Two Fire Claims, One Point of Combined Ratio: What Large-Loss Severity Means for Fire Underwriting

In its first-quarter FY27 results reported on 15 July 2026, ICICI Lombard disclosed that two large fire claims had cost it a net Rs 63 crore, an amount it put at roughly one point of the quarter's combined ratio. Two claims. One point. In a book that processes an enormous volume of small motor and health claims every quarter, two fire losses moved the underwriting result by a full percentage point on their own.

That ratio, two events to one point, is the whole argument of fire underwriting compressed into a single disclosure. Fire programmes are not undone by the count of claims. They are undone by the size of the largest one or two. A property account can run a clean loss ratio for years on attritional damage and then surrender several years of margin to a single well-developed fire in a single warehouse or plant.

This post is not about how a fire claim is investigated, and it is not about the rating mechanics of burning cost. It is about severity as an underwriting signal: what a rising single-risk loss does to the assumptions a fire underwriter relies on, why the probable maximum loss estimate is the number that actually matters, and what all of this means for large industrial buyers even while the commercial market stays soft. The specific fires behind ICICI Lombard's disclosure are the insurer's own aggregate figure, with no verified loss-event detail attached, so the interest here is the pattern, not the particular losses.

Actuaries model a line of business as a distribution of frequency (how often losses happen) and severity (how large they are when they do). Different lines sit in very different places on that map, and fire sits at the difficult end.

Motor own-damage and health are high-frequency, low-severity: many claims, each small, each predictable in aggregate because the law of large numbers does the work. You can price them tightly because the average is stable. Commercial fire is the opposite: low-frequency, high-severity. A large industrial account may go years without a serious loss, so the recent history looks reassuring, and then a single fire produces a claim that dwarfs every premium the account has ever paid.

The consequence for the underwriter is that the loss history is a weak guide. A thin run of good years on a large fire risk carries almost no information about the tail, because the events that matter are precisely the ones that have not happened yet in that account's short record. Pricing a jumbo fire risk off its own three or four clean years is close to pricing it off no data at all. What the underwriter is really pricing is the plausible worst case, and the plausible worst case is an estimate, not an observation.

The number that carries the weight in large fire underwriting is not the sum insured. It is the probable maximum loss, the PML: the largest loss the underwriter expects a single event to produce at a given location, taking account of separations, fire walls, sprinkler and hydrant protection, compartmentation and the way a fire would realistically spread before it is controlled.

The distinction matters because the sum insured and the PML are usually very different numbers. A plant insured for Rs 500 crore might have a PML of Rs 120 crore if it is well divided into fire-separated blocks, or Rs 400 crore if it is a single undivided shed where one ignition can take the whole floor. The PML, not the sum insured, is what the underwriter uses to decide how much of the risk to keep, how much to reinsure, and what rate the exposure demands.

A PML is an engineering judgement dressed as a number, and it can be wrong in two directions. Set it too high and the account looks unattractive and gets priced out of the market. Set it too low and the underwriter keeps more of the risk than intended and rates it below its true exposure. Because the PML feeds retention, reinsurance and price all at once, an error in it propagates through the entire structure of how the risk is carried. This is why the credible PML study, backed by a physical survey, is the single most valuable document in a large fire file, and why an underwriter who accepts a broker's optimistic PML without independent support is quietly taking on more than the file admits.

A large-loss quarter is, in effect, the market discovering that some PML assumptions were light. When a fire develops beyond the estimated probable maximum loss, several things go wrong at once, and they all point the same way.

First, the insurer keeps more than it planned to. Retentions and reinsurance are structured around the PML, so a loss that runs past it lands disproportionately on the layers closest to the insurer, including its own net account. Second, the recovery from reinsurance can fall short of what the gross loss implied, because the programme was sized for the loss the PML predicted, not the loss that occurred. Third, and most durable, the event reprices the assumption. Once a site or a class of sites demonstrates that fires there can exceed the modelled maximum, every similar risk on the book carries a higher implied PML, and the rate has to move to match.

