Chapter 9: Risk Assessment: Evaluating Hazards, Controls, and the Probability of Loss

"Inspect what you expect." — A maxim worn smooth in loss-control and underwriting offices for generations; its origin is unknown, but every seasoned underwriter has learned the hard way what it means. The application tells you what the insured believes. The assessment is where you find out what is true — and the gap between the two is where the losses live.

Overview

You have the file in front of you. In the last chapter you ordered the information — the application, five years of loss runs, an inspection, financials, the motor vehicle reports, the statement of values — and you noted what was missing. Now comes the part that separates an underwriter from a clerk who files paper. None of that information underwrites itself. A loss run is not a decision. An inspection report is not a grade. The job of this chapter is the act of assessment: turning a stack of facts into a judgment about how good this risk really is — how often it is likely to produce a loss, how bad that loss is likely to be, and whether the controls in place are enough to make the answer acceptable at a price you can charge.

Assessment is where the underwriter's core mental move from Chapter 6 — exposure, then hazard, then controls — finally gets its hands dirty on a real account. You will take the exposures you identified, find the hazards that drive their frequency and severity, weigh the controls that push back, and arrive at a risk grade: a structured opinion, defensible to your manager and your auditor, of where this risk sits relative to others of its kind. That grade is the input to everything that follows — the math in Chapter 10, the price in Chapter 11, the terms in Chapter 12, the decision in Chapter 13. Get the assessment wrong and every number downstream is built on sand.

We will build the assessment in seven steps. First, a framework — a repeatable way to move from information to grade that keeps you honest and keeps you consistent across a book of business. Then COPE: construction, occupancy, protection, and exposure, the four-letter spine of every property assessment ever written. We will read physical hazards off an inspection the way a practitioner actually does it, then turn to the harder, human hazards — moral and morale — and the signals that betray them. We will separate the loss controls that genuinely change a risk from the ones that only look good on paper. We will turn all of it into a frequency-and-severity estimate. And we will close by writing the two things this whole chapter produces: an overall risk quality grade, and a loss-control recommendation that tells the insured exactly what would have to change to make them a better risk.

In this chapter, you will learn to:

• Apply a risk-assessment framework that moves from submission information to a defensible risk grade without skipping steps.
• Run COPE — construction, occupancy, protection, exposure — on a property risk and read what each letter reveals.
• Translate construction class and fire protection class into a frequency-and-severity expectation.
• Distinguish physical hazard from moral and morale hazard at the assessment stage, and read the signals of each.
• Identify the loss controls that actually change a risk, and tell frequency controls from severity controls.
• Produce an overall risk quality/grade and a written loss-control recommendation.

Learning Paths

This chapter is the engine room of the book. Whatever line you end up in, you will run some version of this assessment a thousand times — so read all of it, but weight it as your path suggests.

🏠 Personal Lines: COPE (§9.2) and the protection-class read (§9.2, §9.3) are exactly how a home is assessed; the moral/morale-hazard signals in §9.4 are the heart of how an experienced personal-lines underwriter reads an application that "smells wrong." Watch how a grade becomes an accept/decline. 🏢 Commercial Lines: This entire chapter is your daily work. COPE on the building, the inspection read, the loss-control negotiation — this is the assessment you will perform on Harbor Steel and every account like it. §9.5 (controls) is where you earn your keep by turning a marginal risk into a writable one. 📊 Analytics: Note §9.6 carefully — the assessment is a structured, human-built frequency × severity estimate, and it is exactly the target a predictive model later tries to reproduce at scale. Understanding what the human grade captures tells you what a model must, and what it cannot. 📜 Certification: COPE, construction classes, protection class, and the hazard taxonomy are core AINS/AU/CPCU property and risk-analysis content, tested directly. The key terms here recur on every exam.

9.1 The risk-assessment framework

Before a single building gets graded, settle a discipline question: what are you actually doing when you assess a risk? The honest answer is that you are forming an estimate — under uncertainty, on incomplete information, against the clock — of two quantities you will never know for certain in advance: how often this risk will produce a loss, and how bad those losses will be. Everything in this chapter is in service of that single estimate. Risk assessment is the process of converting submission information into a structured opinion about the frequency and severity of future loss, and ultimately into a grade you can price and defend.

Why a framework and not just judgment? Because judgment without structure does not survive contact with a book of business. One account assessed brilliantly is worth little if the next forty are assessed by feel, because then your loss ratio is a function of your mood and your file is impossible to audit. A framework does three things at once: it makes sure you look at everything that matters (you do not forget the exposure on the third side of the building because the fire history grabbed your attention); it makes your assessments consistent, so that two similar risks get similar grades and your portfolio means something; and it makes the assessment defensible, so that when the account goes to a loss two years later — or, just as important, when it runs clean and the auditor asks why you charged a debit — the reasoning is on the page. Recall from Chapter 7 that the underwriting file is a legal and professional record; the assessment is the part of that record that shows your work.

Here is the framework this book uses. It is the same move from Chapter 6 — exposure → hazard → controls — extended through to a grade and a recommendation.

