PRIVILEGED AND CONFIDENTIAL ATTORNEY-CLIENT COMMUNICATION ATTORNEY WORK PRODUCT
Memorandum
I. Executive Summary
You have asked us to analyze whether data outputs, models, scores, indices, and analytics created by Quorum Insurance Holdings Corp. (“Quorum”) using licensed data from Meridian Credit intelligence, Inc. (“Meridian”) as an input constitute “derivative works” under the Copyright Act, 17 U.S.C. § 101 et seq., and to evaluate the parties’ competing intellectual-property positions in the ongoing negotiation of the proposed Master Data License Agreement between Quorum and Meridian (the “MDLA”).
For the reasons set out below, we believe Quorum has the stronger legal position under U.S. copyright law. Under prevailing doctrine, factual data compilations — including credit-file, tradeline, payment-history, commercial-credit, and alternative-data records — receive thin copyright protection at most. Copyright does not protect facts themselves. It protects only original selection, coordination, or arrangement of facts, and even then only to the extent that those elements contain the minimal degree of creativity required by Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991).
Accordingly, outputs generated by Quorum’s proprietary analytical processes — including machine-learning model outputs, blended risk scores, propensity models, fraud indicators, underwriting indices, and enriched datasets — are unlikely to qualify as “derivative works” of Meridian’s underlying data under the Copyright Act. This is especially true where Quorum contributes substantial independent technical, analytical, and business judgment, including model architecture, feature engineering, training methodology, validation logic, weighting decisions, and integration with Quorum’s own first-party claims, policy, telematics, and underwriting data. We refer to these outputs collectively in this memorandum as “Derived Data.”
That said, the practical answer will be driven primarily by contract, not by background copyright law. Because U.S. copyright law provides limited protection for factual databases, and because the law regarding machine-learning model weights and outputs remains developing, the IP allocation provisions in Article VII of the MDLA will likely be the controlling framework between the parties. New York law, which governs the MDLA, generally permits sophisticated commercial parties to allocate ownership and licensing rights by contract, including rights that may go beyond what copyright law would supply by default.
The issue is commercially significant. Quorum intends to use Meridian data as one input into QuorumScore™, Quorum’s proprietary predictive underwriting model. QuorumScore™ is the subject of a pending USPTO application, Serial No. 97/482,116, and is expected to be central both to Quorum’s underwriting advantage and to Quorum’s planned B2B licensing revenue line. Management projections indicate B2B licensing revenue from QuorumScore™ and related analytics of approximately $40–50 million by 2027. Any concession granting Meridian a perpetual license-back, discretionary approval right, or revenue share on QuorumScore™ outputs would materially impair Quorum’s commercial strategy and could dilute the value of Quorum’s independent technology investment.
Meridian’s current position is materially broader than what copyright law supports. Meridian asserts that any data output that “incorporates,” is “based on,” or is “traceable to” Meridian raw data is a derivative work of Meridian’s data and remains subject to Meridian’s ownership and control. Meridian has demanded: (i) a perpetual, royalty-free license-back to all Derived Data; (ii) a prohibition on Quorum selling or sublicensing Derived Data externally unless the data has been “sufficiently transformed,” with Meridian having sole discretion to determine sufficiency; and (iii) a 15% revenue share on net revenues from third-party commercialization of approved Derived Data. We see no copyright-law basis for these demands.
Our principal recommendations are:
1. Quorum should maintain its position that Quorum owns all IP in Derived Data created by or on behalf of Quorum.
2. Quorum should reject Meridian’s proposed perpetual, royalty-free license-back.
3. Quorum should reject Meridian’s proposed “sufficient transformation” standard if Meridian retains sole discretion over the determination.
4. Quorum should reject Meridian’s proposed 15% revenue share.
5. The MDLA should expressly carve out QuorumScore™ and similar Quorum proprietary models, model weights, parameters, inference outputs, scores, indices, and analytics as Quorum-owned IP not subject to any license-back, revenue share, or approval requirement.
6. If a compromise is necessary, Quorum should consider a narrow covenant not to redistribute Meridian raw data or Derived Data in a form from which Meridian raw data can be extracted, rather than accepting Meridian ownership or approval rights over Quorum’s analytical outputs.
II. Background and Scope
A. The Transaction
Quorum is negotiating a five-year enterprise data-licensing arrangement with Meridian. The proposed arrangement is valued at $60,000,000 over the term, consisting of $10,800,000 in Year 1 and $12,300,000 per year in Years 2 through 5. We understand there is also a separate $1,500,000 implementation and onboarding fee under discussion, although that amount has not been the focus of the Article VII IP negotiations.
The licensed Meridian data products are expected to include the following:
Quorum intends to ingest Meridian’s data into its internal data environment and use the data to enhance automated underwriting models, claims-fraud detection algorithms, renewal-pricing tools, customer propensity scoring, lapse-risk modeling, and cross-sell analytics across Quorum’s property and casualty, life, and specialty lines. Quorum’s intended use is not a pass-through resale of Meridian’s data. Rather, Meridian’s data would be one category of input into Quorum’s broader analytics platform, which also includes Quorum’s first-party claims experience, policyholder history, telematics data, underwriting determinations, claims-resolution outcomes, payment and billing data, and actuarial assumptions.
