Insurance underwriters

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Blackboard ledger + physical inspection1700–1850

  Blackboard ledger + physical inspection (Lloyd's market and early American fire underwriting era)

  The first insurance underwriters had no tools beyond pen, paper, and personal knowledge. At Lloyd's Coffee House the risks were literally written on blackboards: cargo manifest, destination port, route, season, captain's reputation. The underwriter assessed the risk entirely from personal experience of maritime trade, weather patterns, and the trustworthiness of the people involved. In American fire insurance, early underwriters would walk the premises being insured, inspecting building construction, proximity to neighbors, and the availability of water. The Friendly Society's "Hand in Hand" mark -- a plaque placed on buildings it insured -- was both a marketing device and a practical tool: it told fire brigades (which were often insurance-company brigades) which buildings to prioritize saving.

• Actuarial tables + mortality schedules…1850–1940

  Actuarial tables + mortality schedules + medical examination (life insurance scientific era)

  The mid-19th century brought the first scientific tools to underwriting. Edmund Halley's 1693 survival table was the conceptual foundation, but it was James Dodson's 1755 demonstration that age-based premiums could make life insurance viable for any applicant that transformed the practice. The Equitable Life Assurance Society, founded in London in 1762, was the first insurer to systematically use mortality tables to price policies. By the mid-1800s, American life insurers were employing company physicians to examine all applicants -- a physician's letter of assessment was required for any policy above a minimal amount. The underwriter worked from two tools: the actuarial table (which set the premium range for a given age and health class) and the physician's report (which placed the applicant into a class). Fire underwriting developed parallel rate-manual tools: large rate manuals produced by organizations such as the National Board of Fire Underwriters (1866) specified the premium for any building type, construction quality, and occupancy class.

  Effect on the work

  Actuarial tables expanded the underwriting market enormously: any literate clerk could look up a rate once the tables were established, reducing the cognitive barrier to entry. This drove the creation of large insurance company workforces where underwriting clerks applied standardized rules to a high volume of applications rather than exercising individual judgment on each one.

• IBM punched-card tabulation + paper rat…1940–1970

  IBM punched-card tabulation + paper rating manuals (clerical processing era)

  Metropolitan Life Insurance was among the first carriers to adopt IBM punched-card tabulating equipment -- renting IBM machines for actuarial and policy-record applications. By the 1940s, large insurers ran their policy records on tabulating machines: premium billing, lapse tracking, and mortality statistics were among the first functions mechanized. For the underwriter, the key shift was organizational: the volume of applications processable by a carrier exploded, requiring large clerical underwriting departments to handle intake, rating, and policy issuance. Travelers Insurance installed one of the first IBM mainframes in the insurance industry in the early 1960s, automating batch rating for personal auto and homeowners policies. Underwriters worked from thick paper rate manuals and applied the tables manually, with tabulating machines handling the record-keeping and billing rather than the underwriting decision itself.

  Effect on the work

  IBM tabulation and early mainframe computing expanded the insurance market rather than contracting the underwriting workforce. More policies could be processed per carrier, driving employment growth rather than displacement. The workforce grew from roughly 35,000 in the 1920s to an estimated 55,000 by the early 1940s and substantially more by the 1960s.

• Mainframe rating systems + personal com…1970–1995

  Mainframe rating systems + personal computers + early automation (digital rating era)

  The 1970s and 1980s brought mainframe-based rating systems that automated the mechanical application of rate tables: a clerk could enter a vehicle year, make, model, driver age, and ZIP code and receive a computed premium in seconds, replacing the manual table lookup that had dominated for decades. Applied Systems, founded in the early 1980s, was one of the first companies to build dedicated software for insurance agency management and rating -- their tools spread from large carriers down to independent agents by the late 1980s. For life underwriting, the 1980s AIDS crisis drove a significant expansion in medical testing requirements, adding new data inputs to the underwriting process. Preferred-rate programs emerged as carriers recognized that healthy applicants deserved lower premiums, requiring underwriters to segment risk more finely. Personal computers arrived in insurance company underwriting departments in the mid-1980s, beginning the shift from batch mainframe processing to interactive desktop rating.

  Effect on the work

  Mainframe rating automation did not significantly reduce underwriting headcount in this era -- instead, it expanded the volume of business each underwriter could process, supporting premium growth without proportional headcount growth.