THE RISK-ASSESSMENT FRAMEWORK                                    [the practitioner's standard sequence]

  1. EXPOSURE      What, exactly, is at risk? The values, the operations, the people, the perils
                   they face. (You inventoried these in Ch. 6; you confirm them against Ch. 8 info.)
        │
        ▼
  2. HAZARD        What raises the frequency or severity of loss to those exposures? Physical
                   (the roof, the wiring, the hot work), moral, and morale hazards. (§9.3, §9.4)
        │
        ▼
  3. CONTROLS      What is already in place — or could be required — that reduces frequency or
                   severity? Sprinklers, alarms, programs, maintenance, management. (§9.5)
        │
        ▼
  4. FREQUENCY &   Given hazards net of controls, how OFTEN and how BAD? Read the loss runs against
     SEVERITY      the class; separate signal from small-sample noise. (§9.6)
        │
        ▼
  5. RISK GRADE +  An overall quality grade (e.g., above-average / average / below-average / decline)
     RECOMMENDATION and a written loss-control recommendation: what must change to improve it. (§9.7)

Walk down the arrows and notice that each step feeds the next. The exposures determine which hazards even matter (no hot work, no hot-work hazard). The hazards net of controls determine the frequency and severity. The frequency and severity, read against the class and the price you could charge, determine the grade. And the grade, paired with the recommendation, becomes the thing the rest of the underwriting process acts on.

📋 At the Desk The single most common way an assessment goes wrong is anchoring on the loud fact. A \$1.2 million fire on the loss run grabs your eye and the whole assessment becomes about that fire — meanwhile the aging roof in a named-storm zone, which is a far larger potential loss, gets a sentence. Discipline against this is mechanical: work the framework in order, every time, and force yourself to inventory all the exposures before you let any single loss drive the grade. The loss run tells you what has happened; the exposure-and-hazard analysis tells you what could, and the second list is usually longer and more important. An assessment that is only a reaction to the last claim is not an assessment — it is a memory.

One more framing before we get concrete. Assessment is comparative, not absolute. You are not asking "is this risk dangerous?" in some cosmic sense — every business has risk, that is why it is buying insurance. You are asking "is this risk better or worse than the typical risk of its class, and by how much?" A metal fabricator is a more hazardous occupancy than an accounting office; that is not a mark against Harbor Steel, it is simply the class it lives in, and the class rate already reflects it (Chapter 11). Your assessment's job is to place this fabricator within the universe of fabricators — is it a well-run, well-protected example that deserves a credit, or a tired, poorly-maintained one that deserves a debit or a decline? Keep that comparative frame and the grade will mean something. Lose it, and you will either decline every hazardous class (and write nothing) or accept every risk that "isn't on fire today" (and blow up your book).

9.2 COPE: construction, occupancy, protection, exposure (the property core)

If risk assessment has a single most-used tool, it is COPE. The acronym stands for Construction, Occupancy, Protection, and Exposure, and it is the four-part framework underwriters and loss-control professionals have used for more than a century to assess a property risk — chiefly its fire risk, though the same four letters extend naturally to other perils. COPE is to property underwriting what a physical exam is to medicine: a structured, repeatable survey that makes sure you look at the body of the risk in the right order and miss nothing important. Learn it cold, because you will use it on every building you ever underwrite, and it organizes the inspection report, the statement of values, and your own grade.

Take the letters one at a time.

C — Construction. What is the building made of, and how will it behave in a fire (or a windstorm, or an earthquake)? Construction class is the categorization of a building by its structural materials and their fire resistance — it is the single most important physical fact about a property, because it largely determines how a fire spreads and whether the building survives. The industry uses a standard ladder of classes, from most combustible to most fire-resistant. The exact labels vary by rating system, but the ascending logic is universal:

The counterintuitive lesson lives in the third row. New underwriters assume "metal building = won't burn = good risk." Not so. Bare structural steel does not burn, but it loses strength rapidly at fire temperatures and can buckle and collapse — sometimes faster than heavy timber, which chars on the outside and retains its core strength for a while. Non-combustible is not the same as fire-resistive. A metal-frame fabrication plant like Harbor Steel is typically classed somewhere in the joisted-masonry / non-combustible range depending on its exact walls and roof support — better than frame, but a long way from the protected-steel, fire-resistive end of the ladder. Construction tells you, more than anything else, how severe a fire is likely to be once it starts.

O — Occupancy. What happens inside the building? Recall from Chapter 6 that occupancy is the use to which a property is put — and it is the chief driver of how often a loss occurs, because the occupancy brings the ignition sources, the combustible contents, and the hazardous processes. The identical building is a wildly different risk as an accounting office, a restaurant, a furniture warehouse, or a welding shop. Occupancy drives frequency the way construction drives severity. A metal-fabrication occupancy carries welding and cutting (open flame and sparks — "hot work"), combustible dusts and oils, flammable gases, electrical loads, and material-handling equipment. None of that is disqualifying; it is simply the hazard profile of the class, and your job is to assess how well this particular operator manages it.

