Claims Adjusters and AI Tools: Documenting Your Own Judgment Under MA DOI Rules

In December 2022, Black homeowners Jacqueline Huskey and Riian Wynn sued State Farm Fire & Casualty Company in federal court in Chicago (Huskey v. State Farm Fire & Casualty Co., Case No. 1:22-cv-07014, N.D. Ill.), alleging that automated fraud-triage algorithms routed Black policyholders' roof and water-damage claims into slower investigation tracks, extra paperwork demands, and lower payouts (Civil Rights Litigation Clearinghouse). On September 11, 2023, Judge Virginia M. Kendall denied State Farm's motion to dismiss the Fair Housing Act disparate-impact claim under 42 U.S.C. 3604(b), opening discovery into how software flags steered human adjusters (Justia Opinion, Sep 11, 2023). Neither Huskey nor Estate of Gene B. Lokken v. UnitedHealth Group (D. Minn., challenging the nH Predict post-acute algorithm) has produced a final dollar damages verdict as of October 2026. Yet coverage counsel warn that bad-faith plaintiffs already subpoena adjuster override rates, screen-dwell timestamps, and supervisor scorecards that penalize departing from AI estimates (Zelle LLP via JD Supra).

Valuation software litigation has already yielded eight-figure settlements where automated markdowns went unchecked. In Volino v. Progressive Casualty Insurance Co. (Case No. 1:21-cv-06243, S.D.N.Y.), policyholders won a $48 million class settlement over automated "Projected Sold Adjustment" deductions inside Mitchell International's WorkCenter Total Loss platform (Insurance Journal). Meanwhile, over twenty state insurance departments have adopted the National Association of Insurance Commissioners (NAIC) Model Bulletin on Artificial Intelligence, including Massachusetts via Division of Insurance (DOI) Bulletin 2024-10 on December 9, 2024 (MA DOI Bulletin 2024-10 PDF).

This guide serves property and casualty desk and field adjusters, auto appraisers, Special Investigations Unit (SIU) analysts, claims supervisors, and agency liaisons across the United States and Boston's insurance hub as of October 29, 2026. Because ed.cafe at No Worker Left Behind (a Boston 501(c)(3) nonprofit) publishes workforce education rather than legal advice, confirm carrier procedures with your compliance and legal teams. Policyholders appealing health-plan denials can use our medical claim algorithm appeal checklist.

How AI estimating, fraud-scoring, aerial-imagery, and GenAI tools enter a modern claim file

Across property, auto, and casualty lines, claims adjusters and AI tools meet at four software checkpoints, each capable of quietly swapping a statistical guess for a verified fact in the claim diary.

First, estimating and total-loss platforms (CCC Intelligent Solutions' CCC ONE, Mitchell WorkCenter, and Verisk Xactimate photo modules) score vehicle or roof photos and pull dealer listings, applying automated condition or negotiation markdowns. Second, SIU fraud-scoring engines (Shift Technology and FRISS) ingest claim metadata to generate anomaly scores holding files out of payment. Third, aerial computer-vision tools compare pre-storm and post-storm satellite tiles to label missing tabs "pre-existing wear." Fourth, generative-AI summarizers condense 400-page bodily-injury demand packets into short reserve memos.

Picture a municipal building inspector signing off on a fire-damaged triple-decker from a satellite photo and a contractor's lookup sheet without walking the framing or checking moisture meters. Where that inspector analogy breaks down is courtroom telemetry: a paper lookup table shows fixed numbers, whereas claims software blends computer-vision scores, price haircuts, and LLM summaries while logging to the second whether the adjuster opened the underlying attachments before clicking "Approve."

Compare task error rates before trusting machine summaries in how often AI is wrong by task.

What Massachusetts DOI Bulletin 2024-10 and Bulletin 2025-02 require from insurers and vendors

Massachusetts Division of Insurance Bulletin 2024-10 (December 9, 2024) and Bulletin 2025-02 (April 30, 2025) put carriers and claims teams on formal notice that automated tools never relax statutory fair-claims duties (MA DOI Bulletin 2024-10 PDF; MA DOI Bulletin 2025-02 PDF).