This is why a couple of large fire claims are read as a signal rather than as noise. A single point of combined ratio from two events is affordable in any one quarter. What it tells the underwriter is that the severity of the tail may be heavier than the book was priced for, and that the PMLs across a whole class of industrial risks may need revisiting. The claim is a data point about the distribution, and the distribution is what the rate is built on.

No insurer carries a large industrial fire risk alone. The mechanism that lets it write the account at all is reinsurance, and the way it is bought is dictated by the PML.

Up to a point, the insurer's treaty reinsurance absorbs the risk automatically: the account fits inside the capacity the insurer has arranged for the year, and the net retention, the slice the insurer keeps for its own account, is a planned figure. Above that point, on jumbo risks where the PML exceeds treaty capacity, the underwriter has to buy facultative reinsurance, negotiating cover for that specific risk in the market. Facultative capacity is priced by reinsurers who look at the same PML, the same survey and the same protections, so a risk that a reinsurer regards as poorly protected is either expensive to place facultatively or cannot be placed at the terms the insurer quoted the client.

This creates a chain the buyer rarely sees. The rate the insured is offered depends on the net retention the insurer is willing to hold and the facultative terms it can secure, both of which depend on the PML, which depends on the physical risk features and the survey. A large-loss period tightens this chain from the reinsurance end: reinsurers reassess severity, facultative terms firm up on the risks they consider exposed, and the primary insurer's willingness to hold net retention on marginal risks contracts. The buyer feels a hardening that originated two steps upstream, in the reinsurance market's reading of severity, not in their own claims record.

The commercial market in 2026 has been soft, with plentiful capacity and competitive rates on most classes. It is tempting to assume a soft market means large fire risks are cheap across the board, but severity does not respect the cycle in the same way frequency-driven lines do.

A soft market compresses rates fastest on the risks reinsurers regard as well understood and well protected: the clean, divided, sprinklered plant with a credible PML and a good survey history. On those risks, capacity competes and the price falls. It behaves very differently on the risks the severity trend has put a question mark over: the large undivided shed, the high-hazard occupancy, the site with unaddressed survey recommendations. Even in a soft market, capacity for those risks is selective, and a run of large losses makes it more selective still.

The result is a widening gap inside the same soft market. Well-managed large fire risks continue to see keen pricing, because the market is competing for exactly that business. Poorly protected or poorly documented large risks find that the soft market was never really open to them, or closes to them first when severity rises. For a large buyer, the lesson is that the price of a jumbo fire programme is set less by the general market direction than by where the individual risk sits on the severity map, and a soft market is not a substitute for a well-managed site.

The clearest way a severity trend changes underwriting behaviour is in the treatment of risk-improvement recommendations. A fire survey does not just set a PML; it produces a list of recommendations, from housekeeping and hot-work controls to sprinkler upgrades and storage separation, ranked by how much they reduce the exposure.

In a benign period, insurers often accept these recommendations as advisory, noting them and renewing anyway. As single-risk severity rises and reinsurers tighten, the recommendations stop being advisory and become conditions. Capacity is offered subject to compliance within a stated timeframe, priority recommendations become warranties or conditions precedent, and continued non-compliance becomes a reason to reduce the line, load the rate or decline the renewal. The survey report shifts from a document the insurer keeps on file to a document the buyer is expected to act on.

For a large industrial buyer, this reframes risk improvement as a commercial matter, not a safety formality. The recommendations that were easy to defer are the ones that determine, at the next hard turn, whether the market will offer full capacity at a workable rate or a reduced line at a loaded one. The buyers who treat the survey as a live compliance programme keep their access to capacity when severity tightens the market. Those who treat it as paperwork discover the cost of the deferred recommendation at renewal, or at the claim.

Two fire claims and one point of combined ratio is a small headline. Underneath it is the entire logic of fire underwriting: that the line lives and dies on the largest loss, that the PML is where the judgement sits, and that the discipline of severity, protection and risk improvement is what separates a fire book that holds from one that surrenders a year of margin to a single well-developed fire.