Meridian is a Georgia corporation. It is privately held and majority-owned by Breckenridge Capital Partners Fund IV, L.P., which holds approximately 62.5% of Meridian’s equity. Breckenridge Capital Partners Fund IV is a growth-equity fund managed by Breckenridge Capital Management LLC. We understand that managing partner Elaine Drummond and partner Caleb Rourke oversee Breckenridge’s Meridian investment. Meridian’s investment bank, Whitmore Zanthor & Co., is advising Meridian in connection with a planned 2025–2026 IPO process. Rachel Ng-Castillo, managing director at Whitmore Zanthor, is the principal banker on that engagement.
Meridian is represented in the MDLA negotiations by Kessler Pruitt & Ashe LLP (“KP&A”), with Graham Whitley serving as lead partner.
B. The Derived-Data Dispute
The principal dispute addressed in this memorandum concerns ownership and permitted use of Derived Data created by Quorum using Meridian licensed data as one input.
Quorum’s position is that it should own all intellectual property rights in Derived Data created by Quorum or on Quorum’s behalf using the Meridian data as an input. Quorum has further taken the position that it should have unrestricted rights to use, reproduce, analyze, modify, improve, sublicense, license, distribute, and commercialize such Derived Data, including to third parties outside Quorum’s corporate group, without additional fees, approval rights, or consent from Meridian. Quorum’s position is that Meridian is paid for access to the raw licensed data through the substantial enterprise license fees under the MDLA, while Quorum owns the independent outputs generated through Quorum’s own technology, data assets, and analytics.
Meridian’s position is substantially broader. Meridian insists that any Derived Data that incorporates, is based on, or is traceable to Meridian’s raw data constitutes a “derivative work” under copyright law and remains subject to Meridian’s underlying IP rights. On that basis, Meridian has demanded:
1. Perpetual royalty-free license-back. Meridian would receive a perpetual, irrevocable, worldwide, royalty-free license-back to all Derived Data for Meridian’s own use.
2. External commercialization restriction. Quorum would be prohibited from selling, sublicensing, distributing, or otherwise commercializing Derived Data to any third party unless the Derived Data has been “sufficiently transformed” such that no Meridian raw data can be reverse-engineered from it.
3. Meridian sole discretion. Meridian would have sole discretion to determine whether Derived Data has been “sufficiently transformed.”
4. Revenue share. If Quorum commercializes approved Derived Data externally, Meridian would receive 15% of net revenues attributable to such Derived Data.
The MDLA form delivered by KP&A on November 4, 2022 reflected Meridian’s position. Quorum’s comprehensive redline delivered by C&H on December 2, 2022 rewrote Article VII to vest all IP in Derived Data in Quorum and to remove Meridian’s proposed license-back, sole-discretion approval right, and revenue share. At the first in-person negotiation session in New York on January 18, 2023, the parties formally identified MFN pricing and Derived Data rights as the two principal open issues.
As discussed below, Meridian’s negotiating position appears driven at least as much by commercial and valuation considerations as by legal doctrine. We understand that Breckenridge and Whitmore Zanthor have advised Meridian’s board to resist unrestricted licensee commercialization of Derived Data because such rights may be perceived in IPO due diligence as weakening Meridian’s “data moat.” We further understand that Meridian’s board formally resolved on August 22, 2023 against granting unrestricted Derived Data commercialization rights, influenced by Breckenridge and Whitmore Zanthor. That board resolution is relevant to the negotiation dynamic, but it does not create legal entitlement under the Copyright Act.
C. Scope of This Memorandum
This memorandum addresses the following questions:
1. The copyright-law treatment of factual data compilations, including the effect of Feist and the “thin copyright” doctrine.
2. The likely copyrightability of Meridian’s data products, including MCI Consumer Credit File™, MCI CommercialView™, and MCI AltSignal™.
3. Whether Quorum’s Derived Data constitutes “derivative works” of Meridian’s data products under 17 U.S.C. § 101.
4. The distinction between machine-learning model weights and parameters, on the one hand, and inference or prediction outputs, on the other.
5. Trade secret considerations under the Georgia Trade Secrets Act, O.C.G.A. § 10-1-760 et seq., and the federal Defend Trade Secrets Act, 18 U.S.C. § 1836 et seq.
6. The role of contractual IP allocation as the controlling framework in the MDLA.
7. The legal and commercial weaknesses of Meridian’s proposed license-back, “sufficient transformation,” and revenue-share demands.
8. Recommendations for Quorum’s negotiating position and for revised Article VII drafting.
III. Copyright Analysis: Factual Data Compilations
A. Copyrightability of Factual Data Under *Feist*
The starting point is Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991). In Feist, the Supreme Court rejected the “sweat of the brow” theory of copyright protection for factual compilations. The Court held that facts are not copyrightable, regardless of the labor, expense, or diligence required to collect them. Copyright protects original expression, not facts, discoveries, or information.
The Court explained that factual compilations may be copyrightable only if the author’s selection, coordination, or arrangement of facts is sufficiently original. Even then, the copyright is thin. It extends only to the original manner in which facts are selected or arranged, not to the underlying facts themselves. The Court stated:
“The most fundamental axiom of copyright law is that no author may copyright his ideas or facts. The copyright is limited to those aspects of the work — termed ‘expression’ — that display the stamp of the author’s originality.”
