AI as IP™ Framework
A Practical Guide to Classify, Protect, and Monetize AI Assets (Part I of II)
AI serves as a double-edged sword, presenting economic risks and the potential to disrupt and harm various industries, while simultaneously enabling significant innovation and growth across many sectors. Yet, for small and mid-sized enterprises (SMEs), the economic value of AI systems is often invisible in financial reporting, lacks legal protections, and is under-leveraged in financing and valuation efforts. In this two-part article, the authors provide a practical guide that builds on earlier articles in the AI as IP™ series and provides SMEs with a step-by-step framework for identifying, classifying, protecting, and monetizing AI assets.
Executive Summary
Artificial Intelligence (AI) serves as a double-edged sword, presenting economic risks and the potential to disrupt and harm various industries, while simultaneously enabling significant innovation and growth across many sectors. Yet for small and mid-sized enterprises (SMEs), the economic value of AI systems is often invisible in financial reporting, lacks legal protections, and is under-leveraged in financing and valuation efforts. This practical guide builds on earlier articles in the AI as IP™ series and provides SMEs with a step-by-step framework for identifying, classifying, protecting, and monetizing AI assets.
AI assets can be grouped into five primary categories: Training Data, Model Assets, Algorithmic Frameworks, Computational Infrastructure, and Deployed Applications. In the following sections, we explain how to recognize the asset, protect it under appropriate IP strategies, and extract measurable economic returns. We conclude by positioning AI as a new form of enterprise capital—an identifiable, controllable, and monetizable intangible asset that should be listed on balance sheets—and by providing SME executives with a “how-to” manual for converting AI innovation into defensible value and investor confidence.
Introduction: Turning AI from Expense into Capital
In recent years, AI has changed how businesses create, deliver, and capture value.[1] And yet most SMEs continue to treat AI investments as an operational expense rather than strategic capital.[2] This mindset obscures the true value of AI projects, limits their protection under law, and diminishes access to financing or M&A acquisitions.
Larger corporations have the capital to retain IP attorneys, accountants, and data scientists with the express purpose of managing AI assets as a part of their broader intangible capital portfolio. SMEs rarely have this luxury.[3] Instead, they need to adopt a focused, efficient approach towards keeping track of AI assets, securing appropriate protections, and establishing systems for ongoing valuation and governance.
When properly classified and protected, AI assets can meet international accounting standards for recognition as intangible property under IAS 38, satisfying the tests of identifiability, control, and future economic benefit.[4]
AI systems can then be amortized, insured, licensed, and sold, thereby capitalizing on innovation from a sunk cost into balance-sheet strength. This shift also enables SMEs to speak the same financial language as investors, acquirers, and regulators, bridging the current AI recognition gap between market value and book value.[5]
AI Asset Framework for SMEs
AI components should be classified into discrete, ownable categories. Without a standardized taxonomy, protection and valuation can be inconsistent and incomplete. The following tables can serve as a practical “AI Asset Framework” for SMEs, comprised of five primary categories and supported by five secondary categories:
Table 1: Primary Asset Categories
|
Training Data Assets |
Curated datasets, inputs, and data pipelines forming the raw material of AI. |
|
Model Assets |
Trained architectures, weights, embeddings, and fine-tuned networks that convert data into value. |
|
Algorithmic Frameworks |
Reusable code logic, optimization routines, and orchestration layers that define how AI operates. |
|
Computational Infrastructure |
Hardware, middleware, and orchestration systems that enable scalable AI performance. |
|
Deployed Applications |
User-facing implementations that deliver AI-driven value to customers, employees, or partners. |
Table 2: Secondary Asset Categories
|
Synthetic Data Generators |
AI systems that create privacy-preserving or domain-balanced datasets. |
|
Prompt Libraries and Retrieval-Augmented Generation (RAG) Architectures |
Retrieval and interaction layers linking language models with knowledge bases. |
|
Evaluation and Benchmarking Systems |
Mechanisms for measuring model accuracy, fairness, and explainability. |
|
Autonomous Agent Frameworks |
An orchestration system enabling AI to act independently or collaborate with other agents. |
|
AI Governance and Risk Models |
Protocols ensuring responsible and compliant AI development and deployment. |
These 10 categories, when combined, can constitute an AI Capital Stack™, a conceptual model for organizing the technological, legal, and financial layers of AI enterprise value. Each layer supports the succeeding layer: data feeds models, models rely on algorithms, algorithms that require infrastructure, and all converge into products deployed for business and consumers.
