I have been working on a number of active State Farm lawsuits involving its claims practices. My research came across a recently filed class action lawsuit in December alleging systemic racial discrimination in its claims processing.1 The press release to the lawsuit indicated the following:
State Farm’s homeowners insurance claims process discriminates against Black policyholders and causes them to incur much greater burdens to have their claims processed and paid out in violation of the Fair Housing Act, according to a class action lawsuit filed today in Illinois. Applying what is believed to be the first ever independent, company-specific statistical analysis of racial disparity in State Farm claims, the suit alleges that the discriminatory impact is driven by series of policies in effect at State Farm that make Black homeowners subject to more scrutiny, delays, and paperwork requests, among other administrative burdens.
‘The data behind this lawsuit indicate that Black homeowners are subject to a fundamentally different and worse claims process at State Farm. State Farm should be a ‘good neighbor’ to all its policyholders, regardless of their race,’ said Aisha Rich of Fairmark Partners, LLP, one of the law firms leading the case.
The complaint alleges that State Farm relies on internal and external automated fraud detection and claims processing tools that use biased historical data and troves of suspect personal consumer data to assign higher risk scores to Black policyholders, resulting in additional scrutiny and delay of their claims. The complaint details specific State Farm vendors, technologies, and practices that likely contribute to their discriminatory treatment of Black policyholders.
‘The complaint alleges that the algorithms, vendors, and data sources that State Farm uses to process and screen claims create disproportionate scrutiny, delay, and burden for Black policyholders. State Farm could examine the algorithmic bias that inflicts enormous harm on Black policyholders, but its failure to do so necessitates this litigation….
A reminder to everybody is that allegations in a lawsuit have to have evidence if the lawsuit is to be successful. This lawsuit alleged in part:
15. In the insurance industry, as elsewhere in our society, racial discrimination has shifted from overt to covert. Even though race-based redlining is now illegal, discrimination has persisted through practices such as using credit-based insurance scores, and discriminatory underwriting guidelines that use age and home value as a proxy for race. Inequitable practices such as these make it more difficult for Black homeowners to build wealth through homeownership at the same rate as white homeowners.
16. Today, discrimination is perpetuated by the modern trend toward automation and data mining. Institutions like insurance companies use algorithmic models to quickly analyze vast troves of publicly available information (‘data mining’) to detect patterns and assist in making future decisions (‘data analytics’). So-called ‘machine-learning’ algorithms are expressly designed to learn based upon the algorithm’s access to a designated data set or an algorithm-driven search for data residing on the internet or in a confined database.
19. As described in more detail below, State Farm’s claims processing methods use algorithmic decision-making that disproportionately subjects the claims of Black policyholders to greater suspicion—and thereby greater administrative process and delay. In this way, State Farm is reproducing and exacerbating existing patterns of race discrimination.
Where did the Class obtain the evidence that the claims practices result in racial discrimination? It seems they did so with claims polling data:
20. In 2021, YouGov—a reputable online polling provider that employs industry best practices and an internal quality assurance process—surveyed about 800 white or Black homeowners with State Farm homeowners insurance policies across the Midwest to assess whether there were racial disparities in State Farm’s homeowners insurance claim submission and adjudication process. Among other data points, the survey collected data on where the homeowners were located; how long their claims process took; whether they were asked to submit additional paperwork after making a claim; how many interactions with a claims handler were required to resolve the claim; and the ultimate outcome of their claim submission.
What about doing this systematically? The complaint alleges:
27. The claims processing aspect of the insurance industry is notoriously opaque, with little oversight or regulation geared toward increasing transparency for consumers and regulators alike. Insurance companies such as State Farm do not typically publish their claims processing guidelines, their claims processing algorithms, their claims outcomes, or other related data. What the industry readily confirms, however, are the basics of the traditional home insurance claim process: A homeowner who has suffered a loss submits a claim; the insurance company assigns an adjuster to assess the loss and apply the wording of the homeowner’s policy to the adjuster’s interpretation of the facts of the case. The adjuster can choose whether to request more information, go to the scene, request other evaluations, or meet with witnesses.
28. Increasingly, however, State Farm and other insurers are automating claims processing, or major aspects of it, and replacing human judgment with algorithms. A significant number of claims are now handled without conversations with the homeowner or follow-up investigation. State Farm instead relies on data analytics and machine-learning algorithms to process the claims it receives. These tools are designed to, among other things, predict the likelihood of fraud and to sort ‘no touch’ or ‘low touch’ claims (which are paid out immediately or near immediately) from ‘high touch’ claims (which trigger additional scrutiny).
