Industries / Finance and business services / Insurance / Insurance brokerage
Help with claims
When a client has a loss, the broker reports it to insurers, argues the client's case, and can be held liable for any gap in the cover it arranged.
7
Milestones achieved, from 1997 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1997
No duty to advise
In Murphy v. Kuhn, New York's highest court held that brokers "have a common law duty to obtain requested coverage for their clients within a reasonable time", but "no continuing duty to advise, guide or direct a client to obtain additional coverage". The client had never asked for higher car liability limits before a crash in 1991. In Voss v. Netherlands Insurance in 2014 the court listed situations that can create a wider duty, such as "when a broker is paid separately for consultation", while calling them "the exception, not the norm".
LessonIn New York, a client who wants advice on how much cover to buy has to ask for it, or pay for it.
Credit
New York Court of Appeals
2006
One attack or two
No final policies had been issued on the World Trade Center when it was destroyed on 11 September 2001, so "the wording of the binding forms" decided whether the attack was one occurrence or two. A binder is a short temporary contract given while the full policy is drawn up. The WilProp form of Willis, the lead broker, treated a series of occurrences as one loss; the Travelers form did not define the term. Juries in 2004 split along those lines, and in October 2006 the US Court of Appeals for the Second Circuit upheld both verdicts. Settlements in 2007 brought the total recovery to about $4.55 billion, "97.2% of the total maximum", by Business Insurance's count.
LessonThe wording of a temporary binder can decide billions, so settling it before a loss is part of the broker's job.
insurancejournal.comamp.claimsjournal.comclaimsjournal.combusinessinsurance.com
Credit
Willis; the juries; the Second Circuit
2011
Floods at the suppliers
Floods in Thailand in 2011 "inundated seven major industrial estates", with "total insured losses exceeding USD 15 billion". Contingent business interruption cover, which pays when a supplier's or customer's site is damaged, was often written on "a limited perils basis, covering only fire, lightning, and explosion", so many losses were not covered where "brokers and insureds expected policy cover", by the law firm Wotton Kearney's account.
LessonEach extension of cover has to match the risks the client's supply chain actually faces.
Credit
Wotton Kearney's 2025 review
2018
Claims help in the rules
India's Insurance Brokers Regulations of 2018 list among a direct broker's functions "Assisting in the negotiation of the claims" and "Maintaining proper records of claims".
Credit
IRDAI
2022
A broker's wording in court
On 17 October 2022 the English Commercial Court gave judgment in three COVID-19 claims by hospitality groups, Stonegate, Greggs and Various Eateries, all insured "under the Marsh Resilience wording" used by the broker Marsh. The insurers "sought to apply a single aggregate limit of £2.5m"; the judge, Mr Justice Butcher, held that the losses aggregated to "multiple occurrences, specifically each of the government measures".
LessonAs in the World Trade Center case, the words of a broker's form that group losses together decide how much a client recovers.
Credit
Commercial Court
2025
Paid for the lost chance
Norman Hay claimed that its broker, Marsh, had not arranged "worldwide non-owner auto cover" for an employee sent to the US for work who was in a crash in a hired car. The Court of Appeal of England and Wales held that such claims "should be decided on a loss of a chance basis": the client need not prove that an insurer would certainly have paid.
Credit
Court of Appeal of England and Wales
2025
Ombudsmen and brokers
After a flood claim in 2023 was declined, the UK Financial Ombudsman Service (FOS) found that Boon Brokers had answered "No" to a question about earlier claims on the customer's behalf, and ordered it to pay "£45,000 to compensate her for financial loss": "Given that Boon was the professional here, I think it would be fair to attribute more of the responsibility for what went wrong to it." In Australia, the Australian Financial Complaints Authority (AFCA) received 788 complaints about general insurance brokers in 2024-25, up from 447, and its chief ombudsman, David Locke, said "clients are unaware they can complain about a broker in addition to an insurer".
