Key takeaways
- In a brokerage, every AI question resolves to an errors and omissions question. Decide that boundary first.
- The time is in submission preparation, loss run summarisation and correspondence, not in placement judgement.
- Never let an AI reading of a policy form become a coverage statement to a client without a producer verifying it.
- Account managers are the heaviest users and the least trained. Write the guidance for them specifically.
A commercial brokerage runs on document handling. Applications, loss runs, schedules, quotes, binders, policies, endorsements, certificates. An account manager spends much of the week moving information between formats and chasing the pieces that have not arrived. That is precisely the shape of work AI handles well, which makes brokerages one of the better fits in professional services.
It also runs on advice, and advice in this business carries errors and omissions exposure. Getting the value without importing the risk is a matter of drawing one line clearly and then configuring around it.
The line: description versus assertion
Everything a brokerage does with AI falls on one side of a single boundary. On one side is describing, reorganising and summarising material the brokerage already holds. On the other is asserting what a policy covers.
The first side is safe and valuable. A summary of a loss run is checkable against the loss run in under a minute. A submission narrative assembled from the client's application can be read against the application. If it is wrong, someone sees it immediately and fixes it before it goes to a carrier.
The second side is where brokerages get hurt. An AI-generated coverage comparison reads beautifully. It uses the right terminology, structures the exclusions sensibly, and sounds exactly like a producer who has read the form. Sometimes it is describing a form it has inferred rather than the form in front of it. The client cannot tell the difference and neither can a busy account manager at five o'clock on a renewal deadline.
That is the E&O scenario, and it is not hypothetical. It is the ordinary consequence of fluent output meeting time pressure.
The rule to write down and enforce: no statement about what is covered, excluded, sublimited or conditioned reaches a client unless a licensed producer has verified it against the actual policy form. AI can draft the explanation. It cannot be the source of the coverage position.
Where the hours actually are
| Task | Fit | E&O exposure |
|---|---|---|
| Turning a client application into a submission narrative | Strong | Low, source is in hand and carrier reviews it |
| Summarising loss runs into a trend narrative | Strong | Low, figures are checkable in minutes |
| Extracting a schedule of locations or vehicles into a usable format | Strong | Low, verified against source |
| Drafting the missing information chase to a client | Strong | None |
| Carrier correspondence and follow-up drafting | Strong | None |
| Preparing renewal meeting talking points from your own notes | Good | Low if the producer supplies the substance |
| First pass at a claims narrative for advocacy | Good | Low, reviewed before submission |
| Drafting the plain-English cover letter for a proposal | Good | Moderate, do not let it characterise coverage |
| Comparing two policy forms clause by clause | Poor | High, this is the classic failure |
| Answering a client question about whether something is covered | Unsuitable | High, this is advice |
| Recommending limits or carriers | Unsuitable | High, judgement plus licensing |
The value concentrates in the top half of that table, and the top half is where account managers spend their week. A brokerage that captures only the strong rows and never touches the bottom three has taken the large majority of the available benefit with almost none of the risk.
Account managers are the whole rollout
Producers are the visible seniority in a brokerage but account managers do the document work, and the document work is what AI touches. A rollout aimed at producers will look successful in a meeting and change nothing operationally.
What works is task guidance written for the account management function specifically, tied to the workflow they actually follow:
- At submission. Turn the completed application and supporting documents into the narrative and exposure summary carriers expect, in your brokerage's house format.
- At marketing. Draft the follow-up correspondence to underwriters chasing quotes, and summarise what has come back so far.
- At renewal. Assemble the prior year picture from your own files into a briefing for the producer, with the changes flagged for a human to interpret.
- At service. Draft certificate and endorsement request correspondence, and summarise long client email threads into what actually needs doing.
- At claims. Turn the client's account of an incident into a structured first notice, and summarise adjuster correspondence for the producer.
Five tasks tied to five points in a workflow they already know beats any amount of general enthusiasm about the technology.
The measurement that convinces owners: pick one renewal cycle and track the elapsed time from complete application to submission out the door, before and after. Brokerages typically find that number moves more than any per-task time saving would suggest, because the delay was never the typing. It was the queue.
Client data obligations
Brokerages hold non-public personal information, health information in benefits practices, and commercially sensitive material about client operations. State insurance data security laws based on the NAIC model impose written information security programme requirements and third-party service provider oversight in a growing number of states, and carrier agreements often add their own terms.
The practical consequence is the same one that applies to any regulated professional firm. An account manager pasting a loss run into a personal consumer account has introduced a service provider that the brokerage did not select, cannot audit and has not documented. The tool is not the problem. The account is.
Fixing it is a short list:
- One brokerage account on a business tier, with every account manager, producer and service staff member on it.
- Training on your inputs disabled, with the setting recorded as evidence.
- Retention configured to match how you retain other client correspondence.
- Administrator-managed access, so departures and role changes are handled centrally.
- The provider added to your information security programme documentation as a service provider.
- A spending ceiling, so the number is known rather than discovered.
That is a few days of work and it converts an undocumented exposure into a line item you can point at during a carrier audit or a client due diligence request.
What about the carriers
Worth knowing that the regulatory attention in insurance has largely been directed at carriers rather than brokers, through the NAIC model bulletin on the use of AI systems and various state adoptions of it. Brokerages feel this indirectly. Some carrier agreements now include terms about technology used in handling their business, and some larger clients extend their vendor requirements down the chain.
The sensible preparation is the same either way. Know which provider you use, know what settings are applied, know who has access, and have it written down. A brokerage that can answer those four questions in an email has already handled ninety percent of what anyone is likely to ask.
Realistic expectations
What changes is throughput in the service function. Submissions go out faster, renewals are prepared earlier, correspondence stops accumulating, and the account management team spends less of the week on document reformatting. In a brokerage of any size that shows up as capacity, which usually means growth without proportional hiring rather than a reduction in headcount.
What does not change is the placement judgement, the carrier relationships, the coverage expertise or the advice. Those are what the brokerage sells, and they remain human. Anyone promising otherwise is describing a product that would generate an E&O claim within a quarter.
Frequently asked questions
Can we put client submissions and loss runs into an AI tool?
On a brokerage-controlled business account with training on inputs disabled, retention configured and administrator-managed access, yes, and you should record the provider in your information security documentation. On personal consumer accounts, no. Brokerages hold non-public personal information and state insurance data security requirements apply to how it is handled.
What is the main errors and omissions risk?
An AI-generated description of coverage becoming a statement to a client without anyone checking it against the form. The output is fluent and confident and occasionally describes an inferred policy rather than the actual one. Any coverage statement must be verified by a licensed producer against the form before it leaves the brokerage.
Where should a brokerage start?
Submission preparation and loss run summarisation, aimed at account managers rather than producers. Those tasks are high volume, the source documents are already in hand, and any error surfaces within minutes. They also sit entirely on the safe side of the advice boundary.
Will this reduce our service headcount?
Usually not. Most brokerages find it absorbs growth rather than removing people, because the constraint was service capacity rather than demand. Submissions go out sooner, renewals are prepared earlier and the backlog shrinks, which shows up as the ability to write more business with the team you have.
Get your service team moving faster on renewals
We choose the provider, disable training on your inputs, set retention and access properly, cap the spend, and write task guidance for account managers tied to submission, marketing, renewal, service and claims. Your operations manager runs it afterwards. Fixed price, live in 30 days or less.