P — Protection. What will fight the fire if one starts? Protection comes in two flavors, and you must assess both. Public protection is the fire service and water supply available at the location, summarized in the fire protection class — a numerical grade (commonly on a 1-to-10 scale, with 1 the best and 10 effectively unprotected) assigned to a location based on the quality of the responding fire department, the water supply (hydrants, flow, distance), and the emergency communications. A property in a class-2 city with hydrants every few hundred feet is a different severity proposition than the same building in a class-9 rural area with a volunteer department and a tanker. Private protection is what the building itself brings: automatic sprinklers (the single most effective severity control in all of property insurance), fire alarms, standpipes, extinguishers, fire doors, and detection. Protection is overwhelmingly a severity lever — it rarely stops a fire from starting, but it determines whether a fire is a \$50,000 event or a total loss.

E — Exposure. What is next door? External exposure is the risk to the insured property from neighboring properties and hazards — the adjacent occupancies, their construction and contents, and how close they are. A pristine building is a far worse risk if it shares a wall with a fireworks distributor or sits downwind of a refinery, because fire (and explosion, and the smoke and water that follow) does not respect property lines. Exposure also includes the natural environment: in COPE's modern use, the catastrophe exposure — the named-windstorm zone, the flood plain, the wildfire interface, the seismic zone — is read here too. For a coastal risk like Harbor Steel, the E is doing heavy work: the plant's largest single loss potential is not an internal fire at all but the hurricane that the whole region shares.

📋 At the Desk Run COPE in order, and let each letter set up the next. Construction tells you how bad a loss gets; occupancy tells you how often one starts; protection tells you what stops it from becoming total; and exposure tells you what the neighbors and the sky can do to you regardless of how clean your own house is. A strong risk is strong on all four. A typical real risk is mixed — good construction, hazardous occupancy, decent protection, ugly exposure — and the art is weighing the letters against each other and against the controls. Write the four letters down the side of a page and force a finding on each; the empty box is where the missed hazard hides.

⚖️ Compliance Corner COPE is a risk-assessment tool, not a free pass to consider anything you like. Everything in COPE must bear on the physical risk of loss — construction, use, protection, neighboring hazards. It must never become a back door for factors the law forbids you to price on. The characteristics of the building and its operations are fair game; the protected-class characteristics of the people who own or occupy it are not, and a loss-control report that drifts into commentary about a neighborhood's demographics rather than its fire-protection infrastructure is both useless and dangerous. The line between assessing a territory's protection class (legitimate, risk-based) and redlining a neighborhood (illegal) is one the book treats in full in Chapter 35; flag it in your mind every time the E in COPE starts describing a place rather than a peril.

Here is COPE applied to the account on your desk, rendered the way you would summarize it at the top of the assessment.

📄 Read the Submission

text FIGURE 9.1 — "COPE on Harbor Steel" [the Underwriting File] THE SUBMISSION Harbor Steel & Fabrication: a 50,000 sq ft metal-fabrication plant, Port Hadley (Gulf Coast); property limits $20M building / $8M equipment / $10M business income. THE CONTEXT C: metal frame / joisted-masonry, built 1994; original built-up roof (~30 yrs). O: welding, cutting, fabrication — hot work, combustible oils/dust, heavy electrical. P: fire protection class 4; nearest hydrant ~600 ft; ORIGINAL wet-pipe sprinklers. E: coastal, named-windstorm exposed; storm-surge zone nearby; no severe adjacent risk. WHAT IT SHOWS Severity drivers stack up: hot-work occupancy + aging roof + named-storm exposure. Protection exists (class 4, sprinklered) but the sprinklers are 30 years old and unverified. The two fires fit the occupancy's known hazards exactly. WHAT IT DOESN'T COPE alone doesn't tell you whether the sprinklers still WORK, whether the 2023 fire's corrective actions took, or how the operator manages hot work day to day — the inspection, the sprinkler cert, and the loss-control read settle that (§9.3, §9.5). THE DECISION Assessable, not yet gradable: the four letters frame the risk; the grade waits on the inspection read and the loss-control findings later in this chapter. THE LESSON COPE organizes the risk so nothing is missed; it sets up the assessment, it does not finish it. Construction and exposure set the severity; occupancy sets the frequency; protection is the lever you can require the insured to improve.

9.3 Physical hazards and the inspection read

COPE tells you what to look at; the inspection tells you what is actually there. Recall from Chapter 8 that an inspection report (and, for a risk like this, a fuller loss-control survey) is the on-site verification of what the application claims — and that it frequently finds what the application omits. The skill this section teaches is the inspection read: how an experienced underwriter consumes a loss-control report and converts it into a hazard finding rather than just skimming the photos.

