Nine Hours to Build One 1/1 Submission Packet. In 2026 You Assign the Outcome, Not the Task List.

Nine hours. That is what a full commercial submission packet costs an account manager at a mid-size independent agency, and eight of those hours are not underwriting judgment. They are gathering.

Count the pieces on a real 1/1 renewal for a 40-employee electrical contractor. ACORD 125 and 126 rebuilt from the expiring policy. ACORD 130 with class codes and payroll by state. ACORD 140 with the location schedule and construction, occupancy, protection and exposure detail. Five years of loss runs, ordered from three different carrier portals, each with its own password reset. The current experience modification worksheet. A schedule of 22 vehicles with VINs and radius of operation. A driver list with MVR dates. Current dec pages. A one-page narrative on the two open claims. And a cover email to four appointed markets that says why this account is worth an underwriter's Tuesday.

None of that requires a license. All of it has to be right.

Why the 2024 version of this stalled out

Your agency tried this. Someone on staff opened a chat box, pasted in a loss run, got a decent summary, then had to paste in the next document, then had to copy the answer back into Epic by hand. It worked, technically, and it saved maybe twenty minutes on a good day. Nobody kept doing it after the first week of December, because the copying and pasting was the job.

The problem was never the quality of the writing. It was that the thing lived in a box on a screen and could not reach into the account folder, the carrier portal or the agency management system. Every step needed a human to carry a document across the gap.

What Claude Cowork and ChatGPT Work do differently

On 12 January 2026, Anthropic released Claude Cowork. You give it a goal, it connects to your files and applications, breaks the job into steps, works on its own, and hands back finished work — a completed spreadsheet, a document set, a packet. It expanded to mobile and web in July. On 9 July 2026, OpenAI released ChatGPT Work on GPT-5.6, doing the same shape of thing: a goal in, hours of unattended work, a finished artifact out.

Both were aimed at people who do not write software. That matters more than any feature list. The person who assigns the goal is your commercial account manager, not your IT provider.

Assigning a goal means telling the software what finished looks like — "a complete 1/1 submission packet for the Herrera Electric account, ready for my review Thursday morning" — instead of walking it through one document at a time and carrying the output between systems yourself.

flowchart TD
  A["Producer sets the goal: 1/1 packet for Herrera Electric by Thursday"] --> B["Opens the expiring file in the agency management system"]
  B --> C["Pulls five years of loss runs from carrier portals"]
  B --> D["Rebuilds vehicle and driver schedules from the fleet list"]
  B --> E["Fills ACORD 125, 126, 130 and 140"]
  B --> F["Drafts the narrative on the two open claims"]
  C --> G["Assembles the submission packet"]
  D --> G
  E --> G
  F --> G
  G --> H["Account manager reviews, corrects and signs off"]
  H --> I["Emailed to four appointed markets"]

Tuesday, October 14, in a real agency

8:05am. The producer on the Herrera account writes four sentences: build the 1/1 submission packet, here is the expiring file, use the fleet list the client emailed on the 9th, markets are the four we are appointed with for electrical, have it back Wednesday afternoon.

By 11:00 the loss runs are pulled and read, the claims are summarised by policy year with reserves and open status, and the ACORD 130 payroll grid is filled and compared against last year with the two class codes that shifted flagged. The fleet list — a spreadsheet with inconsistent VIN formatting and three trucks listed by nickname — is reconciled against the expiring auto schedule, and the four vehicles it cannot match are listed as questions, not guesses.

Wednesday 1:30pm. The account manager sits down with a packet, not a pile. She spends 45 minutes on the four vehicle questions, rewrites the claims narrative because she knows the client fired the foreman involved in the 2024 injury and the underwriter needs to hear that, checks the additional insured forms against the two general contractor subcontracts in the file, and adds the schedule of scheduled equipment the packet under-listed. Then it goes out.

The difference is not that the machine is smarter than she is. It is that she spent her 45 minutes on the parts that require her license and her memory of the account.

What the owner has to change, which is the hard part

The technology is the easy half. The half that fails is management, and it fails in a specific way: agencies keep assigning tasks after they have bought something that accepts goals.

Three changes, and they are all yours to make.

Define what finished means, in writing, once. Your agency already has an unwritten standard for a submission packet — the one the good account manager follows and the new hire doesn't. Write it down: which ACORDs, how many years of loss history, what the narrative covers, what gets attached. That document is now the instruction, and it is also the training material for the next hire. Most agencies discover the standard was never actually agreed on.

Move the review gate to the end, and make it real. The account manager stops being the person who assembles and becomes the person who signs. That is a different job and it needs a checklist: forms against contract requirements, loss run totals against the summary, class codes against the operation as it actually is today. Put her initials on it. Your E&O file needs a human name on every submission and that does not change.

Stop measuring the desk by hours in the file. If your account managers are evaluated on activity volume, they will keep doing the gathering, because gathering is visible. Measure accounts shopped, submissions out by the deadline, and hit ratio.

The arithmetic: this is a capacity number, not a payroll saving

Nobody gets laid off. What changes is how many accounts actually see a market instead of getting rolled over with the incumbent because there was no time.

Two caveats before you take that to a staff meeting. Shopping an account you would otherwise have rolled sometimes costs you the account, so the retention side has to be watched, not assumed. And service work will absorb some of the freed hours whether you plan for it or not — certificates and endorsement requests do not pause for renewal season.

Where this still needs a licensed human, without exception

Market selection stays with the producer. The software will happily assemble a submission for a market you are not appointed with, or for a carrier that dropped that class two months ago after a bad quarter — it does not know your appointments or the underwriter who told you at the state association meeting that they are off contractors until spring.

Coverage recommendation stays human. A packet built from the expiring policy reproduces the expiring policy, gaps included. If the client bought two new locations in a coastal county and the wind deductible needs to change, no amount of document assembly surfaces that. Only somebody who talked to the client does.

And the underwriter relationship is the whole business. The reason your submission gets looked at Tuesday instead of the following Monday is that your producer called. That call is not a step you can assign.

Frequently asked questions

Will it work with Applied Epic, EZLynx or HawkSoft?

It depends on how your system is reached and what your vendor permits, and that is a real conversation to have with them before you buy anything. Where agencies start successfully is the file-and-portal side of the job — documents in the account folder, loss runs from carrier portals, spreadsheets from the client — with the finished packet handed back for a person to file. That covers most of the nine hours without asking your agency management vendor for anything new.

How do I know it did not invent a number?

You check the ones that matter, every time, and you make that check a named step. Loss run totals, payroll by class code, limits and deductibles on the expiring dec page, and vehicle count. Four checks, five minutes. If those four tie out and the flagged questions are answered, the rest of the packet is formatting.

What is a realistic first job to hand over?

Not the largest account. Take a mid-size renewal you have already completed this year, assign the goal, and compare the packet against the one that actually went out. You will learn more from the differences in one afternoon than from a month of demos.

Meanwhile, the phone

The catch with unattended work is that it runs while the front of the agency stays exactly as loud as it was. During 1/1 season the service line does not slow down, and an account manager deep in a review gate is the worst person to interrupt. CallSphere builds AI voice and chat agents that answer the agency's phone and website chat, take certificate and change requests, capture new business inquiries, and book the callback with the right producer — so the hours you just freed on renewal packets do not get eaten by the queue.