That principle has direct application to credit-bureau and alternative-data products. Individual consumer tradelines, commercial credit records, payment histories, balances, payment dates, account statuses, delinquency indicators, account-open dates, public-record events, trade references, utility-payment records, telecom-payment records, rental-payment records, and similar attributes are factual in nature. They may be commercially valuable, sensitive, regulated, and difficult to assemble, but they are not copyrightable expression.
Meridian may have a copyright in the particular selection, coordination, and arrangement of its large databases, including approximately 310 million consumer records in MCI Consumer Credit File™, approximately 42 million business records in MCI CommercialView™, and approximately 85 million alternative-data records in MCI AltSignal™. However, that copyright would not give Meridian ownership over each constituent fact. Nor would it give Meridian copyright control over every analytical output generated by a licensee that uses those facts as inputs.
This is particularly important because Quorum’s contemplated use is analytic, not expressive copying. Quorum is not seeking to republish the MCI Consumer Credit File™ in Meridian’s format, copy Meridian’s database taxonomy for external distribution, or offer a substitute database containing the same selection and arrangement. Instead, Quorum seeks to use factual data points as inputs into Quorum’s own models and systems.
There is a separate point regarding Meridian’s proprietary scoring algorithms and models, if any. If Meridian licenses outputs from proprietary Meridian scoring models, or if Meridian delivers software, source code, model documentation, or expressive explanatory materials, those items may receive separate copyright or trade secret protection. But based on the MDLA materials we have reviewed, the principal licensed feeds consist of factual compilations and data attributes, not Meridian source code or model architecture.
B. The “Thin Copyright” in Database Compilations
Even where a factual compilation qualifies for copyright protection, the scope of protection is limited. Courts regularly describe database and directory copyrights as “thin” because infringement requires copying protected selection, coordination, or arrangement, not merely using the facts contained in the compilation.
This distinction is critical to the Derived Data dispute. Quorum’s extraction, transformation, and loading of Meridian data into Quorum’s own systems — followed by use of selected factual attributes in proprietary underwriting, fraud, and propensity models — does not, without more, implicate Meridian’s compilation copyright. The relevant act is the use of factual information, not the copying of Meridian’s expressive compilation structure.
Several cases illustrate the point:
• In Assessment Technologies of WI, LLC v. WIREdata, Inc., 350 F.3d 640 (7th Cir. 2003), Judge Posner explained that copyright in a database does not prevent others from obtaining and using the facts contained in the database. The copyright owner could not use copyright to control access to public factual data merely because the data appeared in copyrighted software or a copyrighted compilation.
• In Matthew Bender & Co. v. West Publishing Co., 158 F.3d 674 (2d Cir. 1998), the Second Circuit held that West’s enhancements to judicial opinions, including certain star pagination and citation elements, did not prevent competitors from using unprotectable factual and legal materials. The decision reinforces the principle that copyright protection in factual or legal compilations is narrow.
• In BellSouth Advertising & Publishing Corp. v. Donnelley Information Publishing, Inc., 999 F.2d 1436 (11th Cir. 1993), the Eleventh Circuit held that a telephone directory lacked sufficient originality in selection and arrangement to support infringement claims based on use of business listings. The court applied Feist to reject protection for noncreative factual listings.
• In Key Publications, Inc. v. Chinatown Today Publishing Enterprises, Inc., 945 F.2d 509 (2d Cir. 1991), the Second Circuit recognized that a directory could be copyrightable where there was originality in selecting included businesses and categories, but emphasized that protection extended only to the creative selection and arrangement, not the underlying business information.
The lesson from these cases is that a data owner may protect the organization and expressive structure of a compilation, if sufficiently original, but cannot prevent downstream use of the underlying factual information through copyright alone.
Here, Quorum’s ETL process may involve temporary copying and internal storage of Meridian files, but those acts are licensed under the MDLA for internal use. Once Quorum uses factual attributes from the licensed data to generate scores, model outputs, risk classifications, or enriched internal attributes, the copyright question becomes whether Quorum has recast, transformed, or adapted Meridian’s protectable expression. In most contemplated use cases, it has not.
C. The Database Protection Gap
The United States does not have a sui generis database right comparable to the European Union Database Directive. U.S. copyright law does not protect the investment, labor, or expense of assembling a database as such. The Supreme Court expressly rejected that “sweat of the brow” approach in Feist.
The practical consequence is that Meridian cannot rely on copyright law alone to control all downstream uses of factual data extracted from its compilations. Meridian may rely on contract, confidentiality, trade secret law, data-protection law, and regulatory restrictions, where applicable. But absent contractual restrictions, copyright does not give Meridian a general right to control every score, index, model, or analytical product generated from facts contained in its databases.
This database-protection gap explains why Article VII of the MDLA has become a central negotiation point. Meridian is attempting by contract to obtain control rights that copyright law likely does not supply. That is not unusual in data-licensing transactions; licensors often use contract to protect data assets beyond the baseline provided by copyright. But Quorum should recognize the distinction between Meridian’s legal entitlement and Meridian’s commercial ask. The former is limited; the latter is expansive.
IV. “Derivative Works” Analysis Under 17 U.S.C. § 101
A. Statutory Definition
The Copyright Act defines a “derivative work” as:
“a work based upon one or more preexisting works, such as a translation, musical arrangement, dramatization, fictionalization, motion picture version, sound recording, art reproduction, abridgment, condensation, or any other form in which a work may be recast, transformed, or adapted.”
