When a commercial insurance application reaches the Attune scoring engine, the first question is not "is this business risky?" It is: what type of business is this, and what does risk look like for this class? Two applications that appear identical on revenue and years in operation can carry fundamentally different risk profiles once the business classification is resolved. That distinction is where a great deal of underwriting technology falls short, and it is the starting point for understanding how the model actually reads what you submit.
The business class as the load-bearing anchor
The NCCI class code or ISO general liability classification is the primary anchor for everything the scoring model does next. A restaurant at $600,000 in annual revenue and an electrical contractor at the same revenue band are not comparable risks, and treating them as comparable produces scores that look precise but consistently mislead the broker.
The restaurant's exposure concentrates in slip-and-fall incidents, food preparation liability, and liquor-related claims if that coverage is included. The electrical contractor's exposure sits in property damage from faulty installations, completed-operations liability, and bodily injury on client premises. The tail of a large electrical installation claim can extend far longer than the tail of a premises incident in a hospitality setting, which changes how the model weights reserve adequacy against the quoted premium.
Getting classification right before applying any other weighting is not a technical detail. It is the prerequisite for the score having meaning at all. A revenue signal that pushes risk upward in one class might be neutral or mildly favorable in another. Three years in operation is a warning flag for a staffing agency and close to irrelevant for a retail florist. The class defines what the other signals mean.
Revenue as a relative signal, not an absolute one
Revenue appears on every commercial application because it correlates with exposure volume. A higher-revenue food service business typically means more operating hours, more covers served, more transactions per day, and more exposure events per year. But revenue is only informative when measured against what is typical for the class, not against a universal dollar threshold.
At Attune, we treat revenue as a ratio against the class distribution rather than a raw figure. A $1.2 million electrical contractor is a mid-size operation in most Northeast markets. A $1.2 million independent restaurant is at or above the median for owner-operated dining. The absolute number is less informative than where it places the business within its own class peer group.
There is also directional information in revenue trends. A business reporting 40 percent year-over-year growth is not automatically a better risk. In the construction trades, rapid growth sometimes correlates with accelerated hiring, expanded work categories, or geographic expansion into markets that stretch a contractor's established competency. The model accounts for the direction of change, not just the current period value, which is why the application asks about prior-year revenue as well.
Years in operation and what the seasoning effect actually measures
Operating history is a proxy for process maturity, established safety culture, and the kind of accumulated stability that comes from running a business long enough to work out its rough edges. Newer businesses do fail at higher rates than established ones. That is not a harsh position. It is a property of commercial lines loss data across virtually every class.
A business in its first two years has not yet built a meaningful claims history. The scoring model is working primarily from the business profile, the class risk baseline, and whatever background the owner provides about prior experience in the trade. The confidence interval on that decision is inherently wider than for a seasoned operator, and the model reflects that explicitly in the output rather than masking it with false precision.
The maturity signal varies considerably by class. A general contractor in year two is still assembling a workforce and developing site safety processes. A professional services firm in year two may already have an established principal whose prior-firm experience transfers directly. A decade of electrical contracting experience carries across a new business formation in a way that a first-time restaurant opening with no hospitality background does not. The model reads years in operation in the context of the class, not in isolation.
Attune applies wider decision confidence intervals on early-stage businesses and reflects that uncertainty in the decision rationale block. A broker seeing a qualification note on a newer business is seeing the model's honest read of what the available application data can support, not a failure to return a clear answer. Honest uncertainty is more useful to a broker than a falsely narrow band.
Prior claims: frequency versus severity
A single large claim tells a different story than three small claims at the same total incurred value. In general liability and BOP underwriting, frequency is often the more predictive signal. A business with three claims in four years is more likely to have systemic process or safety gaps than a business with one large claim from an unusual event.
Severity matters more in specific coverage categories. In commercial auto, a single high-severity accident may be a one-time event with limited predictive value for the rest of the book. In professional liability, a large claim may indicate a structural quality-control problem that raises the forward-looking risk significantly and warrants a different underwriting response entirely.
The Attune scoring model weights claims differently depending on which coverage type is being scored. Frequency patterns on general liability receive heavier weighting than isolated severity events at the same dollar level. This is not a novel actuarial position, but it matters when you are processing applications in real time and the weighting logic needs to be consistent and auditable across every submission the engine handles, not reconstructed by hand each time.
Location as a modifier, not a primary driver
Location risk is real. A commercial property in a coastal flood zone carries different catastrophe exposure than the same property 40 miles inland. A general liability risk in a state with elevated average jury verdicts for premises liability claims should carry different pricing parameters than a comparable risk in a more moderate legal environment.
The Attune model treats location as a modifier applied on top of the class-level risk baseline, not as an independent scoring dimension that overrides the application-level signals. A well-run restaurant in a higher-risk jurisdiction still scores as a well-run restaurant. The location modifier adjusts the indicative premium band and the carrier appetite flags. It does not change the binary assessment of whether the risk is quotable at all.
What location is not, in our model, is a shortcut for avoiding a close read of the application. A strong risk in a difficult market is still a strong risk. The broker's job is to find the right carrier for that combination. The scoring model's job is to flag the location context clearly so the broker can act on it, rather than discovering a jurisdictional issue after a submission has already been declined and the relationship cost has been paid.
What the score does not see
The score Attune returns is a structured interpretation of the application as submitted. It does not know what the application does not say. An ownership change not reflected in the claims history is invisible to the model. A new work category the business began recently, not yet generating claims, will not appear in the signals. A key employee departure affecting site supervision is not captured in any standard application field.
This is not a limitation unique to algorithmic scoring. A manual underwriter reviewing the same application has the same blind spots. The genuine advantage of a structured model is consistency: the same signals receive the same interpretation across every application, without variation from underwriter workload, broker familiarity, or the sequence in which submissions arrive on a given afternoon.
The broker's role is to supply the context that application fields cannot capture. Attune is built as a tool for brokers, not a replacement for broker judgment. The model handles pattern recognition on quantifiable signals. The broker handles the qualitative context, the client relationship, and the final submission decision. We are not building toward making those functions interchangeable. We are building toward making the quantifiable part faster and more consistent so the broker has more time for the part that actually requires human judgment.
