IRDAI Does Not Collect Repudiation Reasons: Scoring a Group Health Panel Without the Settlement Ratio

In July 2026, the government told Parliament something that group-health buyers should read as a procurement instruction. As reported by Insurance Business Asia on 27 July 2026, IRDAI collects data only on overall claim repudiation rates, not the reasons behind individual denials. Insurer-wise data on claims rejected on grounds such as non-medical necessity or exclusion for the preceding three years is simply not available. The regulator can tell you what fraction of claims each insurer refused; it cannot tell you, and does not know, why.

This matters because the health insurance claim settlement ratio is still the metric most corporate group-health RFPs score on. Procurement teams pull the published settlement and repudiation percentages from the IRDAI annual report or insurer disclosures, rank the panel, and treat the exercise as due diligence. The Parliament answer confirms what claims practitioners have long known: the number being ranked is a headline with no underlying detail. Two insurers with identical 92% settlement ratios can behave completely differently at the moment your employee is on a hospital admission desk, and no regulator-published figure will distinguish them.

This post sets out what a buyer should score instead. Not another explainer of what the settlement ratio means, but its successor: the metric set that a procurement team can actually put into a group-health RFP, now that the old anchor metric has been officially confirmed as hollow.

The aggregate picture is worth stating, because it frames the size of the problem the settlement ratio was supposed to solve. Citing the IRDAI Annual Report 2024-25, the same July 2026 reporting records that Indian insurers processed 3.26 crore health claims in the year ending March 2025. Of these, 87% were settled, 8% were repudiated, and about 5% remained pending at year end. The same reporting characterises that as roughly one in every 12 health claims not resulting in payment.

That one-in-12 figure is real money and real disputes, but the aggregate hides everything a buyer needs to know:

• It says nothing about reasons. A repudiation for documented fraud and a repudiation for a contested reading of an exclusion both count as one repudiated claim. The Parliament answer confirms the reason-level split does not exist anywhere in the regulatory dataset.
• It says nothing about timing. A claim settled after eleven months of attrition counts the same as one settled in a week. The 5% pending bucket is where slow insurers park difficulty.
• It says nothing about claim size. An insurer can hold a strong ratio by settling high volumes of small claims quickly while contesting the large ones that actually threaten an employee's finances.
• It mixes retail and group, cashless and reimbursement. A group buyer's experience is dominated by cashless pre-authorisation behaviour, which the final repudiation number barely reflects.

The most important substitute metric is the cashless denial rate at pre-authorisation, because that is where a group health programme is actually experienced. Final repudiation statistics count claims that were formally lodged and formally refused. They do not count the claim that never became a claim: the cashless request denied or whittled down at the pre-authorisation desk, after which the employee pays the hospital, and either files a reimbursement claim into a harder process or absorbs the cost and never files at all. Deterrence at pre-authorisation never appears in a repudiation rate.

The RFP should therefore demand, for the insurer's corporate group-health book over the last two to three years:

1. Pre-authorisation requests received, approved in full, approved with reduction, and denied, as counts and percentages.
2. The share of denied or reduced pre-authorisations that later converted to paid reimbursement claims. A high conversion share means the initial denial was wrong often enough to be a process problem, not a fraud control.
3. Median time to pre-authorisation decision and to final discharge authorisation, since a slow yes functions as a no when a family is standing at a billing counter.

Insurers hold this data; the regulator just does not collect it in this form. An insurer that declines to produce it for a corporate bid is telling you how it will behave when you ask for claims MIS mid-policy. Treat non-disclosure as a scored negative, not a blank.

The second metric is structural: who actually adjudicates and services the claims. An insurer's published numbers blend the performance of whichever administration model sits behind each account, but your programme will be run by one specific desk, either the insurer's in-house health claims team or an appointed third-party administrator.

The two models fail differently. An in-house model concentrates accountability, since there is no interface between the risk carrier and the claims servicer where a file can stall. A TPA model brings scale and specialised cashless infrastructure but adds a party between the insurer and the member, and TPA quality varies widely. What a buyer must score is not the label but the named desk: which TPA or in-house unit will administer this account, its cashless network depth in the postcodes where employees actually live, its pre-authorisation turnaround record, and its escalation structure. The detailed evaluation framework is in our companion piece on choosing a TPA for a corporate health programme.

For RFP purposes, the scoring question is simple: has the insurer committed, in writing, to a named administrator for this account with defined service levels, or is administration an unnamed back office that can change mid-term? A bid that names the desk, states its cashless ratio on comparable corporate accounts, and accepts service-level commitments should outscore a bid with a better settlement ratio and no such commitment.

Third: what policyholders themselves report. IRDAI's Bima Bharosa grievance portal recorded 257,790 complaints in FY2024-25, up about 20% from 215,569 the previous year, with claim-related issues accounting for approximately 69% of grievances in the general and health segments, per the July 2026 reporting. Unlike repudiation reasons, complaint volumes are visible insurer by insurer, which makes grievance load one of the few genuinely comparable public signals of claims behaviour.