17 U.S.C. § 101.
The statutory examples are instructive. A translation recasts a literary work in another language. A musical arrangement adapts a musical composition. A dramatization adapts a story into dramatic form. A motion-picture version adapts a preexisting work into film. In each case, the derivative work incorporates protected expression from the preexisting copyrighted work and adds new original expression.
A derivative work therefore requires more than causal dependence. It is not enough that the later work was inspired by, informed by, trained on, or factually enabled by the earlier work. The later work must incorporate protected expression from the preexisting work in some recognizable or legally cognizable form. Where the preexisting work is a factual compilation, the relevant protected expression is not the facts themselves but the original selection, coordination, or arrangement of those facts.
B. Application to Quorum’s Derived Data
Quorum’s Derived Data includes several categories of outputs:
1. Blended risk scores that combine Meridian credit attributes with Quorum’s first-party claims, policy, telematics, actuarial, and underwriting data.
2. Propensity models predicting likelihood of purchase, renewal, lapse, cross-sell conversion, claims frequency, claims severity, or channel preference.
3. Enriched datasets in which Quorum appends internally generated attributes, risk flags, model scores, segmentation categories, or inferred characteristics to Quorum customer, prospect, or policy records.
4. Fraud-detection indicators generated by Quorum’s claims analytics models.
5. QuorumScore™ outputs, including predictive underwriting scores, model-generated classifications, and related confidence or explanation attributes.
In each case, the strongest view is that these outputs are not derivative works of Meridian’s databases. They may be derived from Meridian data in a colloquial or business sense, but they are not “derivative works” in the Copyright Act sense.
The reason is straightforward. Meridian’s protectable copyright interest, if any, lies in the original selection, coordination, or arrangement of its database. Quorum’s Derived Data is not a recasting, transformation, or adaptation of that selection, coordination, or arrangement. Quorum is not turning Meridian’s database into a different expressive format. It is using factual inputs from the database to generate new analytical results.
A simplified example illustrates the distinction. If Quorum copied Meridian’s Consumer Credit File™ and sold a competing credit-file database with substantially the same fields, structure, taxonomy, and record selection, Meridian would have a stronger copyright and contract claim. But if Quorum uses a consumer’s payment-history facts as one input into an underwriting score that also considers claims history, telematics behavior, policy tenure, geography, and actuarial assumptions, the resulting score does not reproduce Meridian’s expressive compilation. It is a new analytical fact or business conclusion generated by Quorum’s own system.
This distinction is consistent with Feist and its progeny. Facts may be used to create new works, including analyses, rankings, models, indices, reports, and scores. Copyright does not give the compiler of the facts a downstream veto over those uses. To hold otherwise would revive the rejected “sweat of the brow” doctrine.
C. Meridian’s “Derivative Work” Argument — Critique
Meridian’s position conflates three distinct concepts:
1. Meridian’s ownership or control of its raw data feeds under contract.
2. Meridian’s possible thin copyright in the selection, coordination, and arrangement of its data compilations.
3. Quorum’s independent ownership of analytical outputs generated using facts from those feeds.
Meridian’s proposed standard — that any output “based on” or “traceable to” Meridian raw data is a derivative work — is not the statutory test. “Traceability” in a causal or technical sense is not a copyright-law standard. Copyright’s derivative-work right requires incorporation of protected expression, not mere use of facts or statistical dependence on factual inputs.
If Meridian’s position were accepted, routine downstream uses of factual data would become derivative works of the data sources from which the facts were obtained. Actuarial tables generated from claims data, underwriting risk assessments generated from credit attributes, public-health indices generated from reported health data, economic forecasts generated from financial data, and market analytics generated from transaction data could all be characterized as derivative works of the underlying datasets. That result would be inconsistent with Feist and would dramatically expand copyright beyond its statutory boundaries.
Meridian may attempt to argue that Quorum’s Derived Data “incorporates” Meridian data because some Quorum outputs may include appended attributes or may preserve transformed versions of original data fields. That argument requires careful factual analysis. If Quorum were to redistribute Meridian raw fields, lightly modified Meridian attributes, or a data product from which Meridian’s raw records can be reconstructed, Meridian would have a stronger contractual concern and potentially a trade secret or confidentiality concern. But that is materially different from claiming ownership over all Quorum-generated scores, model parameters, and analytical outputs.
For that reason, Quorum can reasonably offer a narrow non-redistribution or no-reverse-engineering covenant as a commercial compromise, while rejecting Meridian’s broader ownership and revenue-share claims.
V. Machine-Learning Model Outputs: A Special Case
A. The Technical Dispute
The parties also disagree on how to treat outputs of Quorum’s machine-learning models. The issue has two related but distinct parts:
1. Whether trained model weights and parameters created using training datasets that include Meridian data are “derived data” or “derivative works.”
2. Whether live inference outputs generated by those trained models are “derived data” or “derivative works,” particularly where the live input data may or may not include Meridian data.
At a high level, a machine-learning training process involves exposing a model architecture to a training dataset and iteratively adjusting numerical parameters to minimize prediction error or optimize a defined objective function. Meridian data may be included in the training dataset, but it is not the only input. Quorum’s training environment also includes Quorum’s own first-party claims data, policy data, telematics data, underwriting outcomes, lapse and renewal history, fraud determinations, and other third-party or public data sources.