Start with the category. Physical hazard — defined in Chapter 6 as a tangible condition of the property or operation that increases the chance or size of loss — is what the inspection is built to find. For a property risk, physical hazards cluster into a handful of recurring themes, and a good loss-control report hits all of them. Read the report looking for each:

• The structure and its condition. Not just the construction class but its state: the age and condition of the roof (the single most-cited property hazard — a roof at the end of its life is a wind and water loss waiting for a storm), the condition of the exterior envelope, signs of deferred maintenance, prior unrepaired damage.
• Electrical. Old, overloaded, or improperly modified wiring is a leading cause of industrial fire. The inspection should note the age of the system, the panel condition, evidence of overload or amateur modification, and — increasingly — whether an infrared (thermographic) scan of the panels has been done to find hot spots before they ignite. (Both of Harbor Steel's known prior fires touched electrical and hot-work causes; the IR scan is precisely the control that addresses one of them.)
• Heat- and ignition-producing processes. This is the occupancy made concrete: the welding and cutting stations, the location of flammable-gas cylinders, the oil and solvent storage, the dust collection. The inspection should describe where hot work happens, how it is controlled, and how housekeeping is maintained around it.
• Housekeeping. This unglamorous word is one of the most predictive in all of loss control. A clean, orderly plant — combustibles stored away from ignition sources, aisles clear, waste removed, no oil-soaked rags piling up — signals a managed operation. A cluttered, greasy, disorganized one signals the opposite, and housekeeping is a leading indicator: it predicts losses that have not happened yet.
• The protection equipment, verified. Does the sprinkler system actually exist, cover the whole building, and carry a current inspection/test tag? Are the extinguishers charged and current? Is the alarm monitored? An "original 1994 sprinkler system" on the application is a claim; whether it would flow water at the right pressure today is a finding the inspection (or a dedicated sprinkler certification) must establish.

📋 At the Desk Read the inspection for what it doesn't say as carefully as for what it does. A loss-control report that is all glossy exterior photos and no comment on the electrical panels, the hot-work controls, or the sprinkler test date is not telling you the risk is clean — it is telling you the inspector did not look, or did not say. Absence of a finding is not a finding of absence. When a report is silent on a hazard you know the occupancy carries, that silence is itself a gap to close, usually with a pointed question back to the broker or a request for a more detailed survey. The best underwriters keep a mental checklist of "the hazards this class always has" and tick each one against the report; the unticked box is the next phone call.

A subtler point: the inspection read is also a read on management, and management is the most important uninspectable hazard. You cannot photograph a safety culture, but you can infer it. A plant with current sprinkler tags, a posted hot-work permit procedure, clean housekeeping, and an electrical system that has been maintained is telling you something about the people running it that no single data point captures: that this is an operation that takes loss prevention seriously. The same building with expired tags, ad-hoc welding wherever the work happens to be, and clutter is telling you the opposite. Two physically identical fabrication plants can be a good risk and a bad risk entirely on the strength of how they are run — and the inspection is your best window into which one you are looking at. We will sharpen this into the moral- and morale-hazard read in the next section, because the management signal sits exactly at the seam between physical and human hazard.

🔍 Check Your Understanding 1. An inspection report on a fabrication plant includes a dozen exterior photos, confirms the sprinkler system "is present," and says nothing about the electrical panels, the hot-work procedures, or the sprinkler test date. Is this a clean risk? What do you do next? 2. Why is "housekeeping" considered a leading indicator of loss rather than a lagging one — and what does a consistently messy plant predict that the loss runs may not yet show?

9.4 Moral and morale hazard: fraud signals and indifference to loss

Physical hazards are in the building; the hardest hazards are in the people. Chapter 6 defined the human hazards — moral hazard, the increased chance of loss from an insured's incentive to cause or exaggerate a loss (Chapter 1 owns the foundational definition), and morale hazard, the increased chance of loss from carelessness or indifference once a risk is insured. At the assessment stage these stop being abstractions and become a question you must actually answer about the account in front of you: is there something about this insured's situation or attitude that makes a loss more likely than the physical risk alone would suggest? This is among the most consequential and least teachable parts of the craft, because the signals are indirect and the stakes — a fraudulent or indifferent insured can torch your loss ratio and your good faith both — are high.

Take moral hazard first, because its sharpest form, the incentive to cause a loss, is what every underwriter quietly screens for. You are not accusing anyone; you are noticing structural situations where a loss would benefit the insured, because those situations correlate with losses whether or not any individual insured acts on them. The classic signals, none conclusive on its own:

• Over-insurance relative to value. A request for limits well above what the property is plausibly worth, or coverage on a struggling asset for more than it could be sold for. A warehouse worth more burned than operating is the textbook moral-hazard setup.
• Financial distress. A business losing money, behind on obligations, or in a declining industry has, at the margin, more to gain from a loss and less to lose from carelessness. The financials you pulled in Chapter 8 are a hazard document, not just a premium-audit input.
• Coverage timing that fits a loss too neatly. A sudden interest in a coverage just before a known exposure materializes, or a pattern of coverage that lapses and revives around losses.
• A loss history that doesn't add up. Repeated similar losses, losses with murky causes, or a pattern that suggests claims are a revenue line rather than accidents. (The full red-flag and fraud-detection toolkit is Chapter 33's; here you are assessing, not investigating.)

Now morale hazard, which is quieter and, on most accounts, more common. No one is trying to cause a loss; they have simply stopped trying to prevent one, because it is the insurer's problem now. The signals are the inverse of the management signals from §9.3: deferred maintenance, expired sprinkler tags, lapsed safety programs, a "we're covered" attitude that surfaces in how the insured or the broker talks about risk. Morale hazard is why a generously-insured, first-dollar risk can quietly deteriorate even when nothing about the building has changed — the attitude changed. And morale hazard is the one you can most directly fix with structure: a deductible (Chapter 12) restores the insured's skin in the game, and a loss-control requirement restores the discipline.

