Industries / Finance and business services / Insurance / InsurTech
Data for underwriting
The outside data and software insurers buy to decide whom to cover and at what price, from shared claims records and prescription histories to aerial images of roofs.
7
Milestones achieved, from 1994 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1994
A shared claims history
ChoicePoint built the Comprehensive Loss Underwriting Exchange (CLUE) at insurers' request, which a 2004 Insurance Journal article dated to "around 10 years ago". Insurers that subscribe report their claims to it and can see up to five years of an applicant's claims on home and car policies, according to the Utah Insurance Department. ChoicePoint's sale to Reed Elsevier was completed on 19 September 2008, and CLUE is now part of LexisNexis. In 2004 Texas lawmakers objected that customers who only asked about their cover were being recorded. State Representative Craig Eiland said: "Calling for an inquiry is not the same as consumer seeking benefits under the policy."
LessonWhat a shared database counts as a claim can decide who is offered cover.
Credit
ChoicePoint, now LexisNexis
2012
Prescriptions as evidence
Milliman IntelliScript, which says it has supplied insurers since 2001, sells life insurers the histories of the prescriptions an applicant has filled. In a 2012 study the reinsurer RGA found that some prescription information was known for over 70% of applicants. Among applicants with a prescription record, those prescribed high-risk drugs died at 163.5% of the expected rate, against 85.1% for the rest. RGA wrote that electronic histories made such information available "in a more structured, verifiable, and cost-effective manner".
Credit
Milliman IntelliScript; RGA (Scott Rushing and Timothy L. Rozar)
2015
Life cover online, then closed
MassMutual launched Haven Life in May 2015, saying it was "the first solution for buying quality term life insurance entirely online". An algorithm gave an instant decision, though a medical exam was still required within 90 days. In November 2023 MassMutual decided to wind it down, citing "lack of consumer adoption" and "high costs associated with customer acquisition". It stopped taking applications on 3 January 2024.
LessonA faster decision did not lower the cost of finding customers.
Credit
MassMutual; Yaron Ben-Zvi, co-founder and chief executive
2021
Roof measurements in court
EagleView sells roof and wall measurements taken from aerial images to insurers and builders. In 2019 a US jury awarded it $125 million against Verisk and its unit Xactware for infringing its patents, and an order stopped sales of their Roof InSight and Geomni Roof products. In 2021 Judge Renée Marie Bumb tripled the damages to $375 million. The companies settled all their litigation in November 2021.
LessonMethods for measuring roofs from images could be patented, and a patent could stop a rival's product.
Credit
EagleView
2022
Exams waived by data
Accelerated underwriting lets a life insurer skip the medical exam for some applicants and check data such as prescription histories instead. A Society of Actuaries survey of 24 companies, by Allen Klein and Justin Li of Milliman, found that the average share of applications eligible rose from 51% in 2019 to 59% in 2022. The average largest policy offered this way rose from $800,000 to $2,100,000. Random holdouts, in which some eligible applicants are still given the full exam to check the shortcut, were run on average 4% of the time in 2022, down from 8% in 2019.
LessonAs the programs grew, the checks on them shrank.
Credit
Society of Actuaries Research Institute; Allen Klein and Justin Li
2023
Rules for outside data
In January 2019 New York's Department of Financial Services told life insurers that an insurer "may not simply rely on a vendor's claim of non-discrimination" when it uses outside data. In July 2024 it issued a circular letter on AI in underwriting and pricing that requires testing for proxy discrimination, in which a factor stands in for race or another protected trait. The NAIC, the association of US state insurance regulators, adopted a model bulletin on insurers' use of AI on 4 December 2023. Its map of 13 November 2026 lists 28 states, the District of Columbia and Puerto Rico as having adopted it.
Credit
New York Department of Financial Services; NAIC
2024
Photos are not inspections
On 29 May 2024 the Pennsylvania Insurance Department reminded insurers of "their obligation to conduct a physical inspection" to confirm roof damage seen in aerial images before acting on it, for example by not renewing a home policy. It said the images it had seen often showed only "discoloration, streaking, or other cosmetic issues". Commissioner Michael Humphreys said some images "barely identify the structure of the home", and that the department "does not seek to broadly restrict the use of aerial imagery".
LessonA stain seen from the air is not proof that a roof is failing.
Credit
Pennsylvania Insurance Department
Watch for
Roof age from the air
Nearmap says its Roof Age Gen2 dates a roof's replacement from aerial images with "95% accuracy within one year of the roof replacement date". It gives no method or test set.
Achieved whenan outside party publishes a test against public building-permit dates.
Claimed by
Nearmap, October 2025
Life cases read by AI
Sixfold launched an AI underwriting assistant for life and health insurers. Peter Tilocca of ClearView, an insurer using it, said: "We've seen a 55% reduction in time to evaluate a case."
Achieved whenan outside party measures the time saved and the error rate against human underwriters on a stated number of cases.
Claimed by
Sixfold and ClearView, September 2026
Open
01
A roof-age test set
A public set of at least 2,000 homes in Austin, Texas, each with the date of its last re-roofing permit from the city's open data, paired with public aerial photos from before and after that date. Anyone can then check a roof-age score against it: how often the score is within one year of the permit date.
The briefCredit · unclaimed02
Prescription reports checked
A study, run with volunteers' consent, that measures how often the prescription-history reports sold to life insurers list drugs a person never took, or miss drugs they did take. Each volunteer requests their own report and compares it line by line with their pharmacy's records. Only totals are published.
The briefCredit · unclaimed
Open 01 · the brief
A roof-age test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of at least 2,000 homes in Austin, Texas, each with the date of its last re-roofing permit from the city's open data, paired with public aerial photos from before and after that date. Anyone can then check a roof-age score against it: how often the score is within one year of the permit date.
Why it is still open
Vendors such as Nearmap publish accuracy figures without a test set. The largest public roof dataset found, ZRG, covers "over 20k properties" for roof shape, not age.
Built with what exists
Austin's building permit records through the city's open data API, which list "Reroof" permits with issue dates and coordinates; photos from the US National Agriculture Imagery Program (NAIP) on AWS, public domain, at 30 to 100 cm a pixel, updated every two to three years; Python with rasterio to cut out each roof.
First step
Pull one year of Austin re-roofing permits from the open data API and match the first 50 addresses to NAIP images.
Done looks like
2,000 re-roofed homes and 2,000 with no roof permit in the same years, each with images and a link to its permit. A simple baseline model is scored and its share within one year published. A second person checks 100 homes by eye and agrees with the labels for at least 95.
Open 02 · the brief
Prescription reports checked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A study, run with volunteers' consent, that measures how often the prescription-history reports sold to life insurers list drugs a person never took, or miss drugs they did take. Each volunteer requests their own report and compares it line by line with their pharmacy's records. Only totals are published.
Why it is still open
In 2026 Milliman agreed to a proposed $15.5 million settlement of a lawsuit under the US Fair Credit Reporting Act, covering an estimated 780,965 consumers, over allegedly inaccurate reports. The lead plaintiff's 2020 report "included 13 medications Healy said he had never taken". No published measure of how often such reports are wrong was found.
Built with what exists
the copy of their own file that people can request from a consumer reporting agency under the Fair Credit Reporting Act; a simple form on which each volunteer marks every line as matched, not theirs or missing; an encrypted spreadsheet kept offline; Python for the totals.
First step
Request your own prescription-history report and pharmacy record, and mark it up as the first test case.
Done looks like
100 volunteers, each report marked by the volunteer and checked by a second volunteer against the pharmacy record. The share of reports with at least one wrong or missing drug is published with a confidence range, and no personal data are published.