During training, the model learns statistical patterns and relationships. The trained model weights and parameters are the encoded result of that training process. They are numerical values that influence how the model processes future inputs. They are not a human-readable copy of the training data.
At inference time, the trained model receives new input data and produces predictions, classifications, scores, or other outputs. Depending on the use case, the live inputs may include current Meridian data, Quorum first-party data, applicant-submitted information, telematics data, or non-Meridian third-party data. The output may be a score, probability estimate, risk tier, underwriting recommendation, fraud flag, or similar analytic result.
B. Copyright Analysis of Trained Model Weights
Courts have not yet squarely decided whether machine-learning model weights trained on copyrighted works constitute derivative works of the training data. The law is therefore unsettled. Nonetheless, the stronger view, particularly in the context of factual data, is that trained model weights are not derivative works of the training data.
Model weights are numerical parameters. They encode statistical relationships learned during training. They do not ordinarily reproduce, display, or incorporate underlying training records in recognizable form. Even if the weights are “based on” the training data in a causal sense, causal dependence is not enough under 17 U.S.C. § 101. The derivative-work right requires recasting, transformation, or adaptation of protected expression.
That point is especially strong here because Meridian’s protectable expression is limited. The training data consists primarily of factual data attributes. To the extent the model learns correlations between payment history, credit utilization, claims frequency, and underwriting outcomes, it is learning relationships among facts. Copyright does not protect those facts or relationships.
Meridian may contend that trained weights reflect the statistical imprint of Meridian’s database and therefore should be treated as Meridian-controlled. That argument sounds more in contract, confidentiality, or trade secret policy than in copyright. It does not map comfortably onto the statutory derivative-work framework.
This analysis should be stated with appropriate caution. The treatment of AI and machine-learning systems under copyright law is actively developing, and future courts may adopt different approaches in cases involving expressive training data, memorization, or output substitution. But for the specific MDLA context — factual credit and alternative-data feeds used as one component in Quorum’s proprietary insurance analytics models — the weight of analogous authority supports Quorum’s position that trained weights are not derivative works of Meridian’s databases.
C. Copyright Analysis of Inference Outputs
Inference outputs generated using new, non-Meridian inputs are even further removed from Meridian’s data. In that scenario, the trained model applies learned numerical parameters to entirely new input data and produces a score, prediction, or classification. The output does not incorporate Meridian raw data, and it does not incorporate Meridian’s selection, coordination, or arrangement.
Accordingly, inference outputs generated from non-Meridian inputs should not be considered derivative works of Meridian data under any reasonable interpretation of the Copyright Act.
Where inference outputs are generated using Meridian data as current live inputs — for example, where Quorum retrieves current credit attributes from Meridian and uses them to produce a real-time underwriting score — the analysis is slightly closer but still favors Quorum. The output uses Meridian factual data as input, but transforms that input through Quorum’s proprietary algorithms into a fundamentally different analytic product. A credit attribute is not being republished as a credit attribute; it is being used as one variable in a broader underwriting model.
Under Feist, using facts to create new analytical outputs is not the same as creating a derivative work of the compilation containing those facts. The result is not an adaptation of Meridian’s expressive selection or arrangement. It is Quorum’s model-generated conclusion.
D. Implications for QuorumScore™
QuorumScore™ is the most important practical application of this analysis. As we understand it, QuorumScore™ is Quorum’s proprietary predictive underwriting model that blends Meridian credit data with Quorum’s own first-party claims, policy, and telematics data. Quorum has filed a USPTO application for QuorumScore™, Serial No. 97/482,116.
Under the analysis above, QuorumScore™ outputs — including trained model parameters and live scores — should not be treated as derivative works of Meridian’s data under copyright law. QuorumScore™ reflects substantial independent creative and analytical contribution by Quorum’s data-science team, including:
• model architecture and selection;
• feature engineering and feature-selection methodology;
• training-set construction and weighting;
• hyperparameter tuning;
• validation and back-testing protocols;
• model-governance procedures;
• bias and adverse-impact testing;
• integration of claims, policy, and telematics data;
• underwriting interpretation and business-rule overlays; and
• deployment methodology across Quorum’s insurance lines.
Those elements are Quorum’s original work product and may also constitute Quorum trade secrets. Granting Meridian a license-back or revenue share on QuorumScore™ outputs would effectively allow Meridian to appropriate value created by Quorum’s independent investment. Copyright law does not compel that result, and Quorum should not accept it absent a deliberate commercial decision supported by equivalent economic concessions.
VI. Trade Secret Considerations
A. Meridian’s Potential Trade Secret Arguments
Even if copyright law does not support Meridian’s position, Meridian may argue that its data feeds contain trade secrets. Potential theories include that Meridian’s feeds reflect proprietary data-collection methodologies, confidential source relationships, non-public commercial information, proprietary data-cleaning processes, scoring logic embedded in data products, or unique compilations not generally known to competitors.
Under the Georgia Trade Secrets Act, O.C.G.A. § 10-1-760 et seq., and the Defend Trade Secrets Act, 18 U.S.C. § 1836 et seq., Meridian would need to show that the relevant information derives independent economic value from not being generally known or readily ascertainable by proper means and that Meridian takes reasonable measures to maintain its secrecy.
