How a Market Conduct Examination Works
In a market conduct examination, state examiners test a sample of an insurer's files against state law and write a report.
- For
- People who work with or inside insurers and have never been through an examination: data science, model risk, product and vendor teams, and compliance staff arriving from other industries.
- Read if
- You keep hearing that a rule is enforced through market conduct examinations and want to see how one runs, from the first data request to the order that ends it.
In this article
A market conduct examination is a state insurance department’s review of how an insurer treats its policyholders.1 Examiners take a set of the company’s files, such as policies it canceled or claims it handled, check a sample of them against that state’s insurance law, and write a report on what they find. A financial examination is the other kind, and it asks a different question: whether the insurer is solvent, meaning able to pay its claims.1
One examination shows how the steps fit together. The Pennsylvania Insurance Department began examining Lemonade Insurance Company in June 2024. The property policies it tested covered homeowners, tenants and condominium owners. Eleven months later it issued its report, together with an order in which Lemonade agreed to pay $70,000 and to carry out the report’s recommendations.2
Who can examine an insurer
Lemonade was incorporated in New York. Pennsylvania examined it all the same, for the business it did in Pennsylvania.3
Insurance is regulated state by state, and a state can generally examine any insurer it has licensed. Pennsylvania acted under its own law: the report cites two sections of the state’s Insurance Department Act of 1921.4
Under the 1921 act the department may examine any company as often as the commissioner, the department’s head, “in his sole discretion deems appropriate.”5 It must examine every licensed insurer at least once every five years, though not necessarily in a market conduct examination, because the text does not say which kind. For an out-of-state insurer it may instead accept the home state’s report, subject to conditions.5
By our comparison, those provisions follow the Model Law on Examinations (the examination model) nearly word for word.6 A model law is a template statute drafted by the National Association of Insurance Commissioners (NAIC), through which state regulators coordinate, and it binds only where a state enacts it. The NAIC’s 2019 chart counts most states as having adopted this one.6
What gets a company examined
In some states a statute requires market conduct examinations themselves at regular intervals, according to the NAIC’s Market Regulation Handbook, which is guidance for examiners, not law.7
Elsewhere the department chooses, and Florida’s statute shows what can move a company to the front. It tells the state’s insurance regulator to give priority to insurers meeting any of several conditions. One is a number of claims-handling complaints out of proportion to the insurer’s market share. Another is a result marking the insurer “a negative outlier” in the Market Conduct Annual Statement, the uniform claims and underwriting data that insurers file each year.8
A department with a concern can do less than examine. The Handbook says that unless the law requires an examination there are often more efficient and cost-effective ways to respond, though at times an examination will be the best choice. A second NAIC model law, the Market Conduct Surveillance Model Law (the surveillance model), lists steps a commissioner may consider before an on-site examination, among them correspondence, interviews and written questions. Few states have enacted it whole, by the NAIC’s 2021 chart.9
State departments’ own counts, compiled by the NAIC, suggest the lighter steps are more common. For 2022 they come to 468 examinations covering market conduct and about four times as many market actions that were not examinations.10 The comparison is rough: some jurisdictions did not report, and an action can be as small as one letter.10 The Lemonade report does not say why the company was chosen or whether anything milder was tried first.
How examiners test the files
The Lemonade examination covered 2023, and its scope listed seven areas, among them underwriting, claims, complaints and policy forms. Pricing, which the report calls rating, was tested too. A typical test takes a defined population of files and measures it against a named provision of law.11
Take the test of policies canceled midterm. The report says it measured compliance with a Pennsylvania provision on when a policy may be canceled and what form the cancellation notice must take. The population was every property policy that ended midterm during 2023: 19,292 of them. From those, 175 files were selected for review, by a method the report does not describe.12
Some of those files held cancellation notices that left out the amount due, the date it was due, or both. The regulation the report cites requires a notice to state them when the reason for canceling is an unpaid premium. In all, 24 of the 175 files had at least one violation. The report divides files with violations by files reviewed and calls the result an error ratio, here 14 percent.13
Is 14 percent too many? No national rule answers that. The Handbook describes two kinds of state law. Under some laws against unfair practices, one standard for a violation is frequency: conduct that occurs often enough to amount to what they call a general business practice. Many others do not depend on frequency, and each instance is a separate violation.
