, internal team cost for portfolio analytics function at INR 8 to 25 crore annually (covering 15 to 40 staff including catastrophe modellers, exposure data engineers, ML engineers, analysts, and managers), and consultancy support at INR 3 to 8 crore annually for specialised work. The total annual operating cost runs INR 17 to 50 crore at a top-five insurer, with the variation reflecting depth of capability and breadth of coverage. The investment case rests on improved treaty economics (typically INR 50 to 200 crore of reinsurance economics improvement annually for a top-five insurer), improved underwriting selection (translating to combined ratio improvement of 1 to 4 percentage points on the portfolios where the analytics drives action), and improved capital efficiency through more accurate solvency capital calculation.

Mid-sized insurers face different economics. The catastrophe model licence cost is similar (the providers price more on coverage breadth than insurer size). The portfolio analytics platform is harder to justify on annual operating cost alone, leading many mid-sized insurers to use platform-as-a-service offerings or to share platform capacity with consultancy partners. The internal team is smaller (8 to 18 staff at a mid-sized insurer), with the team often combining the catastrophe modelling, exposure data, and analytics functions in a single integrated unit. The mid-sized insurer's investment case rests primarily on the treaty economics and on regulatory capital efficiency rather than on the broader segmentation and pricing optimisation that the larger insurers prioritise.

The governance architecture for portfolio analytics in Indian commercial insurers has consolidated through 2022 to 2026 around a recognisable pattern that aligns the analytical function with the underwriting, actuarial, treaty, and finance functions.

The board-level governance typically operates through the risk management committee with portfolio analytics as a recurring agenda item. The committee receives the portfolio risk profile reporting, the major scenario stress outputs, the treaty performance against the renewal expectations, and the model documentation and validation outputs. The committee provides direction on risk appetite, approves significant analytical methodology changes, and reviews the IRDAI submissions on risk-based supervision matters.

The executive-level governance typically operates through a portfolio governance committee chaired by the chief risk officer or chief underwriting officer, with the appointed actuary, the head of treaty reinsurance, the chief financial officer, the head of catastrophe modelling and portfolio analytics, and key underwriting line heads as members. The committee meets monthly with deeper quarterly reviews, manages the cross-functional coordination, and escalates material decisions to the board committee.

The operational-level execution sits with the portfolio analytics team accountable for the running of the analytical pipeline, the model performance, and the production of the periodic reporting. The team interfaces with the underwriting function on pricing and selection, with the actuarial function on reserving and IBNR, with the treaty team on cession strategy, and with the IT function on data and infrastructure.

The IRDAI engagement on portfolio analytics has progressed through three streams. The risk-based capital framework consultation through 2023 to 2025 surfaced expectations on portfolio risk measurement that insurers are now operationalising. The enterprise risk management guidelines released by IRDAI through 2024 with subsequent updates establish board and senior management responsibilities for risk identification, measurement, and management at the portfolio level. The stress testing direction in IRDAI circulars expects insurers to conduct portfolio stress tests on a defined cadence with documented methodology and board-level review of outputs. The portfolio analytics function delivers the technical capability for these expectations, with the documentation and review supporting the IRDAI dialogue.

The model validation and governance within the IRDAI framework expects insurers to document their analytical models, to validate them periodically through independent review, to maintain version control with documented change history, and to apply expert judgement to outputs rather than treating them as deterministic. Indian insurers in 2026 increasingly have model validation functions separate from the model development function, with the validation reporting to the chief risk officer or directly to the board risk committee.

The 2026 to 2028 outlook sees several developments that insurers should anticipate. First, the IRDAI Risk Based Capital framework movement from consultation to implementation, with the framework's emphasis on portfolio-level risk measurement reinforcing the investment case for portfolio analytics. Indian insurers preparing for the transition are already running parallel analytical pipelines that support the existing solvency framework and the prospective RBC framework. Second, the integration with climate risk reporting under the SEBI BRSR Core framework and emerging IRDAI climate guidance creating an explicit link between portfolio analytics, climate physical risk assessment, and external sustainability disclosure. Insurers that maintain integrated analytics that support both the underwriting decisions and the climate disclosure capture efficiency that separate pipelines would not.

Third, the AI explainability requirements that are likely to be operationalised through 2026 and 2027 will affect the analytics methodology choices. Black-box ML models in portfolio analytics will face increasing scrutiny on explainability, with the regulatory expectation favouring methods that produce attributable outputs (such as tree-based models with SHAP explanations, generalised additive models, and ensemble methods with transparent component contributions) over deep neural networks where the analytical content has comparable predictive performance. Fourth, the expansion of analytics to mid-sized insurers through platform-as-a-service offerings and through Indian provider specialisation will democratise the capability, with mid-sized insurers in 2027 and 2028 operating portfolio analytics at a depth that previously required top-five resources.