The phrase "real-time risk scoring" appears in a lot of vendor conversations right now, and it covers a wide range of actual capabilities. On one end of the spectrum, you have batch-processing tools that vendors describe as real-time because they return results the same business day. On the other end, you have systems that genuinely score a risk in under a minute from submission. When we use the term at Attune, we mean the second thing: a broker submits a commercial application and receives a scored decision before needing to schedule a follow-up call with the client.
That specificity matters because it changes what you can do operationally. A same-day result and a sub-minute result are not interchangeable. They support entirely different broker behaviors at different stages of the client conversation.
What the scoring engine actually reads
When a Business Owner Policy application comes in, the scoring engine processes the fields a commercial underwriter would open first: the business classification code (SIC or NAICS), gross annual revenue, years in operation, number of employees, prior claims history over the preceding three years, and the primary business location at ZIP code level.
Each of these feeds a different component of the model. Business classification sets the baseline risk band. A licensed plumbing contractor and a bar may both submit standard BOP applications, but the underlying exposure profiles are categorically different. The classification code is the first cut that determines how every subsequent signal is weighted.
Revenue band functions as an exposure proxy. A landscaping operation generating $180,000 annually carries different property, liability, and operational risk characteristics than one generating $1.4 million. Revenue alone does not determine the score, but it calibrates what the other signals mean in context. A prior GL claim on a $200,000 revenue account reads differently than the same claim on a $900,000 revenue account.
Years in operation is a signal that experienced underwriters weight deliberately, and the model replicates that weighting. A business with a five-year track record presents differently than one that opened eighteen months ago. Both can be quoted, but the risk model treats them as distinct risk classes, not just slightly different versions of the same profile.
Prior claims history carries the highest individual weight for standard BOP applications. A single at-fault general liability claim in the past two years moves a score meaningfully in one direction. Two or more claims in three years is a different underwriting tier. The model also distinguishes frequency from severity in how it weights historical claims, which matters because two small property claims often signal different risk characteristics than one large one.
Location scoring draws on ZIP-level data for property hazard categories: flood zone designation, wildfire risk band for applicable states, and catastrophe exposure factors for coastal geographies. These run as modifying signals on the location-dependent components of the premium estimate, not as binary accept-or-decline inputs.
How signals translate into a scored decision
The output of the model is not simply a number. It is a labeled tier with an indicative premium range and a short list of the primary factors that drove the score. The format matters as much as the underlying calculation.
To make this concrete: a plumbing contractor in Newark, New Jersey, with a clean three-year claims history and $420,000 in gross revenue would return a score labeled Moderate, with an indicative annual BOP premium in the $3,200 to $3,800 range. The decision summary would note that the trade classification and local property hazard band are the primary contributing factors, with the clean claims history acting as a positive modifier. A broker can walk a client through that in two minutes without guessing at the logic.
The carrier appetite check runs in parallel with the scoring model. Before the result is returned, the system identifies which carriers are currently positioned to write a risk of that class, location, and revenue band. That step tells the broker which two or three carriers to approach first, rather than submitting to five and waiting to find out which ones will review it at all. Eliminating the speculative submission round is where a significant portion of the time savings comes from.
Why timing is the competitive variable
A broker who receives a scored decision back in fifteen to twenty minutes can give the client a preliminary picture in the same conversation. Not a final bound quote, but a real indicative price from a genuine scoring process based on actual application data. That changes the dynamic of the client interaction in a way that a same-day or next-day result simply cannot.
The alternative is the standard commercial workflow: application submitted, enters the underwriting queue, broker waits two or three business days, follows up, receives the quote, schedules a second call to present it. By that point, a client may have heard from a broker working through a direct channel, or decided that the process is too slow and moved on. Neither outcome is recoverable for the broker who waited.
The conversion dynamic shifts when the broker can discuss coverage scope and a ballpark price while still sitting across from the client. The client experiences a different kind of professionalism: the broker has the tools to answer the question when it is asked, not two days after. This is not a subtle workflow improvement. It is a change in when deals close and how often they do.
What real-time scoring does not mean
We want to be direct about scope here. This technology does not replace underwriter judgment. The scoring model is a risk classification and pricing estimation tool. It identifies a risk tier, generates an indicative premium range, and surfaces which carriers are positioned to write the risk. The final binding decision is made by a licensed underwriter on the carrier side. Attune is tooling that assists brokers in preparing and routing submissions, not an entity that issues, underwrites, or guarantees coverage.
Real-time scoring is also not a guarantee of coverage. Indicative pricing is a starting point for the broker-client conversation, not a firm offer. Carrier appetite changes, individual underwriters exercise discretion on specific accounts, and application details not captured in standard fields can move a final quote outside the indicated range. Brokers who use the tool effectively communicate this distinction to clients at the moment they present preliminary pricing.
The appropriate scope for the model is the standard BOP and general liability applications that represent the large majority of small commercial submissions. Complex risks, unusual business classifications, high-limit policies, and specialty lines warrant a different process. The scoring model identifies when a submission falls outside its training distribution and surfaces that for manual review rather than generating a low-confidence result and presenting it as settled.
Decision rationale as a usable output
A score of 72 out of 100 is not something a broker can explain to a client without further context. What the client needs to hear is why the premium is where it is, not a number. A decision summary that identifies the risk as scoring moderate because the business class carries higher-than-average property exposure in the ZIP code, offset by six years of operation and no prior claims, gives the broker language they can use directly.
This is the piece of the output design that changed most significantly after early-access feedback. The initial version of the scoring output returned a numeric score and an indicative premium range. Brokers in the pilot told us the score itself was the least useful part of what came back. What they needed was the explanation of the factors driving it, presented in terms they could take into a client meeting without translation. The decision rationale format is a direct result of those conversations, not an initial design assumption.
There is a practical implication for how brokers should think about scoring tools: the benchmark question is not "how fast does it score?" It is "how well does the output explain itself?" A fast score that a broker cannot explain to a client is not a complete product. The explanation is the part of the tool that gets used at the moment it matters most.
The data currency question
One constraint worth naming directly: a scoring model is only as current as the appetite signal data it draws on. Carriers update underwriting guidelines, tighten appetite after significant loss events, and adjust pricing parameters across their commercial books. Appetite data that is six months stale will generate incorrect carrier targets even when the scoring model itself is accurate.
This is a data pipeline challenge, not a model accuracy problem. The scoring model is stable once trained on a business class. The appetite signal layer requires continuous updating to reflect what carriers are actually writing in the current market. Maintaining that pipeline is a significant operational component of making the product useful over time, not just at the moment of initial deployment.
For brokers evaluating any quoting or scoring tool, asking about the appetite data refresh cycle is as important as asking about the scoring methodology. A fast, accurate score routed to carriers who are no longer writing that class or have tightened their appetite for that geography is not a useful result. The freshness of the appetite signal is what makes the routing recommendation actionable.