For states of the first kind the Handbook supplies benchmarks, historically 7 percent for claim practices and 10 percent for other trade practices. An error ratio above the benchmark is presumed to show a general business practice. The Handbook strongly encourages those states to use that test, “absent state case law that may apply an alternative standard.”14
On our reading, Pennsylvania’s law has both kinds: the cancellation provision has no frequency element, and its claims rules do.15 So one cancellation that breaks that provision is a violation, and whether 14 percent is high is not a question it asks.
Both claims tests that found violations came in at 3 percent, which by our comparison is below the Handbook’s 7 percent benchmark, yet they drew a recommendation.15 The report counts any instance of noncompliance as a violation, but neither it nor the order says whether 3 percent amounted to a general business practice.15
From first request to order
The examination began on June 11, 2024 and had no on-site portion.16 Lemonade was sent a preliminary packet that set out the specific data it had to supply. Once it had supplied everything, the department checked the data for accuracy and completeness and recorded no violation.16
Throughout the examination Lemonade’s officials received status memoranda that cited policy numbers and the section of law each was said to violate. So the company knew each finding before it received the report. The department sent the report to Lemonade on April 1, 2025. The report defines its six recommendations as corrective measures the department finds necessary, and the first says the company must make sure its cancellation notices state the amount due and the due date.17
Lemonade replied in writing on April 30, recommendation by recommendation. On the first, its letter says it had revised its notices and filed the new forms with the state on February 25, five weeks before the report arrived. The department then published the report, dated May 22, with Lemonade’s reply and the order that closed the examination attached.18
How an examination ends
Lemonade’s examination ended in a consent order, an order the company agrees to. Lemonade waived its right to a formal hearing and accepted the terms, among them compliance with the report’s six recommendations. The order describes the $70,000 as a payment “in settlement of all violations contained in the Report” and gives no calculation for it.19
A second cost came from pricing. By the report’s account, Lemonade issued policies with an “incorrect premium capping factor,” and each one violated the statute that allows a policy to be issued only in accordance with filed rates. Pennsylvania had Lemonade look for the same error in 2021, 2022 and 2024, outside the year examined. That review produced about $172,000 in restitution, meaning money returned to policyholders: more than twice the payment in the order.20
An examination can also end with “No further regulatory action,” the first item on the Handbook’s list of outcomes. The choice among those outcomes belongs to the department, not to the examiners who test the files.21
Where AI comes in
Neither the Lemonade report nor the Handbook’s 2025 edition uses the phrase artificial intelligence. The NAIC has drafted an AI Risk Evaluation Supplement, an optional set of questionnaires meant to be added to existing market conduct and financial reviews. In a pilot, according to an NAIC working group’s minutes, some states used the Supplement as part of an examination that was already planned, and others as a questionnaire outside the examination process.22
Status as of the morning of November 16, 2026: the Supplement is still a draft. Version 6.0 was exposed for written comments through October 30, and version 7.0 has since been posted; the working group’s agenda for its noon session on November 16 lists 7.0 for adoption (our news item has the details), and no version had been adopted that morning. A draft revision of the Handbook’s examination standards, dated August 18, 2026 and not adopted, would add this instruction for examiners: “Verify that underwriting models or machine learning algorithms accurately assess and price risk.”23
Which documents to have ready is covered separately.
What depends on the examining state
On our reading, four matters turn on the examining state’s law. In Pennsylvania the 1921 act’s article on examinations sets no number of days’ notice but requires a scheduling conference before field work begins.24 The company has not more than thirty days to reply to the report in writing. An adopted report stays confidential for thirty days. Then, unless a court has stayed publication, the department may make it public. The company bears the cost.24
On publication, Pennsylvania’s “may” matches the examination model; the surveillance model says “shall.”25 Washington requires at least sixty days’ notice, except in extraordinary circumstances, and pays for examining insurers based there.25
The published record does not show how much notice Lemonade had or what the examination cost it. The rules on all four came from the law Pennsylvania examined it under. In any examination, then, the first things to learn are the examining state and the law it is acting under.