Some aspects of Meridian’s business may satisfy that standard. Meridian’s source relationships, data-ingestion pipelines, data-normalization logic, matching methodology, quality-control rules, fraud-screening techniques, and proprietary scores may have trade secret value. However, much of the factual information in credit files — tradeline data, payment histories, balances, account statuses, utility-payment data, telecom-payment data, rental-payment data, and commercial trade references — originates from third-party data furnishers or public/commercial sources. The individual facts are not necessarily “secret” in the trade secret sense merely because Meridian aggregates them.
Trade secret law also does not automatically give Meridian ownership of outputs independently generated by Quorum after authorized access to the data. If Quorum obtains the data under license and uses it within the permitted scope, the question becomes contractual. Meridian can require confidentiality and use restrictions for Meridian raw data. But Quorum’s analytical outputs, particularly where they do not disclose Meridian raw data or proprietary methodologies, are properly characterized as Quorum’s own confidential information and trade secrets.
B. Quorum’s Trade Secret Rights in Derived Data
Quorum has strong trade secret interests in its Derived Data, particularly QuorumScore™ model architecture, parameters, feature-engineering logic, validation results, deployment rules, and outputs. These materials derive independent economic value from not being generally known. Quorum also appears to take reasonable measures to protect them, including limiting access within its data-science and underwriting teams, maintaining secure model-development environments, and treating QuorumScore™ materials as proprietary.
Quorum’s data-science team contributes substantial proprietary know-how to the development of Derived Data. That know-how includes:
• determining which data fields have predictive value in insurance underwriting;
• constructing derived features from raw data inputs;
• weighting features by line of business, geography, product, and customer segment;
• integrating first-party claims and policy data with external credit and alternative data;
• designing model architectures and ensembles;
• tuning and validating models for predictive accuracy and regulatory defensibility;
• documenting model-governance procedures; and
• translating model outputs into underwriting and pricing workflows.
A perpetual, royalty-free license-back to Meridian would undermine those trade secret interests. Depending on its scope, the license-back could require Quorum to disclose sensitive outputs or model-derived attributes to Meridian, a third-party data analytics company that may serve other insurers or develop competing insurance-risk products. Even if Meridian were subject to confidentiality obligations, disclosure of broad categories of Derived Data would increase leakage risk and could complicate Quorum’s ability to maintain trade secret protection.
For that reason, Quorum should treat the license-back demand not merely as an economic issue but as a core trade secret and competitive-position issue.
VII. Contractual IP Allocation: The Controlling Framework
A. Primacy of Contract Over Copyright in Data Licensing
Because copyright protection for factual compilations is thin, and because the treatment of machine-learning weights and outputs remains developing, Article VII of the MDLA will likely be the decisive source of rights between Quorum and Meridian.
Under New York law, sophisticated commercial parties generally may allocate ownership and licensing rights by contract. Courts applying New York law will enforce clear provisions assigning ownership, reserving rights, granting licenses, imposing confidentiality obligations, and restricting downstream use. In the data-licensing context, contract often supplies the primary protection because background IP law does not fully protect factual datasets.
Both parties’ conduct confirms this. Meridian is not relying solely on copyright law; it has proposed detailed contractual controls, including license-back rights, transformation approval, and revenue sharing. Quorum likewise has proposed a detailed Article VII redline vesting Derived Data rights in Quorum. The negotiation is therefore about contractual risk allocation, not simply about what the Copyright Act would provide by default.
B. Quorum’s Proposed Article VII — Strengths
Quorum’s proposed Article VII, drafted by C&H and included in the December 2, 2022 comprehensive redline, vests all IP in Derived Data in Quorum and grants Quorum unrestricted commercialization rights. That position is legally defensible and commercially important.
The copyright analysis supports Quorum’s proposed allocation. Meridian has no copyrightable interest in individual factual data elements. Meridian’s compilation copyright, if any, does not extend to Quorum’s independent analytical outputs. Quorum contributes the relevant independent intellectual and technical value through its models, data assets, and business logic.
Quorum’s proposed allocation is also consistent with common market practice in enterprise data-licensing transactions where the licensee contributes substantial independent analytics. Although outcomes vary by industry and bargaining leverage, licensees commonly insist that they own outputs, scores, models, analyses, and aggregated insights generated from licensed factual data, particularly where the licensor is already compensated through significant license fees and where the licensee’s outputs do not disclose or redistribute raw licensed data.
Omar Siddiqui’s separate market survey memorandum on license-back provisions addresses this point in more detail. In brief, we understand that license-backs are more common where the licensor has commissioned custom analytics or where the licensee is acting as a service provider for the licensor. They are less common where an enterprise customer uses licensed factual data to improve its own proprietary products and risk models.
C. Meridian’s Proposed Provisions — Risks to Quorum
Meridian’s proposed Article VII creates significant legal and commercial risks for Quorum.
Perpetual royalty-free license-back. Meridian’s proposed license-back would give Meridian continuing access to Quorum’s proprietary analytical outputs, potentially including QuorumScore™. Because the license is perpetual and royalty-free, Meridian would obtain the benefit of Quorum’s technology investment without paying additional consideration. The provision could allow Meridian to use Quorum outputs to improve Meridian’s own data products, develop insurance-sector analytics, or license enhanced products to Quorum’s competitors. That result would be commercially unreasonable and inconsistent with Quorum’s need to protect its trade secrets.