⚠️ Underwriting Trap The trap is treating the moral/morale-hazard read as either paranoia or gullibility — and most underwriters lean one way by temperament. The paranoid underwriter sees fraud in every loss and declines good accounts that had ordinary bad luck, starving the book and burning brokers. The gullible one takes every application at face value, ignores the financial-distress signal because the building looks fine, and writes the account that has every structural reason to produce a loss. The disciplined move is neither: treat the signals as questions to resolve, not verdicts. Over-insurance? Get the valuation. Financial distress? Read the statements and weigh it against the controls and the price. A murky loss? Ask for the full claim detail. You are not the judge; you are the underwriter deciding whether the questions resolve well enough to write the risk at terms that protect the pool.

For Harbor Steel specifically, the moral-hazard screen comes back fairly clean, and it is worth saying why, because "clean" is a finding too. The company is a going concern with roughly \$45M in revenue and a single long-tenured owner; the limits requested are consistent with the values; the losses have identifiable, ordinary causes (electrical, hot work) consistent with the occupancy rather than suspicious ones; and the account is coming to you not because the insured went shopping after a loss but because the prior carrier non-renewed — a market event, not an insured behavior. That last point matters: the reason a risk arrives shapes the moral-hazard read, and "non-renewed for cat exposure" is a very different story from "shopping the market the week after a fire." The morale-hazard read is the live one here: the original sprinklers and original roof raise the question of whether maintenance has been deferred, and that is exactly what the loss-control requirements in §9.5 are designed to answer and fix.

⚖️ Compliance Corner Assessing moral and morale hazard is legitimate and necessary, but it runs along a sensitive edge. You may assess hazard from conduct and circumstance — the financials, the loss pattern, the condition of the risk, the management of the operation. You may not let the assessment become a proxy for a protected characteristic, and you may not deny coverage based on a credit-based insurance score or a financial factor where state law restricts it, or use a consumer report without the adverse-action and disclosure obligations the Fair Credit Reporting Act (FCRA) imposes (Chapter 8 owns the FCRA mechanics). "This insured is in financial distress, which is a moral-hazard signal I weighed against the controls and the price" is a defensible, conduct-based assessment. "I didn't like the feel of the neighborhood" is not an assessment at all — it is bias, and it is unlawful. Keep the assessment tied to risk, and keep the reason on the page.

9.5 Controls and loss prevention: what changes the risk

Here is where assessment becomes underwriting, and where the best underwriters separate themselves from the rest. A risk is not a fixed quantity you either accept or decline. It is a quantity you can change — by requiring the controls that reduce its frequency or its severity before you put your capital behind it. Loss control is the set of measures — engineering, procedures, and management practices — that reduce the frequency or severity of loss. It is the underwriter's most powerful lever, because it lets you turn a risk you would otherwise decline into one you can write profitably, and it does so by genuinely making the insured safer, which serves the social function of insurance (theme six) at the same time as it protects the pool.

The single most important distinction in loss control is the one new underwriters blur: frequency controls versus severity controls. They do different jobs, and a strong risk needs both.

LOSS CONTROL — TWO JOBS                                          [constructed teaching example]

  FREQUENCY CONTROLS                          SEVERITY CONTROLS
  (reduce how OFTEN a loss occurs)            (reduce how BAD a loss is when it occurs)
  ───────────────────────────────            ──────────────────────────────────────────
  • hot-work permit program                   • automatic sprinklers
  • electrical maintenance + IR scans         • fire walls / compartmentation
  • housekeeping discipline                   • fire alarms + monitored detection
  • machine guarding (WC)                      • emergency response / fire brigade
  • driver selection + training (auto)         • business-continuity / backup plans
  • employee safety training                   • redundant utilities, drainage (cat)

  A fire that NEVER STARTS is a frequency win. A fire that starts but is HELD TO $40K instead
  of a total loss is a severity win. The disciplined underwriter requires controls on BOTH axes,
  because you cannot drive frequency to zero — so you must also be sure the inevitable loss is survivable.

Walk the logic. Frequency controls attack the ignition: a hot-work permit program — a formal procedure requiring a permit, a fire watch, cleared combustibles, and a post-work check before any welding or cutting — directly attacks the exact hazard that produced Harbor Steel's 2023 fire. Electrical maintenance and infrared scans attack the hazard behind the 2021 fire. Good housekeeping attacks the fuel that turns a spark into a fire. None of these can drive frequency to zero — humans weld, wires age — but each meaningfully lowers it, and lowering frequency lowers expected loss directly.

Severity controls attack the consequence: sprinklers are the overwhelming example, the single most cost-effective severity control in the history of property insurance, because a working automatic sprinkler system turns the overwhelming majority of structure fires into small, contained, survivable events instead of total losses. Fire walls compartmentalize so a fire in one area does not consume the whole plant. For the catastrophe peril, severity controls look different — storm shutters, roof tie-downs, elevated equipment, drainage — but the logic is identical: you cannot stop the hurricane (frequency is set by geography), so you manage how much it costs you when it comes.