Understanding this hierarchy of assets can help SMEs map each technical component to specific IP protections. For example, copyright for code, patents for innovations, trade secrets for proprietary processes, trademarks for deployed products, and contracts for legal enforcement. The following sections describe each primary AI asset category in detail and with examples, supporting relationships, and recommended protection strategies.
Training Data Assets
Definition and Strategic Importance
Training Data is the foundational layer of any AI system: it defines the boundaries of what a model can learn, predict, or create.[6] Data quality and exclusivity often represent the single most defensible differentiator for SMEs. Proprietary datasets, which are often collected from business operations, customers, and partners, can constitute an enduring competitive advantage with the proper protections and utilization.
Data becomes commercially valuable in three ways: uniqueness, curation, and continuity.[7] Uniqueness differentiates proprietary data from legally accessible sources. Good curation can help ensure the data is clean, labeled, and relevant. Continuity refers to collecting and updating data, which can help to maintain its relevance and economic value over time. Together, these qualities can meet the accounting tests of identifiability and control required under IAS 38, thereby qualifying Training Data as a recognizable intangible asset.
Table 3: Training Data—Practical SME Examples
|
Retail Analytics SME |
A mid-sized retail data firm compiles multi-year point-of-sale data across independent stores, cleaning and labeling it to reveal hyperlocal purchasing trends. The dataset is then licensed to national chains. |
|
Healthcare Startup |
A diagnostic firm builds anonymized patient datasets that integrate radiology and genomic data under the Health Insurance Portability and Accountability Act (HIPAA) compliant governance. This dataset serves as the foundation for an AI disease prediction engine. |
|
Manufacturing Company |
A family-owned manufacturer deploys IoT sensors to collect equipment telemetry data. The dataset supports predictive maintenance analytics, reducing downtime, and enabling future licensing to other facilities. |
Table 4: Training Data—Supporting Categories
|
Synthetic data generators may be carefully considered for augmenting sparse or sensitive datasets while preserving privacy. |
|
Prompt Libraries and RAG Architectures enable better data retrieval for model retraining. |
|
AI Governance Models document compliance, audit trails, and consent provenance. |
Table 5: Training Data—Protection Strategies
|
Copyright |
Protects creative labeling schemas, annotations, and expressive database organization. |
|
Trade Secrets |
Safeguards proprietary curation processes, preprocessing pipelines, and data lineage documentation. |
|
Contracts |
Data-sharing agreements and non-disclosure agreements (NDAs) restrict unauthorized replication or derivative use. |
|
Patents (Limited) |
Occasionally applies to novel data processing systems or anonymization techniques. |
Practical Advice for SMEs
SMEs should treat Training Data as an owned asset; maintain metadata logs, version histories, and a chain-of-custody record to establish provenance. Use restricted-access repositories with encryption and watermarking, and require third-party contributors to sign data contribution and confidentiality agreements with respect to ownership rights. Whenever feasible, secure data insurance or indemnity coverage for breach of contract or misuse exposures.
Model Assets
Definition and Strategic Importance
Model Assets include trained neural networks, embeddings, weights, and architectures that apply data to provide actionable insight.[8] They embody both creative design and empirical learning, representing a hybrid of software and proprietary know-how. For some SMEs, these assets can represent high revenue potential through licensing opportunities, Software as a Service (SaaS) platforming, and embedded analytics for client products.
Model Assets can meet all four accounting criteria for intangible recognition: they are identifiable (distinct from hardware), controlled (via access credentials or hosting environments), measurable (based on reproducible performance), and yield future economic benefit (through use or licensing). The value of Model Assets can scale exponentially when integrated with proprietary data or domain-specific tuning.[9]
Table 6: Model Assets—Practical SME Examples
|
FinTech Company |
A regional lender develops an AI model that predicts creditworthiness based on behavioral and transactional data, outperforming conventional FICO scoring. |
|
Legal Technology |
Builds a Natural Language (NLP) Model for automatic contract clause comparison, fine-tuned on proprietary case data. |
|
Logistics SME |
Trains a predictive model that forecasts delivery delays using weather and GPS inputs, later licensed to multiple courier partners. |
Table 7: Model Assets—Supporting Categories
|
Evaluation Systems benchmark accuracy, latency, and robustness. |
|
Prompt Libraries enhance interpretability and explainability. |
|
Governance Models document retraining protocols, performance thresholds, and ethical use policies. |
Table 8: Model Assets—Protection Strategies
|
Patents |
Protects model code and documentation. |
|
Trade Secrets |
Covers model weights, hyperparameter tuning, and fine-tuning datasets. |
|
Copyright |
Protects model code and documentation. |
|
Contracts |
Used for API-based access licensing, inference-as-a-service, and model performance Service Level Agreements (SLAs). |
Practical Advice for SMEs
SMEs should retain internal ownership of trained model weights and deploy them via secure APIs or on-premise containers, rather than through direct code transfers; file provisional patents for new model architectures or inference efficiencies to preserve early rights and defer costs; and document version control using reproducible training logs. If models are fine-tuned from open-source bases, SMEs should maintain compliance records to provide derivative legitimacy.
