Industries / Finance and business services / Insurance
Insurance brokerage
Insurance brokers act for people and businesses that want insurance: they work out what cover is needed, find insurers willing to provide it, arrange the policy, and help at renewal and when a claim is made.
1385
A marine insurance contract dated 23 December 1385 covered goods worth 1,200 gold florins on a voyage from Porto Pisano to Palermo, for a premium of 54 gold florins, and names the broker who arranged it, Gherardo d'Ormanno. It survives in the Datini archive in Prato and was studied by an EU-funded research project at Bocconi University.
38
Milestones achieved across 6 domains.
14
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Merchants in 14th-century Tuscany already used brokers to arrange insurance: a contract of 1385 names one. In London, a register of insurance set up by the Crown in January 1575 made brokers jointly liable with the insured for the premium, according to the legal historian Guido Rossi. By 1810 brokers at Lloyd's took each risk first to a lead underwriter and then to others who followed with shares, and they still visit underwriters at their desks, called boxes.
Where it stands
The four largest brokers by 2025 revenue were Marsh ($26.66 billion), Aon ($16.99 billion), Gallagher ($13.17 billion) and WTW ($9.77 billion), by Business Insurance's count, and the large firms keep buying smaller ones: Aon completed its $17 billion purchase of USI on 16 November 2026. In the UK, brokers arrange 94% of commercial insurance, by the count of the British Insurance Brokers' Association (BIBA). Much of the work still runs on email, PDFs and insurers' own websites: in a 2024 survey by IVANS, a firm owned by the broker software maker Applied Systems, agents said a simple small-business quote took about 41 minutes for each insurer.
What is open
Brokers now buy AI tools that fill in forms, compare quotes, place risks and check renewals, but the speed and accuracy figures come from the firms that sell them. The open jobs below build public test sets and tables, such as made-up quotes with planted differences, slips with known answers and ombudsman decisions tagged by what went wrong, so that anyone can check those tools and the way brokers work. Each can be started this month with public documents, open-source code and current AI models.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Finding cover
Finding cover is working out which insurers, Lloyd's syndicates or specialist markets will take a client's risk, and getting the risk in front of them.
6 · 2 · 2
Last achievedAn aggregator turns broker2021
Watch forAppetite found by AI
2
openQuoting
Quoting is gathering a client's details and getting prices and terms back from insurers.
5 · 2 · 2
Last achievedA price without terms2026
Watch forForms that fill themselves
2
openPlacing policies
Placing is getting the chosen insurers to commit to a risk on agreed terms, and recording the contract.
7 · 3 · 2
Last achievedGerman court backs the lead2025
Watch forPlacing in days, not weeks
2
openRenewals
Renewal is the yearly round in which a broker reviews a policy before it ends, takes it back to insurers if needed, and tells the client what has changed in price and cover.
6 · 2 · 2
Last achievedNotices that explain nothing2026
Watch forRenewals without a submission
2
openHelp 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 · 2 · 2
Last achievedOmbudsmen and brokers2025
Watch forA claims desk with dashboards
2
openCompliance
Compliance covers the rules brokers work under: who may act as one, how they must disclose what they are paid, and how they must treat clients.
7 · 3 · 2
Last achievedIndia lifts commission caps2023
Watch forLighter rules for business cover
2
open
Advances to build with
1 research note
Open in insurance brokerage
3 of 12 open deliverables
Open · Finding cover
Stated appetite checked against filings
For one US state and one line of cover, the business owner's policy (BOP) sold to small firms, take each insurer that both publishes an appetite guide and files its underwriting rules with the state. For 100 business classes, such as bakeries or roofers, record whether the insurer's published appetite says it writes the class and whether its filed rules accept it, citing the page of each. Publish the table and the share of cells where the two disagree.
Built with what exists
the QuoteSweep corpus, released under a CC-BY-4.0 license; SERFF Public Access, the NAIC's site through which many state insurance departments, such as Georgia's, let the public view rate and form filings; a document-reading AI model such as Claude or Gemini to pull eligible and ineligible classes from filed rules and cite the page (see the Note "The jobs agents repeat"); a spreadsheet for the hand checks.
Open · Quoting
Quote portals for browser agents
Rebuild the question flows of five insurers' public small-business quote pages as local copies: the same questions, branches and error messages, without their code or branding. Write 100 made-up businesses and score browser agents, AI that operates a web browser, on filling every copy correctly from one submission. Each copy has a marked stop line before any contact or payment page.
Built with what exists
Playwright to build and serve the local copies; Browser Use, an open-source browser agent (see the Note "Deleting 80% to start over"); computer-use models from Anthropic, OpenAI and Google. Local copies mean no made-up data is sent to real insurers.
Open · Placing policies
Slips turned into data, tested
A public test set of 300 made-up MRC slips in mixed formats (Word files, PDFs, scans, slips with endorsements), each with a hand-keyed answer key of CDR fields, and an open-source converter that turns a filled slip into those fields. A public script scores any converter on the set.
Built with what exists
the CDR field list and MRC template published by the London Market Group, subject to their terms of use; document-reading AI models such as Claude and Gemini; Tesseract or docTR for text in scans; JSON Schema for the answer key.
Where it sits
Broader
In outline