Issued by Commissioner Michael T. Caljouw from the December 4, 2023 NAIC Model Bulletin, Bulletin 2024-10 governs predictive models, machine learning, and LLMs across underwriting, rating, marketing, and claims and SIU operations (NAIC Model Bulletin PDF):

1. Written AIS Program: Carriers must maintain a written Artificial Intelligence Systems (AIS) governance framework defining human oversight and override authority for consumer-impacting claim actions.
2. No Vendor Shield: Buying software from CCC, Mitchell, Shift, Verisk, or an aerial vendor leaves the insurer fully responsible under M.G.L. c. 176D. Carriers must vet vendor data, drift controls, and audit rights; see our questions to ask an AI vendor.
3. Market-Conduct Discovery (M.G.L. c. 175 s. 4): During market-conduct examinations, the DOI can demand AIS inventories, validation studies, vendor contracts, adjuster training materials, and override records.
4. Aerial Imagery Rules (Bulletin 2025-02, April 30, 2025): Insurers must verify aerial image clarity and age, cannot rely solely on cosmetic roof streaking, must perform due diligence including a follow-up physical inspection when images are inconclusive, and must review homeowner counter-evidence. Neighboring regulators set parallel rules in Connecticut (March 2024), Pennsylvania (Notice 2024-06), and Rhode Island (Bulletin 2025-3, requiring images under 15 months old).

Why rubber-stamping an AI recommendation triggers M.G.L. c. 176D and Chapter 93A treble damages

In Massachusetts, rubber-stamping an AI claim reduction or fraud hold without conducting an individualized factual review exposes the carrier to unfair-settlement liability under M.G.L. c. 176D s. 3(9) and mandatory double-to-treble damages plus attorney's fees under M.G.L. c. 93A s. 9 (MA General Laws c. 176D s. 3).

Under M.G.L. c. 93A s. 9(1), consumer policyholders hold a direct private right of action for any violation of c. 176D s. 3(9) (MA General Laws c. 93A s. 9). Four clauses collide directly with unverified AI claims handling: Clause (d) bars refusing to pay claims without a reasonable investigation based upon all available information; Clause (f) requires prompt, fair settlements when liability is reasonably clear; Clause (g) bars lowball offers forcing insureds to sue; and Clause (n) requires a prompt reasonable explanation of the basis in the insurance policy in relation to the facts or applicable law for any denial or compromise offer.

Test the exposure math on one Worcester homeowner claim (assumptions stated):

• Actual loss: A windstorm creases roof shingles; a licensed roofer submits an itemized $28,000 replacement scope with brittle-test video and code drip-edge photos.
• Unverified AI shortcut: An aerial plugin labels 18-month-old satellite shading "pre-existing wear" and outputs a $4,200 patch estimate (an 85.0% reduction in the $28,000 claim sample). Chasing a closure quota, the desk adjuster clicks "Confirm" in two minutes without ordering a ladder check or reviewing the video.
• Chapter 93A multiplier: If the homeowner wins a $28,000 trial judgment and the court finds a willful c. 176D s. 3(9)(d) failure to investigate, M.G.L. c. 93A s. 9(3) multiplies the judgment 2x to 3x ($56,000 to $84,000), adds 12% annual statutory prejudgment interest (M.G.L. c. 231 s. 6C rate rule), and awards mandatory plaintiff attorney's fees often topping $75,000.

As coverage counsel at Zelle LLP warn, plaintiffs in bad-faith discovery subpoena audit logs showing whether an adjuster agreed with the algorithm in 95% of assigned claim files, how many seconds elapsed before clicking confirm, and whether KPI bonuses penalized upward overrides (Zelle LLP via JD Supra).

How California, NAIC states, and auto/medical rules restrict sole reliance on claim algorithms

Major US insurance jurisdictions share one rule: software may assist claim triage and pricing research, while a qualified human must own every adverse valuation, delay, or denial.