“Sufficient transformation” standard with Meridian sole discretion. Meridian’s proposed transformation standard is flawed in several respects. First, it lacks objective criteria. Second, it gives Meridian unilateral control over Quorum’s ability to commercialize Quorum-created IP. Third, it creates indefinite uncertainty for Quorum’s business planning and revenue projections. Fourth, whether raw data can be reverse-engineered from transformed outputs is a technical question that should be evaluated objectively, not determined by Meridian in its sole discretion. The separate memorandum critiquing the “sufficient transformation” standard addresses potential technical alternatives, including objective de-identification, aggregation, suppression, and reconstruction-risk testing.
15% revenue share. Meridian’s proposed revenue share would materially reduce the economics of Quorum’s planned B2B licensing strategy. Against Quorum’s projected $40–50 million B2B revenue by 2027, a 15% revenue share would transfer approximately $6–7.5 million annually to Meridian. That amount would be paid on top of the $12.3 million annual license fee in Years 2 through 5 and the overall $60 million MDLA commitment. There is no copyright-law basis for this payment. It is a commercial toll on Quorum’s independent analytics.
Taken together, Meridian’s proposed provisions would convert a data-input license into a continuing participation right in Quorum’s proprietary analytics business. Quorum should resist that structure.
VIII. Evaluation of Meridian’s Likely Counter-Arguments
A. Investment-Based Arguments
Meridian is likely to argue that it has invested heavily in assembling approximately 310 million consumer records, 42 million business records, and 85 million alternative-data records, and that this investment justifies downstream control over Derived Data.
That argument has commercial force but limited copyright significance. Under Feist, investment in data collection is not protected by copyright as such. If Meridian wants to protect its investment, it must do so through contract, confidentiality, trade secret measures, technological controls, and pricing. The MDLA license fees already compensate Meridian for providing access to the data. The contemplated $60 million five-year commitment, plus the $1.5 million implementation fee, is substantial consideration.
A license-back and revenue share would amount to double recovery. Meridian would be paid once for access to its raw data and then again for Quorum’s independently developed outputs. Unless Meridian makes a corresponding commercial concession, such as a material fee reduction, exclusivity, enhanced data rights, or expanded service commitments, there is no obvious economic justification for that structure.
B. IPO-Valuation Arguments
Meridian may also argue that unrestricted Derived Data rights could undermine its IPO valuation. We understand that Breckenridge Capital Partners, as 62.5% owner, and Whitmore Zanthor & Co., through managing director Rachel Ng-Castillo, have advised Meridian’s board to resist unrestricted Derived Data commercialization rights because such provisions could be perceived in IPO diligence as weakening Meridian’s “data moat.” We further understand that Meridian’s board formally resolved on August 22, 2023 against granting unrestricted Derived Data commercialization rights, influenced by Breckenridge and Whitmore Zanthor.
These are legitimate business concerns for Meridian, but they are not legal grounds for asserting IP rights that do not exist under copyright law. Quorum can acknowledge Meridian’s need to protect raw data while preserving Quorum’s ownership of independent analytics. The MDLA can include robust confidentiality provisions, raw-data use restrictions, security controls, audit rights, and restrictions on redistribution of Meridian raw data. Those provisions protect Meridian’s data moat without granting Meridian ownership or control over QuorumScore™ or other Quorum-created outputs.
Quorum should avoid allowing IPO narrative concerns to be reframed as copyright entitlement. The distinction matters in negotiation. Meridian may choose not to grant certain rights for business reasons, but it should not be permitted to characterize Quorum’s independent analytics as Meridian-owned derivative works as a matter of law.
C. “Traceability” as a Legal Standard
Meridian’s “traceability” formulation is particularly vulnerable. The Copyright Act does not ask whether a later work is traceable to a prior factual input. It asks whether the later work recasts, transforms, or adapts protectable expression from the preexisting work.
Traceability is also an unstable practical standard. Many analytic outputs are traceable in some causal sense to many inputs. A risk score may be traceable to credit data, claims history, telematics, actuarial tables, applicant answers, geospatial data, public records, and internal underwriting judgments. That does not make the output a derivative work of each source. If traceability were the test, virtually every analytic product would be subject to overlapping downstream ownership claims by every data supplier whose facts contributed to the analysis.
That result would be commercially unworkable and legally inconsistent with Feist. Quorum should reject “traceability” as an IP ownership standard. If the parties need a technical protection standard, it should focus on whether Meridian raw data is redistributed or reasonably reconstructable from Quorum’s external products, not whether an output is causally informed by Meridian data.
IX. Recommendations
A. Negotiating Position on Derived Data IP Ownership
We recommend that Quorum take the following positions in the MDLA negotiations.
1. Maintain Quorum’s position that Quorum owns all IP in Derived Data. The copyright analysis strongly supports Quorum’s position. Quorum should not concede ownership of Derived Data, model outputs, model weights, scores, indices, enriched attributes, or analytics created by Quorum or on Quorum’s behalf.