Algorithmic Frameworks
Definition and Strategic Importance
Algorithmic Frameworks represent the “logic layer” behind AI: the reusable code, optimization engines, and orchestration modules that determine how models are trained and deployed.[10] For SMEs, frameworks may be the most scalable and licensable form of IP because they have the potential to be applied across multiple industries.
These frameworks are often the foundation for SaaS offerings, enabling modular configuration and integration with client systems. They can also embody operational experience; the heuristics, efficiency improvements, and workflows refined through years of practice and development. As such, they can be technically and economically defensible as tangible assets.
Table 9: Algorithmic Frameworks—Practical SME Examples
|
Quantitative Trading |
Develops a proprietary reinforcement learning engine that autonomously adjusts investment weights in real-time. |
|
E-Commerce |
Utilizes an adaptive pricing algorithm that balances profitability with customer retention. |
|
Energy Technology |
Implements an AI control system for smart grid load management, reducing energy volatility through predictive modeling. |
Table 10: Algorithmic Frameworks—Supporting Categories
|
Evaluation Systems validate algorithm performance, speed, and interpretability. |
|
Autonomous Agent Frameworks extend the logic to dynamic multi-agent systems. |
|
Governance Models ensure transparency and compliance with explainability standards. |
Table 11: Algorithmic Frameworks—Protection Strategies
|
Copyright |
Safeguards source code implementations and related documentation. |
|
Trade Secrets |
Protects unique parameter-tuning processes, heuristics, and efficiency improvements. |
|
Contracts |
Defines ownership in collaborative development and restricts redistribution in SaaS models. |
|
Patents |
Secures novel algorithmic methods, optimization strategies, and orchestrations. |
Practical Advice for SMEs
Document an algorithm’s evolution: the inputs, outputs, revisions, and responsible engineers. Apply version control with immutable audit trails. For complex systems, file defensive patents covering key methods and preserve undisclosed elements as trade secrets. Use clear employment IP assignment agreements to ensure company ownership of algorithms. To maximize return on investment (ROI), periodically assess algorithmic frameworks for potential reuse across new products.
Computational Infrastructure
Definition and Strategic Importance
Computational Infrastructure encompasses the full technology environment that enables AI functionality—from physical hardware to the orchestration layers that manage deployment and deliver results.[11] This can represent a hidden yet critical asset for SMEs: proprietary configurations, cost optimization architecture, or security mechanisms that distinguish their AI capabilities from commodity solutions.
Unlike general IT systems, AI infrastructure includes data pipelines, model serving frameworks, GPU orchestration, and edge deployment architectures.[12] Each can contribute directly to the operational value and performance of AI solutions. When properly documented, such systems can meet the accounting tests of identifiability and control under IAS 38,[13] and can thereby qualify as intangible assets.
SMEs may be prone to underappreciating their infrastructure value because they are embedded in technical operations rather than IP filings. Yet distinctive systems architecture—when it enables new levels of efficiency, compliance, or performance—can be protected and monetized.
Table 12: Computational Infrastructure—Practical SME Examples
|
Edge AI Integration |
A mid-sized manufacturer develops an edge inference pipeline running machine-vision models locally on IoT devices, reducing latency and bandwidth costs. |
|
Hybrid Cloud Architecture |
A health analytics SME builds a privacy-preserving cloud platform ensuring that patient data never leaves its secure domain while still supporting AI model training. |
|
Energy Optimization Platform |
A data center management company uses AI-driven GPU orchestration to balance computational load and reduce energy consumption. |
Table 13: Computational Infrastructure—Supporting Categories
|
Autonomous Agent Frameworks automate workload scheduling and scaling. |
|
Evaluation Systems benchmark throughput, latency, and resilience. |
|
AI Governance and Risk Models oversee system reliability and data security compliance. |
Table 14: Computational Infrastructure—Protection Strategies
|
Copyright |
Protects infrastructure code, orchestration scripts, and system design documentation. |
|
Trade Secrets |
Safeguards architecture diagrams, scaling algorithms, and performance tuning parameters. |
|
Contracts |
Cloud service agreements, uptime SLAs, and cybersecurity clauses define rights and liabilities. |
|
Patents |
Protects innovations in orchestration methods, containerization workflows, or system performance optimizations. |
Practical Advice for SMEs
SMEs should document every infrastructure component as part of an “AI Architecture Register” to ensure systematic documentation and inventory of AI assets.[14] This can include diagrams, data flows, and deployment methods. SMEs should also access logs and cybersecurity protocols as evidence of trade secret diligence, and file provisional patents for proprietary orchestration or automation tools that demonstrably improve efficiency or compliance.