• Rules-based expert systems + early auto…1995–2010

  Rules-based expert systems + early automated underwriting (decision-engine era)

  The mid-1990s saw the first wave of technology that genuinely displaced underwriting labor rather than amplifying it. Rules-based expert systems -- software encoding underwriting guidelines as if-then decision trees -- began handling personal auto and homeowners applications automatically. If an applicant met the carrier's filed guidelines, the system issued the policy without human review. If it fell outside, it was routed to a human. The Y2K crisis of the late 1990s had an ironic effect: carriers faced with replacing legacy systems largely chose to patch rather than rebuild, delaying the modernization of underwriting platforms but also leaving early expert systems in place longer than intended. By 2000, carriers processing personal lines at scale reported that a significant share of clean applications were being issued without human touch -- the first true straight-through processing.

  Effect on the work

  This era produced the first measurable underwriting headcount decline. From roughly 162,000 in 1990, insurance underwriter employment had fallen substantially by 2000 and drifted lower into the 2010s. Each new generation of automated underwriting software extended the share of submissions that could be decided without human judgment -- concentrating remaining human underwriting on the complex and non-standard risks that rule engines could not handle.

• Straight-through processing + predictiv…2010–2022

  Straight-through processing + predictive analytics + insurtech platforms (STP era)

  The 2010s delivered the second, larger wave of underwriting automation. Cloud computing, mobile channels, and modern APIs made it possible to build insurance products that could originate, rate, and bind entirely without human involvement. Guidewire and Duck Creek, founded in the early 2000s, became the dominant policy-administration platforms for mid-size carriers; they incorporated automated underwriting engines as first-class features. Cape Analytics (founded 2014) used aerial and satellite imagery to score property risks at the address level, replacing the physical inspection for standard commercial and personal property accounts. Cytora (founded 2016) automated the triage of broker submissions for commercial lines, routing in-appetite risks to straight-through processing and flagging exceptions for human review. By the late 2010s, personal lines carriers were reporting 70-80% straight-through processing rates for clean submissions. The workforce declined from the 2000 level into the mid-2010s (reaching 89,910 in 2017) before rebounding as specialty commercial lines -- particularly cyber insurance, which emerged as a major product after the 2013-2015 breach wave -- required large numbers of experienced human underwriters that no automated system could replace.

  Effect on the work

  Employment reached its post-2000 OEWS trough at 89,910 in 2017, then grew as specialty lines expansion offset the continued automation of personal-lines processing. The net effect was a workforce bifurcation: carriers reducing personal-lines headcount while simultaneously hiring for commercial, cyber, and specialty roles.

• AI-powered underwriting platforms + lar…2022–present

  AI-powered underwriting platforms + large language models (agentic AI era)

  Generative AI entered insurance underwriting from two directions simultaneously. Large language models (ChatGPT, Claude) gave underwriters tools to parse complex submission documents -- lengthy engineering surveys, D&O prospectuses, contractor qualification packages -- at a fraction of the previous time. Akur8 deployed machine learning pricing engines that could recalibrate rate models from live loss experience in hours rather than weeks. Hyperscience's intelligent document processing automated the extraction of structured data from unstructured broker submissions. By 2025, McKinsey was projecting that 90% of pricing and underwriting tasks for personal and small-business insurance would be automated by 2030, with AI-driven adoption rates rising from 14% of carriers in 2024 toward 70% by 2028. Agentic AI architectures -- where specialized agents handle intake, risk profiling, pricing, compliance review, and decision orchestration as an autonomous pipeline -- were being piloted at leading carriers by 2025-26. The surviving human underwriter role concentrates on exactly what these systems cannot do: bespoke broker negotiation, coverage-term judgment for novel risks, regulatory filing expertise across many state jurisdictions, and the relationship capital that determines whether a specialty broker routes their best business to your carrier or a competitor.

  Effect on the work

  BLS projects a decline in insurance underwriter employment from 2024 to 2034, a much smaller decline than Frey and Osborne's 99% computerization probability predicted. The occupation is proving more resilient than headline automation projections suggested, but the composition is shifting: routine-processing roles are declining while specialty and complex-commercial roles are growing, requiring more education, more experience, and materially better judgment.