The underwriter's move, then, is to assess the controls that exist, identify the ones that are missing or inadequate, and decide which of the missing ones you will require as a condition of coverage — what Chapter 13 will formalize as a subjectivity. This is the heart of turning a marginal risk into a writable one. You do not have to accept the risk as it walks in the door; you can accept the risk as it will be once the controls you require are in place, and price and structure accordingly.

📋 At the Desk When you require a control, be specific, be verifiable, and tie it to a hazard you actually found. "Improve safety" is not a requirement; "implement a written hot-work permit program with fire watch and 30-minute post-work monitoring, and provide the written procedure before binding" is. The test of a good loss-control requirement is that you could audit it: there is a document, a tag, a scan report, or an installation that either exists or doesn't. Vague exhortations make the insured feel nagged and change nothing; specific, verifiable, hazard-linked requirements change the risk and give you a clean record that the risk you bound is the risk you assessed. And always distinguish a requirement that is a condition of binding (no control, no coverage) from a recommendation the insured should adopt but that won't hold up the quote — the difference is whether the hazard is severe enough to be a deal-breaker.

🤖 Model vs. Judgment A predictive model can see that Harbor Steel had two fires and score the frequency accordingly. What a model trained on historical losses generally cannot see is that the controls which would prevent the next two fires are about to be installed — because the future hot-work program, the IR scan, and the sprinkler recertification are not in the historical data; they are conditions you are attaching to the deal. This is one of the cleanest cases where the underwriter legitimately knows something the algorithm doesn't: the model assesses the risk as it was, while the underwriter assesses the risk as it will be under the terms required. A model-driven decline that ignores the controls you can require is assessing a different, worse risk than the one you are actually proposing to write. (When the model formally scores this account in Chapter 32 and the underwriter overrides it, this is a large part of the documented reason.)

9.6 Estimating frequency and severity for this risk

Now assemble the pieces into the estimate the whole assessment is for. Recall the structure from Chapter 6: expected loss is frequency × severity — how often, times how much. Risk assessment is, at bottom, a structured way to form a view on each of those two numbers for this risk, informed by the class it belongs to and adjusted by what the COPE read, the inspection, the hazard analysis, and the controls have told you. You are not doing actuarial science here — the full mathematics of pure premium, loss ratios, and credibility is Chapter 10's, and the pricing build is Chapter 11's. You are doing the underwriter's version: forming a defensible qualitative-to-semi-quantitative expectation that the math will then sharpen.

Work frequency and severity separately, because different facts drive each:

Frequency — how often. The drivers are mostly occupancy and management. The class sets the baseline: a metal-fabrication occupancy has an inherently higher fire frequency than an office, because of the hot work and ignition sources. Then you adjust for this operator: are the frequency controls (hot-work program, electrical maintenance, housekeeping) strong or weak? Harbor Steel's two fires in five years are, on their face, an elevated frequency for the size of the operation — but here is the assessment subtlety, and it is the one Chapter 10 will make rigorous: two events is a very small sample. Two fires in five years could be a genuinely high-frequency risk, or it could be an ordinary fabricator that had two unlucky-but-explainable events. What tips the read is the cause analysis: both fires trace to known, addressable hazards (electrical and hot work), and the controls that address them are exactly the ones you can require. That turns "elevated frequency" into "elevated frequency that is controllable," which is a materially better risk than "elevated frequency, cause unknown."

Severity — how bad. The drivers are mostly construction, protection, and exposure. For the fire peril, Harbor Steel's severity is shaped by non-combustible/joisted-masonry construction (moderate — won't add fuel, but unprotected steel can buckle), a fire protection class of 4 with sprinklers (a real severity mitigant if the sprinklers work — which is why their certification is a requirement, not an assumption), and the \$10M business-income exposure (a fire that idles the plant is not just a building loss; the lost income while production stops can rival or exceed it — business income is Chapter 19's domain, but it belongs in your severity estimate now). For the catastrophe peril, severity is dominated by the E — the named-windstorm exposure on an aging roof is the single largest severity potential in the whole file, and it is largely outside the insured's control (you manage it with terms and reinsurance, Chapters 12 and 27, not with a permit program).

Here is the structured way to summarize the estimate — a frequency-severity read that becomes the spine of the grade.

FREQUENCY × SEVERITY READ — Harbor Steel (fire peril)            [the Underwriting File]

                    BASELINE (the class)        THIS RISK (adjusted)          NET READ
  FREQUENCY         elevated (hot-work          two explained fires;          ELEVATED but
                    occupancy)                  controls available             CONTROLLABLE
  SEVERITY          moderate-to-high            sprinklered (unverified);     MODERATE-HIGH;
                    (steel buckles; large       $10M BI multiplies a          hinges on sprinkler
                    floor area; BI exposure)    production stoppage            verification
  ─────────────────────────────────────────────────────────────────────────────────────
  CAT (separate)    high severity, low          aging roof in named-storm     LARGE; managed by
                    frequency — geography       zone; outside insured's       TERMS + REINSURANCE,
                                                control                        not loss control

Read across the rows and the assessment writes itself: the fire-peril frequency is elevated but driven by hazards you can require controls against; the fire-peril severity is moderate-to-high and hinges on whether the sprinklers actually work, which is why their certification is non-negotiable; and the catastrophe severity is large and must be managed by structure and reinsurance rather than by anything the insured can do on-site. That is three distinct findings, each pointing at a specific action — and that is what an assessment is supposed to produce.