In Massachusetts auto claims, Auto Damage Appraiser Licensing Board rules (212 CMR 2.00) also require licensed appraisers to inspect damaged vehicles and negotiate supplements in good faith with repair shops rather than treating first-pass photo AI scopes as final (Mass.gov 212 CMR 2.00).

Five habits that keep your adjuster license, audit trail, and independent judgment intact

Whether you handle 140 open auto and property files at a regional mutual or review commercial claims at an agency, the ed.cafe Adjuster Independent-Judgment File Note & AI Override Log protects both policyholders and your licensing record in under three minutes per claim.

1. Never paste claim files into public chatbots: Bodily-injury packets, driver's licenses, and bank details trigger Massachusetts 201 CMR 17.00 and HIPAA/GLBA rules; see what not to paste into an AI tool at work.
2. Check raw PDF pages behind GenAI summaries: Verify ER intake notes, imaging, operative reports, and bill totals by page number before setting reserves.
3. Audit auto comparables and roof photo dates: Check trim, mileage, and local dealer availability on total losses, and log aerial photo dates before declining ladder inspections.
4. Never cite "AI score" in a denial letter: Under M.G.L. c. 176D s. 3(9)(n), cite exact policy sections applied to verified facts, never a Shift suspicion score.
5. Log both agreements and overrides: Use the ed.cafe Adjuster Independent-Judgment File Note & AI Override Log to record primary evidence checked, and protect diagnostic habits using keeping skills sharp when working with AI and the ed.cafe Staff Room.

Copy-ready template: The ed.cafe Adjuster Independent-Judgment File Note & AI Override Log

Paste Part A of the ed.cafe Adjuster Independent-Judgment File Note & AI Override Log into your claim diary before issuing any valuation, SIU hold, or coverage decision; supervisors can run Part B quarterly for DOI Bulletin 2024-10 readiness.

PART A: 4-LINE CLAIM-FILE DIARY TEMPLATE (PASTE BEFORE PAYMENT / OFFER / DENIAL)
Claim #: ____________________  Date/Time: ______________  Adjuster: ____________
1. AUTOMATED TOOL OUTPUT RECEIVED (Advisory Input Only):
   Tool / Vendor: [CCC ONE / Mitchell / Xactimate AI / Aerial / GenAI / SIU]
   Raw Output: [e.g., $4,200 patch scope / Aerial wear flag dated __/__]
2. INDEPENDENT PRIMARY EVIDENCE REVIEWED BY ADJUSTER:
   [ ] Insured statement [____]   [ ] On-site / appraiser inspection [____]
   [ ] Shop / roofer supplement, photos & moisture/brittle test [____]
   [ ] Source PDFs verified by page (Pages [__-__] of medical / invoice packet)
   [ ] Local market comparables / NOAA wind & hail data for [ZIP / Date]
3. POLICY PROVISION & FACTUAL FINDING (M.G.L. c. 176D s. 3(9)(d) & (n)):
   Policy Form, Section & Page: [Form #, Section __, Page __]
   Verified Facts: [Summary of physical damage, cause of loss & scope]
4. ADJUSTER DISPOSITION ON TOOL OUTPUT:
   [ ] ADOPTED WITH VERIFICATION: Matched Step 2 primary evidence because: _____
   [ ] MODIFIED / OVERRIDDEN (+$[____] / -$[____] / Ladder Check Ordered):
       Departed from tool output because primary evidence showed: ______________
Adjuster Signature / Appraiser License #: ______________________________________

PART B: SUPERVISOR QUARTERLY KPI & MA DOI BULLETIN 2024-10 AUDIT CHECK
[ ] 1. ZERO OVERRIDE PENALTY: KPI scorecards and bonuses do NOT penalize cycle
       time or variance on evidence-backed adjuster overrides or inspections.
[ ] 2. RUBBER-STAMP CHECK: Audit 20 files per unit where AI scopes were adopted
       unchanged in <3 minutes; confirm Step 2 source records were opened.
[ ] 3. AERIAL CURRENCY (BULLETIN 2025-02): Zero adverse actions relied solely
       on blurry/old aerial images or cosmetic roof streaking.
[ ] 4. DENIAL AUDIT (c. 176D s. 3(9)(n)): Zero letters cite "AI score" or
       unverified software markdowns instead of policy text and facts.