2. Reject the perpetual royalty-free license-back. Meridian’s license-back demand has no meaningful copyright-law basis and would compromise Quorum’s trade secrets and competitive position. If a business compromise becomes necessary, Quorum could consider a narrow, limited-purpose license-back solely for Meridian’s internal data-quality improvement purposes, subject to strict confidentiality, no external use, no product-development use, no competitive use, no sublicensing, and no access to QuorumScore™ model logic, weights, parameters, or customer-level outputs. That should be treated as a last-resort concession, not a starting point.
3. Reject Meridian’s “sufficient transformation” standard with sole discretion. Quorum should not accept a standard under which Meridian alone decides whether Quorum may commercialize Quorum-created IP. If any transformation standard is included, it should be objective, measurable, and subject to joint determination or neutral third-party resolution. The relevant question should be whether Meridian raw data is disclosed or reasonably reconstructable, not whether Meridian believes the output is commercially derivative of its data.
4. Reject the 15% revenue share. There is no copyright-law basis for Meridian’s proposed 15% revenue share. Based on Quorum’s projected 2027 B2B revenue of $40–50 million, the revenue share would transfer $6–7.5 million annually to Meridian, on top of the $12.3 million annual license fee. This would materially impair the economics of QuorumScore™ commercialization and should be rejected.
5. Protect QuorumScore™ explicitly. The MDLA should include an express carve-out recognizing QuorumScore™ and similar proprietary models as Quorum’s exclusive IP. The carve-out should cover model architecture, source code, object code, algorithms, features, feature weights, parameters, trained models, model documentation, validation materials, inference outputs, risk scores, classifications, explanations, and related analytics. These assets should not be subject to any license-back, revenue share, transformation approval, or post-termination destruction obligation.
B. Contractual Drafting Recommendations
We recommend revising Article VII to include the following concepts.
1. Clear definition of “Licensed Data.” The agreement should define Meridian’s “Licensed Data” as the raw data feeds, files, records, attributes, documentation, and related materials provided by Meridian under the MDLA, including MCI Consumer Credit File™, MCI CommercialView™, and MCI AltSignal™.
2. Clear definition of “Derived Data.” “Derived Data” should be defined to include all data, outputs, scores, indices, models, analyses, insights, predictions, classifications, enriched attributes, aggregated data, de-identified data, statistical results, model weights, parameters, and other materials created by or on behalf of Quorum using Licensed Data as one input, excluding Meridian raw data in the form provided by Meridian.
3. Affirmative Quorum ownership clause. Article VII should state that, as between the parties, Quorum exclusively owns all right, title, and interest in and to Derived Data and all IP rights therein.
4. Meridian acknowledgment. Meridian should acknowledge that it has no ownership interest in Derived Data and no right to use, access, license, sublicense, distribute, commercialize, or exploit Derived Data except as expressly provided in the MDLA.
5. No reverse-engineering covenant as compromise. If compromise language is needed, Quorum could agree not to distribute externally Meridian raw data or Derived Data in a form from which Meridian raw data can reasonably be extracted or reconstructed. That covenant should be objective and should not give Meridian sole discretion.
6. Post-termination treatment. Quorum should be permitted to retain and continue using all Derived Data indefinitely after termination or expiration of the MDLA. The current MDLA draft requires Quorum to return or destroy Meridian raw data within 90 days after termination. That obligation should not apply to Derived Data, QuorumScore™, model weights, parameters, scores, outputs, or analytics.
7. Express treatment of ML assets. The definitions should expressly address trained model weights, parameters, embeddings, feature transformations, inference outputs, prediction scores, classifications, confidence scores, and related analytic outputs to avoid later ambiguity.
8. No implied license-back. The agreement should state that no license to Derived Data is granted to Meridian by implication, estoppel, or otherwise.
9. Confidentiality alignment. The confidentiality provisions should separately protect both parties’ confidential information: Meridian’s raw licensed data and Quorum’s models, Derived Data, QuorumScore™ materials, and analytics.
C. Coordination with Related Workstreams
Several related workstreams should be coordinated with the Article VII negotiation.
First, Quorum should coordinate with Caldwell & Strauss LLP, particularly Angela Firth, Quorum’s regular outside IP counsel, on QuorumScore™ patent and trade secret protection. The MDLA should not undermine Quorum’s pending USPTO application or create disclosure obligations inconsistent with Quorum’s patent and trade secret strategy.
Second, Quorum should coordinate with David Cartwright and Lisa Johannsen on CCPA/CPRA implications for Derived Data commercialization. A separate memorandum on privacy and data-protection issues is pending. The privacy analysis may affect permissible external commercialization structures, de-identification requirements, consumer notice obligations, and restrictions on sharing or selling personal information.
Third, Quorum should coordinate with Marcus Weill on Standish National Bank covenant implications. The $60 million MDLA license value exceeds the $25 million threshold in Section 7.12(b) of the Amended and Restated Credit Agreement dated June 15, 2021 by $35 million. In addition, any characterization of Derived Data as licensed rather than owned IP could complicate lender consent analysis, collateral descriptions, and covenant compliance.
Fourth, Quorum should consider the implications of the Pinnacle Ridge Advisory LLC pricing benchmark study delivered February 14, 2023. We understand Franklin Osei, managing director at Pinnacle Ridge, led that study. The benchmark analysis may provide useful leverage in the broader commercial negotiation, particularly if Meridian continues to demand revenue participation in addition to above-market or market-clearing license fees.
X. Conclusion