AI use case

AI summarization of medical evidence for life and health underwriting

AI that reads the medical evidence behind a life or health insurance application (attending physician statements, electronic health records, lab results and disclosures), turns it into a structured, cited summary of conditions, treatments and dates, and maps it to the insurer's underwriting manual so an underwriter can decide faster and more consistently.

By Len Debets · Last verified 27 September 2026 · 2 public deployments

USD 300,000 to USD 1.6 million
Indicative value per year
A life insurer that fully underwrites 20,000 applications a year with medical records. Worked example, see how it is calculated.

What problem does it solve?

For fully underwritten life and health cover, much of the work is reading medical evidence. An attending physician statement or a set of health records can be long, mixing handwritten notes, scans and lab printouts in no particular order. Underwriters or nurse reviewers read it to find the handful of facts that matter (diagnoses, dates, medications, test values, smoking status) and then look them up in the insurer's underwriting manual.

That reading is slow, expensive and can differ between reviewers, and applicants wait while it happens. Electronic health records can make evidence available sooner, but someone still has to read it.

How does it work?

  1. Collect and digitize. Medical records arrive from providers, labs and record retrieval vendors; the system converts scans and handwriting to text and splits the file into encounters.
  2. Extract clinical facts. A model extracts diagnoses, procedures, medications, vitals and lab values with dates, and codes them to a standard vocabulary, each linked to its source page.
  3. Build the timeline. Facts are ordered into a timeline and checked against the applicant's own disclosures to highlight differences.
  4. Map to the manual. Retrieval over the underwriting manual suggests the relevant impairment guidance and any further evidence needed, without setting the final rating.
  5. Underwriter decides. The underwriter reviews the summary, opens the source pages where it matters and makes the decision; simple, clean cases can go to straight through rules that the insurer already governs.
Audience
Employee facing
Autonomy
Copilot
Adoption
Early adopters
Channels
Internal tools

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

No public deployment has disclosed a measurable outcome yet.

Value drivers: Speed and cycle time, Employee productivity, Customer experience, Compliance quality.

Indicative value

A life insurer that fully underwrites 20,000 applications a year with medical records

USD 300,000 to USD 1.6 million

Underwriting review time released per year

How this is calculated

Formula: applications * reviewHours * timeSavedShare * costPerHour. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Applications with medical records reviewed per year applications, applications per year20,00020,000The reference insurer.
Underwriter or nurse reading time per file reviewHours, hours per file12Editorial assumption. Replace with a time study of your own medical evidence review.
Share of reading time the summary removes timeSavedShare, fraction of reading time0.30.5Editorial assumption. No insurer on this page publishes a reading time figure, so replace it with the result of your own parallel run.
Fully loaded cost of an underwriting hour costPerHour, USD per hour5080Editorial assumption. Replace with your own fully loaded cost.

What it leaves out: Time released only. It leaves out the usually larger effect of faster decisions on placement rates (fewer applicants dropping out), record retrieval costs, platform costs and the effort of validating the model for a high risk use.

Who already uses it?

2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

Prudential plc

Hong Kong SAR China · Insurance · 2025

ProductionGrade B

In its 2025 full year results, Prudential said it continues to enhance its health underwriting with AI powered solutions designed to increase underwriting automation and efficiency, and that it launched MedScreen+ in Hong Kong, an AI underwriting tool intended to provide underwriters with a faster, simpler and more transparent process and to support its financial consultants with instant, indicative underwriting results for customers. In its 2024 results it reported that around 74 per cent of new business policies were processed through auto underwriting capabilities. No outcome figures are given for MedScreen+ itself.

No outcome disclosed.

Manulife

Canada · Insurance · 2024

ProductionGrade B

Manulife's 2024 annual report says generative AI in Singapore automates document digitization and summarization in underwriting, improving the accuracy of underwriting decisions and reducing processing time for policy applications, and that in the US it expanded the use of electronic health records and used generative AI to automate preliminary underwriting assessments. In 2025 it partnered with Munich Re Life US on alitheia, an AI driven risk assessment platform, raising the instant underwriting decision eligibility limit from US$3 million to US$5 million. No time or accuracy figures are disclosed.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Historical medical files with the underwriting decisions made on them, for testing
  • The underwriting manual (the insurer's own or a reinsurer's) in a form the insurer may use for retrieval
  • A clinical vocabulary and mapping for impairments used in the manual
  • Consent and authorization records for every medical record used

Systems to integrate

  • New business and underwriting workbench
  • Medical record retrieval vendors and electronic health record sources
  • Document management with page level retention
  • Rules engine for straight through decisions

Complexity: High

Clinical extraction from poor scans is hard, errors carry real consequences for applicants, and in the EU the use is high risk under the AI Act. Special category health data raises the bar on data protection, residency and access control. Expect formal model validation and a long parallel run.

  1. 1

    Start with summarization, not decisions

    The first release produces a cited summary and timeline for the underwriter. Straight through decisions stay with existing, validated rules until the summary is proven.

  2. 2

    Build a gold standard set

    Have senior underwriters and medical officers annotate a few hundred real files so recall of critical facts (for example a cancer history or abnormal lab value) can be measured, not guessed.

  3. 3

    Measure what is missed, not just what is right

    A missed impairment is far worse than an extra one. Track recall on critical conditions separately and set release thresholds with the chief underwriter and medical officer.

  4. 4

    Run in parallel

    For a period, underwriters review files the usual way and compare with the summary. Only reduce reading once differences are understood and documented.

