Where the numbers come from.

The sources and assumptions behind the figures on this site.

Sources and assumptions.

Home: the problem

  • EY via Insurance Journal (Oct 2025): commercial litigation funds and expected cost to insurers.
  • APCIA and Milliman closed claim study (2022, claims closed 2019): litigated share of claims and paid dollars, and the 14x cost of a litigated GL injury claim.
  • Sedgwick (2026): days from first notice to lawsuit.

Production results

  • Production claims at a large commercial carrier, Aug to Sept 2026: about 3,000 new claims over seven weeks, all lines.
  • Pre-suit share, page depth, and accuracy are from a June 2026 review. Accuracy: 5 signal errors in 692 reviewed runs, counting both false alarms and misses.
  • Share of risk findings from inside attachments: about 2,300 new claims over six weeks at the same carrier.
  • Detection time: risks are detected within minutes of a message arriving; the 30-minute median runs from arrival to the handler’s alert, including routing.

Detecting on the graph

  • Claim-system comparison: every critical- or high-urgency risk finding at a large commercial carrier, Feb 12 to Oct 16, 2026 (2,930 findings on 1,770 claims), classified against the standard data model of a leading claim system, without the email and documents. 49% rested on facts with no structured field, 17% on a field that records the event but not what made it critical or high, and 34% on a structured field. Notes and diary entries are discretionary and don’t count.
  • Graph against raw documents: June 2026 tests on a carrier’s book of about 6,300 claims, from 550 graded runs. First-notice fields read from raw files: 70–89% correct across four model tiers, from OCR text or with the model reading the files directly. Small model without structure: an exploratory run at the smallest model tier scored 0.57 on risk detection without the graph and 0.93 with the graph and detection rules. Six of seven tasks: the smallest model tier on the graph against the largest reading raw documents, at about 1/25th the per-token price. Court deadlines: 8 litigation cases; the smallest model tier on the graph against the largest reading the raw papers. Text per question: 408 tokens on the graph against 7,439 for the relevant pages as OCR text.

Lines of business

  • AM Best via Insurance Journal (Sept 2025): commercial auto underwriting results.
  • The employment practices X-ray is a real production finding, with names and details changed. The general liability, commercial auto, and workers’ compensation X-rays are representative.
  • The examples on the Isomer Signals page are production findings, with names and details changed.

Pre-suit litigation risk

  • Sedgwick State of the Line (Mar 2026): share of litigated GL claims with a lawyer within 24 hours of notice.

Reserve accuracy

  • Sedgwick (Summer 2026): bodily-injury settlement and verdict growth in Sedgwick’s own book.
  • Triple-I and CAS (Oct 2025): share of booked losses in commercial auto and GL from legal system abuse, 2015 to 2024.

Impact calculator

  • Context for the model: APCIA and Milliman closed claim study (2022); Sedgwick State of the Line (2026); Triple-I and Casualty Actuarial Society (2025).
  • LexisNexis Risk Solutions claimant survey (2023): 57% of represented auto claimants hired a lawyer before filing.

Illustrative value

  • The $1B book is an illustrative scenario, not a guarantee. It assumes a 68% loss ratio and a 10% defense cost ratio, entered separately so no dollar is counted twice. The line presets come from NAIC 2025 statutory data: direct losses incurred and defense and cost containment expense, each as a share of premium, averaged across each line’s eight largest states and weighted by premium. Commercial auto liability is 78% and 10%, general liability 73% and 10% (NAIC’s other liability, which also includes claims-made lines such as D&O and EPL), and workers’ compensation 47% and 8%. The commercial book is those three blended by premium. All are rounded; enter your own.
  • High-risk claims are 3% of claims and carry 50% of loss, matching the share of liability claims that are litigated and their share of paid dollars cited above. They are assumed to carry 80% of defense cost, since defense counsel is retained mostly on litigated claims.
  • It assumes 50% of high-risk claims are caught in time today and 70% with Isomer. Of claims caught in time, 20% avoid a suit and save 75% of their defense cost, the rest cost 15% less to defend, and settlements come in 7.5% lower.
  • Savings are a run rate on each year of claims. By calendar year, defense savings are paid 60% in the year a claim opens and 40% the next, and settlement savings 20%, 40%, and 40% over three years.
  • The comparisons assume an 8% underwriting margin and $125K per adjuster, fully loaded. Avoided blowups (excess verdicts and bad-faith judgments) and more accurate reserves are not in the total: blowups are too rare to average, so the calculator shows what one $5M blowup is worth, and better reserves book the same dollars earlier rather than saving new ones. A pilot on your own claims replaces these assumptions with your actuals.