Footnotes
-
NAIC, “Market Conduct Regulation,” insurance topics page, last updated May 15, 2025, which sets market conduct regulation against solvency regulation. https://content.naic.org/insurance-topics/market-conduct-regulation ↩ ↩2
-
Pennsylvania Insurance Department, Bureau of Market Actions, “Market Conduct Examination Report of Lemonade Insurance Company,” as of April 1, 2025, issued May 22, 2025, published in one PDF with the Consent Order of the same date (Docket No. MC25-05-001) and the company’s written response. The policy types are those of the files tested (for example, page 20). Page numbers in these notes are those of the PDF. https://www.pa.gov/content/dam/copapwp-pagov/en/insurance/documents/regulations/regulatory-actions/lemonade-insurance-market-conduct-05-2025.pdf ↩
-
Pennsylvania report (note 2), page 17 (incorporation in New York) and page 16 (the purpose of the examination). ↩
-
Pennsylvania report (note 2), page 16, citing Sections 903 and 904 of the Insurance Department Act of 1921 (40 P.S. §§ 323.3 and 323.4). Section 903(b) reaches “any company” and “every insurer licensed in this Commonwealth” (https://www.legis.state.pa.us/WU01/LI/LI/US/HTM/1921/0/0285..HTM). The general rule is our reading of that section and three other texts: the NAIC’s Model Law on Examinations (#390), Section 3A (https://content.naic.org/sites/default/files/model-law-390.pdf), and the statutes of two states that the chart in note 6 does not list as adopters of it, N.Y. Ins. Law § 309(a) (https://www.nysenate.gov/legislation/laws/ISC/309) and Fla. Stat. § 624.3161(1) (http://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0600-0699/0624/Sections/0624.3161.html). ↩
-
Insurance Department Act of 1921 (note 4), Section 903(b), and Section 903(d), which sets the conditions for accepting the home state’s report; Section 902 defines the commissioner as the Insurance Commissioner of the Commonwealth. The five-year requirement is for “an examination,” and the scheduling factors the section lists are mainly financial, such as financial statement analyses and actuarial opinions. ↩ ↩2
-
Model Law on Examinations (note 4), Section 3A; the comparison with Section 903(b) of the 1921 act is ours. NAIC, state chart for the model, dated Summer 2019, which says it is not a formal legal opinion of NAIC staff (https://content.naic.org/sites/default/files/model-law-state-page-390.pdf). “Most states” is our count of its “Model Adoption” column, which lists Pennsylvania’s 40 Pa. Stat. Ann. §§ 323.1 to 323.8. On the NAIC’s role, see the page in note 1. ↩ ↩2
-
NAIC, Market Regulation Handbook, 2025 edition, Chapter 13, page 185, on jurisdictions that require examinations at regular intervals, with financial examinations or separately. https://content.naic.org/sites/default/files/publication-market-reg-hb.pdf ↩
-
Fla. Stat. § 624.3161(8) (link in note 4), paragraphs (b) and (c) of the six it lists. On the Market Conduct Annual Statement: NAIC insurance topics page of that name, last updated September 25, 2025. https://content.naic.org/insurance-topics/market-conduct-annual-statement ↩
-
Handbook (note 7), Chapter 2, page 17. NAIC, Market Conduct Surveillance Model Law (#693), Section 4B (https://content.naic.org/sites/default/files/model-law-693.pdf). Adoption is our count of the “Model Adoption” column in the NAIC’s state chart for that model, dated Fall 2021 (https://content.naic.org/sites/default/files/model-law-state-page-693.pdf). ↩
-
NAIC, Insurance Department Resources Report, Volume One, September 2023 edition, with 2022 data: Tables 14 and 17 (339 examinations of market conduct alone and 129 combined financial and market conduct examinations, in each case where the reporting department was the lead state) and Table 18 (1,939 market actions that were not examinations). Florida reported 487 of the 1,939, most of them, by a note to the table, for insurers that did not file required data on time; without them the ratio is about three to one. The tables do not say how a letter is counted; correspondence with the insurer is the first item on the surveillance model’s list (note 9). https://content.naic.org/sites/default/files/publication-sta-bb-volume-one.pdf ↩ ↩2
-
Pennsylvania report (note 2), page 16, where the period is 2023 “unless otherwise noted.” ↩