  5. 5

    Govern it as a high risk system where applicable

    In the EU, meet the AI Act requirements for high risk systems (risk management, data governance, logging and human oversight) and, as the deploying insurer, carry out the fundamental rights impact assessment that Article 27 requires. Elsewhere follow the insurer's AI governance framework and local rules, such as Colorado's Regulation 10-1-1 where external consumer data or predictive models are involved.

Guardrails

  • Every extracted fact links to the page and passage it came from
  • The summary never sets the rating or declines an applicant on its own
  • Health data processed only in approved regions, with access limited to underwriting roles
  • Explicit consent or legal basis recorded for every record processed
  • No inference of protected characteristics or genetic information beyond what law permits

KPIs to instrument

  • Median days from application to decision, before and after
  • Reading time per file from workbench logs
  • Recall of critical impairments on the audited sample
  • Share of summaries corrected by underwriters, by error type
  • Placement rate (offers accepted) for summarized versus non summarized cases

Human in the loop

Underwriters make every decision and review source pages for any material fact. Medical officers own the clinical vocabulary and the gold standard set, and the chief underwriter signs off release thresholds and reviews a monthly sample of summaries against full reads.

Common failure modes

Missed conditions in poor scans
Handwritten notes and faxed pages drop out of extraction and the summary looks complete. Flag low quality pages and require a human to read them.
Automation bias in the underwriter
A clean summary is trusted and the source is never opened. Sample decisions and show the pages behind each material fact.
Unfair outcomes through proxies
Summaries emphasise facts that correlate with protected characteristics. Test outcomes by group and keep the underwriting manual, not the model, in charge of rating.
Consent and residency gaps
Records are sent to a model outside the approved region or without a valid basis. Route health data through controlled infrastructure only.

What are the risks and rules?

EU AI Act

High risk

Annex III point 5(c): AI intended for risk assessment and pricing in relation to natural persons in life and health insurance. Article 6(3) exempts some purely preparatory tasks, but never a system that profiles natural persons. Extracting an applicant's health conditions and mapping them to the underwriting manual evaluates their health, which is profiling, so treat the system as high risk. Under the timeline as amended, the obligations for Annex III high risk systems apply from 2 December 2027, and Article 27 requires deployers of point 5(c) systems to assess the impact on fundamental rights before first use.

Guidance

Controls to put in place

  • Registration as a high risk system in the EU and a conformity assessment before use
  • Fundamental rights impact assessment under Article 27 of the AI Act before the insurer first uses the system, including when it is bought from a vendor
  • Documented data governance for training and test medical data, including consent
  • Logging of every summary, its sources and the underwriter's decision
  • Periodic fairness testing of decisions made with the summary
  • Data protection impact assessment for special category health data
  • In the US, a valid HIPAA authorization from the applicant for every record requested from a provider: life insurers are generally not HIPAA covered entities themselves, while health insurers acting as health plans are and must meet the Privacy and Security Rules directly

Frequently asked questions

Can AI decide life insurance applications from medical records?
For some applications, AI does decide. Manulife says its partnership with Munich Re Life US on alitheia, an AI driven risk assessment platform, raised instant underwriting decision eligibility from US$3 million to US$5 million. Summarization tools, such as Manulife's generative AI in Singapore, support an underwriter instead, and Prudential launched MedScreen+ to provide a faster, simpler and more transparent process for its underwriters. In the EU, AI used for risk assessment and pricing of individual life or health cover is high risk under the AI Act, so keep an underwriter accountable for decisions the summary informs.
How much faster does underwriting get?
No insurer on this page publishes a figure for medical record summarization. Manulife says its generative AI in Singapore reduces processing time for policy applications but gives no number, and Prudential discloses no outcome for MedScreen+. Measure it yourself in a parallel run, and separate reading time from time spent waiting for records.
What is the biggest risk?
Missing a material condition. Measure recall on critical impairments against a gold standard, keep source pages one click away and sample decisions, because a fluent summary is easy to trust too much.

How to cite this page

Blits.ai AI Use Case Library, "AI summarization of medical evidence for life and health underwriting", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/life-underwriting-medical-record-summarization. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

Insurance

AI copilot for underwriting risk assessment

A copilot that assembles everything relevant to a risk (the submission, loss history, internal guidelines, third party data and public information), highlights exposures and gaps against the insurer's underwriting guidelines and drafts the underwriting narrative or referral note, while the underwriter makes and signs every decision.

Deployments
8 public, best grade B
Reported cycle time reduction
50%
Generali Global Corporate & Commercial, vendor claim
InsuranceHealthcare

AI for health insurance prior authorization and claims adjudication support

AI that reads prior authorization requests, medical claims and appeals with their clinical and billing documents, extracts diagnoses, treatments and costs, checks them against the policy and published clinical criteria, and prepares a summary and recommendation for a clinician or adjudicator, who makes every adverse decision.

Deployments
5 public, best grade B
Reported handling time reduction
about 50%
Acentra Health, vendor claim
Cross industryGovernment and public sector

AI document intelligence for unstructured forms and documents

AI that takes documents in any format, such as scanned forms, PDFs, photos, emails and handwritten notes, splits and classifies them, extracts the required fields with a confidence score, validates them against business rules and source systems, and sends only the uncertain cases to a person before the data enters the downstream process.

Deployments
5 public, best grade B
Reported accuracy
at least 90%
Ancine, vendor claim
Insurance

AI copilot for insurance pricing and actuarial analysis

AI that speeds up the work of pricing and actuarial teams, from automated, transparent risk and demand model building to natural language analysis of rate filings, experience data and reserving diagnostics, while actuaries select the models, sign off the rates and own the professional judgment.

Deployments
5 public, best grade B
Reported productivity gain
5x
Generali France, organization claim