LessonWhen a broker answers an insurer's questions for a client, a wrong answer can cost the claim, and the broker may have to pay.
Credit
FOS; AFCA
Watch for
A claims desk with dashboards
Aon says its Claims Copilot gives its "1,800-strong team of Claims professionals" a shared platform, scores how insurers perform on claims and lets clients track their own claims. It launched first in Germany, and Business Insurance reported in August 2026 that it was "now live in 21 countries". Aon gives no figures on results.
Achieved whenAon's clients or insurers report measured results, such as claim times or amounts recovered, before and after.
Claimed by
Aon, November 2025
Every claim notice automated
FurtherAI, which sells claims intake software, says the broker BCMG is "published as automating 100% of FNOL across its book"; FNOL is the first notice of loss, the first report of a claim. The same page says "Vendor accuracy is measured on vendor test sets."
Achieved whenBCMG or an outside party publishes the volumes and error rates.
Claimed by
FurtherAI, May 2026
Open
01
Ombudsman decisions, tagged
A public dataset of about 600 published FOS decisions about insurance brokers, each tagged by what went wrong (cover not arranged as asked, a wrong answer given for the client, a renewal missed, an insurer's request not passed on, a wording explained wrongly, a claim not pursued), with the outcome and the amount awarded.
The briefCredit · unclaimed02
Claim conditions, checked
A table of the claim conditions in 30 published UK home and small-business policy wordings: how soon a loss must be reported, what proof is needed, and what the policyholder must do to cooperate, each citing its clause. Then 150 made-up losses, for each of which three AI models draft a claim notice and a list of documents to gather; two people grade every draft against the table.
The briefCredit · unclaimed
Open 01 · the brief
Ombudsman decisions, tagged
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public dataset of about 600 published FOS decisions about insurance brokers, each tagged by what went wrong (cover not arranged as asked, a wrong answer given for the client, a renewal missed, an insurer's request not passed on, a wording explained wrongly, a claim not pursued), with the outcome and the amount awarded.
Why it is still open
The FOS publishes its decisions since 1 April 2013 with search filters but no bulk download, and its annual complaints data has no line for insurance brokers. The closest project found, a 2024 brief for University of Edinburgh students set by the broker Simply Business, predicts only "Upheld or Not Upheld" across all products, and says the data is "publicly available but not ready to be used".
financial-ombudsman.org.ukfinancial-ombudsman.org.ukteaching.maths.ed.ac.uk
Built with what exists
the published decision PDFs, downloaded slowly and within the FOS website's terms; a language model for first-pass tags, checked by people; Label Studio for the coding.
First step
Pull the 2024 and 2025 insurance decisions about firms with "Brokers" in their name and code 50 by hand to draft the codebook.
Done looks like
Two people code the same 100 decisions and agree well (Cohen's kappa, a standard measure of agreement, of at least 0.75). The language model matches their tags on at least 90% of 150 held-back decisions. The dataset and codebook are published under an open license.
Open 02 · the brief
Claim conditions, checked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A table of the claim conditions in 30 published UK home and small-business policy wordings: how soon a loss must be reported, what proof is needed, and what the policyholder must do to cooperate, each citing its clause. Then 150 made-up losses, for each of which three AI models draft a claim notice and a list of documents to gather; two people grade every draft against the table.
Why it is still open
Claims tools for brokers publish no public tests. FurtherAI's own comparison of them says "Vendor accuracy is measured on vendor test sets", and Aon gives no figures for Claims Copilot.
Built with what exists
policy wordings that insurers publish on their websites; current AI models such as Claude, Gemini and ChatGPT; Label Studio for grading. The losses are made up, and no notice is sent to an insurer.
First step
Download ten home insurance wordings and table their claim conditions by hand.
Done looks like
The table, with a second person agreeing on at least 97 of 100 entries checked. 450 drafts graded by two people (Cohen's kappa of at least 0.7) on four checks: the right deadline, every required document, nothing that admits fault, and the right recipient. Pass rates are published for each model.