⚠️ Underwriting Trap The classic small-sample error cuts both ways, and both directions cost money. Over-read the two fires as proof of a hopeless risk and you decline an account that, with controls, would have been one of the steadier files on your book — and you cede it to a competitor who read it better. Under-read them as "just bad luck, every shop has fires" and you write an uncontrolled, genuinely high-frequency risk at a price that won't hold. The disciplined path is the cause analysis: two events is too few to trust as a rate, but the causes are signal even when the count is noise. Two fires from two different, addressable causes tells you about the hazards and the management; it does not, by itself, tell you the frequency. Let the causes drive the controls and let Chapter 10's credibility math drive how much the count moves the price.

9.7 The overall risk grade and the loss-control recommendation

Everything in this chapter has been building toward two deliverables: a risk grade and a loss-control recommendation. These are what the assessment produces — the work product that the math, the pricing, and the decision will act on. Write them down explicitly; an assessment that lives only in your head is not an assessment your file can defend.

The risk quality grade. Risk quality (or risk grade) is the underwriter's overall, comparative opinion of how good a risk is relative to others of its class — typically expressed on a simple ordinal scale. The exact labels vary by carrier (some use 1–10, some use letters, some use "preferred / standard / substandard / decline"), but the meaning is constant: where does this risk sit in the distribution of risks like it, and what does that imply for whether and how to write it? A common four-band scale:

The grade is comparative and conditional. Comparative, because it places the risk within its class, not against the universe. Conditional, because the grade you assign should reflect the risk as you propose to write it — with the controls you will require. Harbor Steel as it walked in — aging roof, original sprinklers, two uncontrolled-looking fires, named-storm exposure — grades below-average. Harbor Steel as you propose to write it — with a hot-work program, an IR scan, sprinkler certification, an ACV-roof endorsement until replacement, and a windstorm deductible — is a managed below-average-to-average risk: a hazardous-class account brought under control, writable at a debit-rated price with the cat exposure ceded. That distinction — the as-is grade versus the as-proposed grade — is the single most useful sentence in the assessment, because it tells the rest of the process exactly what work the terms have to do.

The loss-control recommendation. This is the written list of what must change to improve the risk, sorted into what is required (a condition of coverage — a subjectivity, in Chapter 13's language) and what is recommended (the insured should do it, but it won't hold up the quote). For Harbor Steel, the assessment recommendation reads, in substance:

• Required (conditions precedent to binding): a written hot-work permit program with fire watch; certification that the wet-pipe sprinkler system is tested and functional; an infrared scan of the electrical panels with any hot spots corrected. (Roof replacement within 12 months and telematics on the fleet are also conditions, but they belong to the property-terms and auto assessments in later chapters; here you flag the fire-related controls the risk assessment surfaced.)
• Recommended (risk-improvement, not deal-breakers): formalized housekeeping standards around hot-work areas; a documented preventive-maintenance schedule for the electrical and sprinkler systems; employee safety training records.

📋 At the Desk The grade and the recommendation are one deliverable, not two — the grade is the what (how good is this risk?) and the recommendation is the how (here is exactly what would make it better, and which of those we require). Never hand your manager or the broker a grade without the recommendation attached. "This is a below-average risk" invites a decline or a fight; "this is a below-average risk that becomes an acceptable, debit-rated account if we require these three controls" is an underwriting plan — it shows you have not just judged the risk but figured out how to write it profitably. The first is a verdict; the second is the job.

🔍 Check Your Understanding 1. Why does Harbor Steel grade below-average as-is but acceptable as-proposed? Which single sentence captures the difference, and why is that sentence the most useful one in the assessment? 2. Sort these into "required" versus "recommended" for a hot-work-heavy fabricator, and say why: (a) a written hot-work permit program; (b) repainting the exterior; (c) a current sprinkler test certificate; (d) employee safety-training records.

🗂️ The Underwriting File

The assessment. You now do, on Harbor Steel, the work this chapter taught. You run COPE on the plant and write the four findings at the top of the assessment (Figure 9.1): metal-frame/joisted-masonry construction built in 1994 with an original built-up roof at the end of its life; a hot-work fabrication occupancy that brings welding, cutting, combustible oils, and heavy electrical loads; protection at fire protection class 4 with original wet-pipe sprinklers whose current functionality is unverified and a nearest hydrant about 600 feet away; and an exposure dominated not by any neighbor but by the named-windstorm and storm-surge zone the whole coast shares. The plant is not a highly protected risk.