Where this stops being true

Operational claim-diary templates hit boundaries where coverage counsel or special statutes govern:

• Undisputed full-payment fast tracks: When straight-through processing pays 100% of a submitted glass or towing invoice in claims with zero deduction, full payment creates no policyholder harm under M.G.L. c. 176D. Detailed judgment logs matter whenever a tool reduces, delays, flags for SIU, or denies any part of a claim.
• Pending class dockets: As of October 29, 2026, Huskey v. State Farm (1:22-cv-07014, N.D. Ill.) and Estate of Gene B. Lokken v. UnitedHealth Group (0:23-cv-03514, D. Minn.) survived partial dismissal motions without final dollar damages verdicts yet entered; cite them for discovery and Fair Housing Act / contract theories rather than payout totals.
• ERISA self-funded health plans: Employer self-funded benefit plans under federal ERISA (29 CFR 2560.503-1) preempt state M.G.L. c. 93A treble damages, though federal full-and-fair-review rules still require human fiduciary review.
• Licensing vs. quotas: If quotas pressure a licensed adjuster or Massachusetts auto appraiser (212 CMR 2.00) to skip mandated inspections, escalate through internal compliance or consult licensing counsel. This guide offers educational information as of October 29, 2026, not legal advice.

Questions people ask

Key terms

• Massachusetts DOI Bulletin 2024-10: December 9, 2024 directive adopting the NAIC AI Model Bulletin and requiring written AIS governance, vendor oversight, and market-conduct audit readiness.
• Massachusetts DOI Bulletin 2025-02: April 30, 2025 directive requiring image accuracy, physical inspections on inconclusive aerial roof photos, and review of homeowner counter-evidence.
• M.G.L. c. 176D s. 3(9): Massachusetts Unfair Claim Settlement Practices statute requiring reasonable investigation of all available information ((d)) and policy-grounded explanations ((n)).
• Chapter 93A Multiplier (M.G.L. c. 93A s. 9): Massachusetts consumer remedy awarding 2x to 3x damages plus mandatory attorney's fees for willful or knowing c. 176D claims violations.
• Projected Sold Adjustment (PSA): Automated markdown applied to listed comparable-vehicle prices in total-loss software, challenged in the $48 million Volino v. Progressive settlement.
• Rubber-Stamp Discovery: Litigation audit of adjuster screen-dwell timestamps, override rates, and KPI scorecards to test whether human claim review actually occurred.

Sources

1. MA DOI, "Bulletin 2024-10: Use of AI Systems in Insurance" (Dec 9, 2024): Bulletin 2024-10 PDF
2. MA DOI, "Bulletin 2025-02: Aerial Imaging" (Apr 30, 2025): Bulletin 2025-02 PDF
3. NAIC, "Model Bulletin: Use of AI Systems by Insurers" (Dec 4, 2023): NAIC PDF
4. MA General Court, "M.G.L. c. 176D s. 3" (current): c. 176D s. 3
5. MA General Court, "M.G.L. c. 93A s. 9" (current): c. 93A s. 9
6. Mass.gov, "212 CMR 2.00: Auto Damage Appraisal" (current): 212 CMR 2.00
7. Clearinghouse, "Huskey v. State Farm (1:22-cv-07014)" (2024): Case File
8. N.D. Illinois, "Huskey v. State Farm Order" (Sep 11, 2023): Justia Opinion
9. Zelle LLP, "AI Tools and Bad Faith Risk" (JD Supra): Zelle Analysis
10. Insurance Journal, "Progressive $48M Total-Loss Settlement (Volino)" (May 20, 2024): Insurance Journal
11. California DOI, "Bulletin 2022-5: Bias in Claims Practices" (Jun 30, 2022): CDI PDF
12. Rhode Island DBR, "Insurance Bulletin 2025-3: Aerial Images" (Aug 18, 2025): RI Bulletin