-
Pennsylvania report (note 2), pages 19 and 20. The report counts as midterm any policy that ends other than on its twelve-month anniversary date, and does not say who ended the policies in the population. The provision is Section 5(a)(9) of the Unfair Insurance Practices Act (40 P.S. § 1171.5(a)(9)), on canceling or not renewing policies on owner-occupied homes or individuals’ personal property in force sixty days or more (https://www.legis.state.pa.us/WU01/LI/LI/US/HTM/1974/0/0205..HTM). ↩
-
Pennsylvania report (note 2), page 20: 42 violations in 24 files, 18 of them cited to 31 Pa. Code § 59.6(6) (https://www.pacodeandbulletin.gov/Display/pacode?file=/secure/pacode/data/031/chapter59/s59.6.html&d=reduce). In the error ratio a file with several violations counts once (page 14). ↩
-
Handbook (note 7), Chapter 17, page 226. The laws of the first kind are state statutes based on the NAIC’s model acts on unfair trade practices (#880) and unfair claims settlement practices (#900). The same figures also set the tolerance for error when a sample is drawn; the page says that tolerance “has little to do with the real tolerance that a jurisdiction may have for error,” and that a first-stage sample cannot establish with confidence that the true rate of noncompliance exceeds 7 or 10 percent. ↩
-
Unfair Insurance Practices Act (note 12): Section 5(a)(9) has no frequency language, and neither does 31 Pa. Code § 59.6 (note 13); Section 5(a)(10) makes listed claims acts unfair “if committed or performed with such frequency as to indicate a business practice.” 31 Pa. Code § 146.1 says the claims standards, “if violated with a frequency that indicates a general business practice, will be deemed to constitute unfair claims settlement practices”; it does not say a single breach is no violation (https://www.pacodeandbulletin.gov/Display/pacode?file=/secure/pacode/data/031/chapter146/s146.1.html&d=reduce). Sorting these provisions into the Handbook’s two kinds is our reading of their text. Pennsylvania report (note 2), page 14 (the definition of a violation), pages 33 and 34 (the homeowner and tenant claims tests, each with “an error ratio of 3%” and violations cited to 31 Pa. Code § 146.6) and page 42 (recommendation 3). The Consent Order (note 2) states that the § 146.6 violations are punishable under the Unfair Insurance Practices Act. ↩ ↩2 ↩3
-
Pennsylvania report (note 2), page 14 (the starting date and the absence of an on-site portion) and page 41 (the data packet, the department’s check of the data, and no data integrity violations). ↩ ↩2
-
Pennsylvania report (note 2), page 14 (the status memoranda) and pages 42 and 43 (the recommendations; the first cites 31 Pa. Code § 59.6(6)). Consent Order (note 2), page 7 (the report sent to the company on April 1, 2025). ↩
-
Company response (note 2), pages 45 and 46, dated April 30, 2025. Report cover: “As of: April 1, 2025” and “Issued: May 22, 2025.” Consent Order, page 6. ↩
-
Consent Order (note 2), pages 6 to 11, in which the department “orders and Respondent consents to” the terms. ↩
-
Pennsylvania report (note 2), pages 29 and 30, citing 40 P.S. § 1224(a) and (i) of the Fire, Marine and Inland Marine Rate Regulatory Act. Section 1224(i) bars an insurer from making or issuing a policy “except in accordance with the filings or rates which are in effect” for it (https://www.legis.state.pa.us/WU01/LI/LI/US/HTM/1947/0/0247..HTM). The restitution was $171,649.67, and the report says the department has received proof that it was paid. ↩
-
Handbook (note 7), Chapter 16, page 208, which lists the dispositions and says who decides. ↩
-
NAIC, “Artificial Intelligence (AI) Risk Evaluation Supplement,” version 5.0 exposure draft (https://content.naic.org/sites/default/files/inline-files/ai-risk-evaluation-supplement-v5.0-clean.docx). The pilot: August 13, 2026 minutes of the NAIC’s Big Data and Artificial Intelligence (H) Working Group, adopted August 31, which report the statement in paraphrase (https://content.naic.org/sites/default/files/national_meeting/minutes-bdaiwg081326.pdf). The Handbook (note 7) says it reflects guidance adopted as of December 2024. ↩
-
