Retail submissions, read into the file and cleared for your underwriters to review.

Today an underwriting assistant opens the email, saves each attachment, keys the insured, the retail agency, the line, the effective date and the limits asked into the management system, and runs a clearance search. Then the file goes one of three ways: to an underwriter, back to the retailer as already in from someone else, or back with a list of what is missing.

In the proof of concept, AI reads the email and every attachment and drafts the submission record and the clearance result: each possible match with the reason it matched, the fields read from the applications and the schedule, the questions for the retailer, and a note on appetite or authority. Nothing is entered in your system and nothing goes back to the retailer. Your reviewer, the person on your desk who clears the most submissions, compares each draft with what your team did.

One email can hold a dozen files: ACORD applications, a carrier's own supplemental for a habitational risk or a contractor's subcontracting, a bar's liquor liability questions, loss runs from several prior carriers, photos and a cover note with the target premium. The statement of values is usually the retailer's own spreadsheet, one row per building, with column names that change from agency to agency. Construction arrives in words like brick or steel and has to land in one of ISO's six construction classes, from frame to fire resistive, beside a protection class from 1 to 10 and a total insured value. The draft maps each column and lists anything it could not read rather than guessing.

Clearance is a matching problem, not a lookup. The same insured can arrive as an LLC from one retailer and under its trade name from another, with a mailing address in one email and a building address in the next. A good check compares the legal name, every DBA, the FEIN where given, each location on the schedule and the effective date. Markets settle a clash differently: polled by Celent, insurers named the first submission received, the broker judged best placed to write the risk, or no clearance at all. So the draft records when each email arrived, lists every possible match with the reason, notes any broker of record letter in the file, and leaves the decision to your clearance lead.

The record has to land where your system expects it. Vertafore AIM won't let a user add a new insured or begin a submission until an insured clearance search has run, and its results separate all submissions, unbound submissions, in-force policies and all policies for that insured. A submission entered through AIM's Admin Clearance also creates a suspense record for the assigned underwriter. Documents often go to ImageRight, which stores work as drawers, files, folders and documents; one wholesale broker's published setup keeps a file per submission or policy. The draft lays out each field the way your system holds it, Applied Epic included, and we never type it in.

An MGA has one more test to apply: the authority. A binding authority sets the classes, limits, insured values and territory the MGA may write, and anything outside goes to the carrier as a referral. Lloyd's, for its coverholders, requires each contract to state the regulatory risk location, the insured's domicile and the territorial scope in plain terms, and it does not recognize an incidental risk location. The draft checks every location on the schedule against the territory and every class against the authority, then marks the file inside the authority, for referral, or outside appetite, quoting the clause it relied on.

Your reviewer will look hardest at the costly misses. A false clear, a duplicate the draft failed to find, puts two retailers on one risk, so it weighs more than a false match the clearance lead can dismiss in seconds. Then come the gaps a retailer must fill before anyone can price: an unsigned or undated application, loss runs too old for the market, a missing supplemental, an effective date already past, or a retail agency with no broker agreement on file.

When your reviewer says yes, new submissions run through the same job for about 30 days.

Your clearance desk makes every call. A person confirms each record before it is entered, decides each block, and writes or approves every reply to a retailer. The AI never declines a risk, never tells a retail agent a risk is taken, never quotes and never writes into your system.

Live files are scored against the same written criteria the test used, by the same reviewer.