The raw count is misleading on its own, because a large insurer will accumulate more complaints simply by writing more business. The usable metric is complaints per unit of premium or per thousand lives covered, computed from the insurer-wise grievance disclosures against the insurer's health premium base. A small insurer with a modest complaint count can be far worse, normalised, than a large one with a bigger absolute number.

Two refinements sharpen it further. First, weight claim-related complaints above servicing complaints, since the RFP is scoring claims conduct and roughly 69% of general and health grievances are already claim-related. Second, look at the trend across two or three years rather than a single year: national complaint volume rose about 20% in FY2024-25, so an insurer whose normalised load is climbing faster than that is deteriorating against the market, and that trajectory will land on your policy year.

Grievances that fail internal escalation reach the Insurance Ombudsman, and the system-level numbers are stark. Citing the Council for Insurance Ombudsmen Annual Report 2024-25, the July 2026 reporting records that the 17 Ombudsman offices received 53,184 complaints in FY2024-25, of which approximately 71% were resolved in favour of the policyholder. Health disputes made up about 64% of all complaints, against 24% for life and 9% for motor.

Read those two figures together. Health insurance generates nearly two-thirds of India's Ombudsman caseload, and across that caseload roughly seven in ten complaints end in the policyholder's favour. The 71% is reported across all lines rather than for health alone, so it is not a health-specific win rate, but health is the bulk of what produces it. That is the closest thing the public record offers to the missing repudiation-reasons data: a systemic signal that a large share of contested denials does not survive independent scrutiny. It also matters for how repudiations are defended; where a denial rests on alleged non-disclosure, the burden of proof sits with the insurer, and the Ombudsman statistics suggest insurers frequently fail to discharge it.

For the RFP, score two things per insurer: its share of Ombudsman complaints relative to its health book, and its record on implementing awards, meaning whether awards against it are honoured promptly rather than delayed or litigated. Ombudsman procedure is itself tightening, including self-contained note and ex-parte timelines, so an insurer's conduct in that forum is increasingly visible. Ask each bidder to state the number of Ombudsman awards against it in health for the last two years and the average time to implementation, and verify against the Council's published data where possible.

The final substitute metric attacks the averaging problem directly. An overall average settlement turnaround is as blendable as the settlement ratio itself: high volumes of small, fast claims can mask systematic slow-walking of large ones. Since the large claims are the ones that matter most to an employee facing a serious hospitalisation, the RFP should require settlement time disclosed by claim size band, for example under INR 50,000, INR 50,000 to 2 lakh, INR 2 lakh to 5 lakh, and above INR 5 lakh, each with median and 90th-percentile days from document completion to payment.

The 90th percentile is the number to weight. Medians describe the routine claim; the 90th percentile describes what happens when the insurer contests, investigates, or simply queues a file. A widening gap between median and 90th percentile in the top size band is the statistical signature of an insurer that pays small claims to protect its ratio while grinding large ones.

This banding also connects to the buyer's regulatory entitlements. Policyholder-protection norms set defined turnaround expectations for claims decisions, and a buyer that has contracted for band-level reporting can actually detect breaches instead of discovering them anecdotally. The entitlement framework is covered in our post on claim TAT entitlements under the policyholder protection regime.

Assembled, the successor metric set replaces one unverifiable percentage with five observable behaviours. A workable weighting for a group-health RFP claims section, out of 100:

1. Cashless denial rate at pre-authorisation, with reduction and reimbursement-conversion data: 25 points. The single best predictor of employee experience, drawn from the insurer's own corporate book.
2. Named claims desk and service-level commitment (TPA or in-house): 20 points. Full points only for a named administrator, disclosed cashless ratio on comparable accounts, and contracted service levels.
3. Settlement time by claim size band, median and 90th percentile: 20 points. Weight the top band's 90th percentile most heavily.
4. Normalised Bima Bharosa grievance load and trend: 15 points. Claim-related complaints per thousand lives or per crore of health premium, across three years.
5. Ombudsman record: 10 points. Complaint share relative to book, awards against the insurer, and time to implement them.
6. Overall settlement and repudiation ratio: 10 points. Demoted, not deleted. It still screens out outliers; it just cannot rank the credible middle.

Two drafting notes. First, put the data demands in the RFP itself and state that non-disclosure scores zero on that line; insurers respond to scored consequences, not requests. Second, convert the winning bidder's disclosures into contract terms, so the pre-authorisation turnaround and band-level settlement reporting you scored become quarterly MIS obligations you can enforce. The regulator has confirmed it will not collect the data that distinguishes good claims behaviour from bad. Until that changes, the buyer's RFP is the only instrument that can.