Supplement: NAIC, Big Data and Artificial Intelligence (H) Working Group page (https://content.naic.org/committees/h/big-data-artificial-intelligence-wg). As read on the morning of November 16, 2026, its Exposure Drafts section links version 7.0 clean and edited PDFs beside the version 6.0 files, none marked adopted, and the agenda for the group’s Fall National Meeting session that day, noon to 1:00 PM CT, lists as item 4 “Consider Adoption of the AI Risk Evaluation Supplement v7.0.” The parent Innovation, Cybersecurity, and Technology (H) Committee’s agenda, third draft dated November 10, lists “Consider Adoption of the AI Risk Evaluation Supplement” for November 17. As read on October 23, 2026, its Exposure Drafts section exposed version 6.0, “Following up from the October 16, 2026, Big Data and Artificial Intelligence (H) Working Group email notice,” “for a 14-day comment period ending Friday, October 30, 2026,” and the page listed the group’s next meeting for November 2, 2026. Handbook: revised Chapter 20, General Examination Standards, an exposure draft dated August 18, 2026 and listed by the Market Conduct Examination Guidelines (D) Working Group, which still listed it as a draft on November 16, 2026; it has not been adopted, and the instruction quoted is in its Underwriting and Rating section, Standard 4 (https://content.naic.org/sites/default/files/inline-files/chapter-20-general-exam-standards-revised-8-18-26.pdf). ↩
-
Insurance Department Act of 1921 (note 4), Article IX, Sections 901 to 908, among them Sections 904.1 (scheduling conference), 905(b) (a “reasonable opportunity” to reply), 905(e)(1) (confidentiality and publication) and 907 (cost). That the article’s text names no number of days’ notice before an examination is a reading of the statute, not of the department’s regulations or practice. The Lemonade report does not mention a scheduling conference, and the record does not say how many days Lemonade was given to reply. Under Section 907 the company pays the cost “in such manner as the commissioner shall by regulation prescribe.” ↩ ↩2
-
Model Law on Examinations (note 4), Section 5E(1), and Market Conduct Surveillance Model Law (note 9), Section 6I(1): each keeps a report confidential for a period first, and each lets it be opened only if no court has stayed its publication. The comparison with Section 905(e)(1) of the 1921 act (note 24) is ours. Wash. Rev. Code § 48.37.060(8), notice no later than sixty days before the estimated start, “except where the examination is conducted in response to extraordinary circumstances,” and (14), cost (https://app.leg.wa.gov/RCW/default.aspx?cite=48.37&full=true). The state pays for examinations within Washington of an insurer domiciled or with its home offices there, other than a title insurer, but not for witness costs; other examinations, and any part requiring travel and services outside the state, are at the expense of the person examined, though a domestic insurer does not pay the commissioner’s examiners for their work outside the state. ↩ ↩2
The Bottom Line
- A market conduct examination tests a sample of an insurer's files against one state's insurance law, while a financial examination looks at solvency.
- In general any state where an insurer is licensed can examine it, and the head of that state's insurance department decides whether to.
- How many violations are too many depends on the state's law, and under many state laws each instance is a separate violation.
- Examinations can end by agreement: Pennsylvania's examination of Lemonade found violations in 24 of 175 sampled files in one test and closed with a $70,000 settlement.
- Notice, response time, publication and cost depend on the examining state's law, so first establish which state is examining and which law it cites.
Simon Li · Founding Editor
I write InsureAI Wire and maintain its 51-jurisdiction tracker. Most of the work is reading: NAIC working group papers, state bulletins, bills, court filings, and public comment letters. Every claim on the site carries the document it came from, so you never have to take my word for it.
Free · Weekly
Track these developments weekly
Every state and NAIC move on insurance AI, with the source linked, in one email a week.
Free weekly · No spam · Unsubscribe anytime
Information aggregation and analysis, not legal advice. See our disclaimer.