The loss-control read of the two fires. You read the loss runs the way Chapter 8 taught — for the story, not just the numbers — and the story is legible: the 2021 fire (~\$180K) was electrical; the 2023 fire (~\$1.2M, no injuries) was hot-work/welding. Two fires in five years is an elevated frequency on a small sample, but both trace to known, addressable hazards of the occupancy — and that is the finding that matters. The moral-hazard screen comes back clean (a going concern, a single long-tenured owner, limits consistent with values, ordinary loss causes, and an account that arrived via a carrier non-renewal for cat exposure rather than post-loss shopping). The morale-hazard question is the live one: the original roof and original sprinklers raise the spectre of deferred maintenance, which is exactly what your required controls are built to resolve.

The grade and the recommendation. You grade Harbor Steel below-average as-is and below-average-to- average as you propose to write it — a hazardous-class account that is controllable. The risk-assessment recommendation requires, as conditions of coverage, a written hot-work permit program, a sprinkler certification, and an infrared electrical scan; it recommends formalized housekeeping and a documented preventive-maintenance schedule. Running disposition: graded average/below-average, controllable — the hazards are real, the causes are known, and the controls that address them are exactly the ones you can require. What this assessment does not settle: it does not price the risk (Chapter 10's math and Chapter 11's rate do that), it does not finalize the terms (Chapter 12), and it does not make the accept/decline/ modify call (Chapter 13). It hands the next steps a graded, control-conditioned risk picture — which is precisely what an assessment is supposed to hand them.

Conclusion

Risk assessment is the act of turning information into judgment — the step where a stack of facts becomes a grade you can price and defend. You run a framework in order — exposure, hazard, controls, frequency and severity, grade and recommendation — so that nothing is missed and the next similar risk is assessed the same way. You run COPE on every property: construction tells you how severe a loss gets, occupancy tells you how often one starts, protection tells you what stops it short of total, and exposure tells you what the neighbors and the sky can do regardless. You read the inspection for what it doesn't say as carefully as for what it does, because absence of a finding is not a finding of absence — and because the inspection is your best window into the most important uninspectable hazard, management. You screen for moral and morale hazard as questions to resolve, not verdicts to pronounce. And above all you remember that a risk is not fixed: the controls you require can move it, and the disciplined underwriter's signature move is to write the risk as it will be under the terms required, not the risk that walked in the door.

Two themes ran hardest through this chapter. Underwriting is judgment (theme one): the assessment is the purest expression of it — a structured, defensible opinion formed under uncertainty that a model can inform but not replace, because the model assesses the risk as it was while you assess the risk as it will be. And pricing follows risk (theme four): the grade you produce here is the input that lets the next chapters charge a premium adequate to the risk actually being accepted — neither the inflated price that loses the account nor the soft price whose losses arrive on schedule two years later. Behind both sits the quiet service of the social function (theme six): the controls you require don't just protect the pool, they make the insured genuinely safer.

In the next chapter we put numbers under the judgment. You graded Harbor Steel's two fires as "elevated but controllable" frequency; Chapter 10 asks the rigorous version of the question you have been circling all chapter — is "two fires in five years" a credible signal of this risk's true frequency, or is it small-sample noise that should be credibility-weighted against the class? The assessment told you the causes are signal; the mathematics will tell you how much the count should move the price.

Key Terms

• COPE — the four-part property risk-assessment framework: Construction, Occupancy, Protection, Exposure; the standard structured survey of a property's risk of loss.
• Construction class — the categorization of a building by its structural materials and fire resistance (frame → joisted masonry → non-combustible → masonry non-combustible → modified fire-resistive → fire-resistive); the chief driver of fire severity.
• Occupancy — the use to which a property is put (office, warehouse, restaurant, fabrication, etc.); the chief driver of loss frequency, because it brings the ignition sources and hazardous processes.
• Fire protection class — a numerical grade (commonly 1–10, 1 best) of a location's public fire protection: the responding fire department, the water supply, and emergency communications; chiefly a severity mitigant.
• External exposure — the risk to the insured property from neighboring properties and hazards, and from the natural/catastrophe environment surrounding the location.
• Loss control — the engineering measures, procedures, and management practices that reduce the frequency or severity of loss; the underwriter's lever to change a risk before writing it.
• Risk quality / grade — the underwriter's overall comparative opinion of how good a risk is relative to others of its class, expressed on an ordinal scale and conditioned on the terms proposed.

Spaced Review

1. Run the four letters of COPE and say, for each, the single thing it most tells you about a fire risk. Which letter drives frequency and which drives severity? (§9.2)
2. Harbor Steel's loss run shows two fires in five years. Explain why the causes of those fires are signal even when the count is too small a sample to trust as a frequency — and name the chapter that will make the small-sample point mathematically rigorous. (§9.6; Ch. 10)
3. (From earlier.) Chapter 6 defined hazard as physical, moral, morale, and legal. At the assessment stage, which hazard does an infrared electrical scan address, and which hazard does a deductible address? (§9.4, §9.5; Ch. 6)
4. (From earlier.) Chapter 1 named adverse selection as the enemy of the pool. How does a rigorous risk assessment — grading and pricing each risk on its own hazards and controls — defend the pool against adverse selection? (§9.1; Ch. 1)
5. (The recurring pricing-discipline question.) You grade an account below-average-as-is but acceptable-as- proposed once three controls are required. Would binding it without making those controls conditions of coverage help or hurt the combined ratio, and why? (§9.5, §9.7; Ch. 3)