29. State Farm does its best to keep the nature of its claims processing methods confidential, asserting that disclosing any details publicly would undermine anti-fraud efforts. But those disclosures that have been made—by State Farm and its third-party vendors—reveal that State Farm harnesses a variety of tools to collect extensive data about policyholders, and to use that data in claims processing and fraud detection. It describes itself as at the ‘cutting edge of analytics’ and it actively recruits and hires employees with a background in data analytics, to ‘turn data into actionable insights by leveraging a combination of Natural Language Processing, Machine Learning, Artificial Intelligence, or other data science tools and concepts.’
What does the complaint allege about how State Farm does this?
32. On information and belief, State Farm uses a combination of internal and third party tools to leverage the vast troves of data it collects to process homeowners insurance claims. Data regarding policyholders and their claims are stored, managed, and accessed primarily through State Farm’s Enterprise Claim System (‘ECS’), a proprietary web-based system used by State Farm claims associates. ECS operates as an electronic file where information about a claim may be found and updated in real time. In addition to providing a web-based interface for the entry and storage of claim data, State Farm executes automated processes based on code and data relationships. As an example, State Farm has disclosed that it licenses a third-party vendor’s proprietary system, Technology Analytics for Claims (‘TAC’), which uses text-based queries of claims data from ECS to help detect and identify claims that might be fraudulent.
33. In fact, State Farm maintains relationships with a variety of third-party vendors that offer software, integrations, and applications focused on insurance claims automation. Especially relevant here, State Farm has a relationship with Duck Creek Technologies, an insurance-specific software and analytics company that provides comprehensive claims management and fraud detection tools.
34. According to Duck Creek’s SEC filings, State Farm was its single largest customer as of fiscal year 2019. On information and belief, State Farm continues to be one of Duck Creek’s largest customers.
35. Per its website, Duck Creek offers a ‘comprehensive claims management solution that helps insurers manage the entire claims lifecycle—from first notice of loss to settlement—in a single integrated solution,’ called Duck Creek Claims. Duck Creek Claims provides ‘end-to end claims workflows that enable high-touch to no-touch claim handling.’ The system is comprised of ‘dynamically-guided workflows, rule-driven automation, personalized user interfaces and data enrichment to automate processes where needed.’ In practice, this means that on receipt of a claim, the system uses predictive modeling or rules-based decision making to determine how a claim should be treated and then structures the claim workflow accordingly. The system’s view of appropriate workflow is impacted by claim type, but also by processing rules (e.g., rules that determine whether a claim is high touch, low touch, or no touch) and by predictive fraud modeling.
38. As explained in an industry presentation on FRISS: ‘Much like the way banks rely on credit scores to assess the risk in lending to an individual, insurance carriers . . . rely on the FRISS Score to assess the likelihood of fraudulent activity in claims.’ A claim with a lower FRISS score is deemed less risky than a claim with a higher risk score. A score of 50 or less, for example, indicates that FRISS has little to no concerns about the claim, and can be quickly reviewed by claims handlers or fast tracked to payment. A score slightly above 50 might indicate there are some anomalies with the claim, and that a claims adjuster should spend a little extra time reviewing these claims and perhaps forward them to a supervisor for a decision before payment can be made. A high FRISS score suggests fraud is likely and the claim should be referred to triage or specific claims adjusters for analysis.
39. According to Duck Creek and FRISS, ‘[a] FRISS Score is generated at every step in the claims process: [first notification of loss], claim contact added, loss details added/changed, before issuance of payment – as well as at any time on demand.’ The software is then capable of auto-generating tasks for the claims adjuster based on a claim’s FRISS Score. As explained in the same industry presentation discussed above, the system is designed such that ‘honest customers receive lightning-quick responses, while questionable claims are guided through further processing with actionable insights and automated workflows’ in the insurer’s claims management system.
Last Spring, I noted that State Farm had discrimination concerns raised in Discrimination Concerns at State Farm.
I will monitor this case. I assume that State Farm will strongly oppose the allegations in this case and try to stifle discovery of its claims practices. However, these secret claims practices are important to understand and probably exist in other insurance companies who tend to share claims practices through vendors and consultants.
Thought For The Day
Three can keep a secret, if two of them are dead.
1 Huskey v. State Farm Fire & Cas. Co., No. 22-cv-7014 (Ill. E.D. [Complaint filed Dec. 14, 2022]).