SMEs should also consider insurance coverage for business interruptions or cyber liabilities to protect against outages or breaches that could compromise AI operations. To further support Environmental, Social, and Governance (ESG) reporting, SMEs can also establish monitoring dashboards that track performance, cost, and environmental impact.
Computational Infrastructure, when treated as an IP asset, enables reliable AI delivery and can signal to investors that the organization possesses mature operational control, which can be a key indicator of scalability and valuation readiness.[15]
Deployed Applications
Definition and Strategic Importance
Deployed Applications are the visible interfaces through which users utilize AI.[16] They apply the upstream components—the data, algorithms, and infrastructure—into commercial outcomes. These applications can often be the primary source of revenue for SMEs and the most publicly recognizable form of IP.
Examples of Deployed Applications can include conversational bots, predictive dashboards, generative design tools, and embedded AI modules. These products embody not just code but also brand identity, user experience, and data-driven differentiation. Therefore, Deployed Applications can often meet all IP protections tests and be commercially licensed.
Table 15: Deployed Applications—Practical SME Examples
|
Conversational Interface |
A consulting firm develops an AI chatbot that provides regulatory compliance guidance to clients in real time. |
|
Generative Design Tool |
An architecture SME deploys an AI system that creates adaptive building layouts based on site and client preferences. |
|
Predictive Dashboard |
A logistics analytics company offers a subscription-based interface forecasting delivery delays and route risks. |
Table 16: Deployed Applications—Supporting Categories
|
Prompt Libraries and RAG Architectures supply context for user queries and improve response precision. |
|
Autonomous Agent Frameworks enable actions such as automatic reporting, alerting, or workflow triggers. |
|
AI Governance Models monitor user interaction quality, safety, and compliance with ethical guidelines. |
Table 17: Deployed Applications—Protection Strategies
|
Copyright |
Protects user interface design, code, and graphical layout. |
|
Trade Secrets |
Protects backend integration logic, deployment processes, and performance data. |
|
Contracts |
Governs licensing, API access, and terms of use. |
|
Patents |
Covers novel user interaction methods, automation flows, or technical integration processes. |
|
Trademark |
Safeguards product names, logos, and taglines, reinforcing customer trust. |
Practical Advice for SMEs
SMEs should register trademarks for any branded AI products to preempt competition, create End-User License Agreements (EULAs), and formalize their Terms of Service to define acceptable AI use, user data handling, and liability limitations. Feedback loops should also be established to collect performance data and enable iterative improvements to the model. SMEs can then implement usage monitoring systems to detect misuse or unauthorized reselling. Finally, all Deployed Applications should be tied to trademarks and corporate brand narratives to link their technology to identity.
Aside from generating revenue, Deployed Applications can also carry reputational significance for a firm by being the tangible proof of AI maturity, and the bridge between technical innovation and user trust. Protecting these assets with layered IP and brand strategies is becoming increasingly essential for both monetization and market credibility.
In the ensuing second part, the authors provide further guidance for SMEs on how to build an AI IP Portfolio.
This article was previously published by J.S. Held Insights, March 2026, and is republished here by permission.
James E. Malackowski is the Chief Intellectual Property Officer (CIPO) of J.S. Held and Co-founder of Ocean Tomo, a part of J.S. Held. In 2025, the Licensing Executives Society International (LES) recognized him with its highest honor: the LES Gold Medal. In 2022, he was inducted into the IP Hall of Fame and received the Q. Todd Dickinson Award for significant contributions to IP as a business asset. He is only the seventh person honored with both the LES Gold Medal and IP Hall of Fame inclusion. Mr. Malackowski has served as an expert on more than 100 occasions on intellectual property economics, including valuation, royalty, lost profits, price erosion, licensing terms, venture financing, copyright fair use, and injunction equities. He has substantial experience as a board director for leading technology corporations, research organizations, and companies with critical brand management issues.
Mr. Malackowski may be contacted at (312) 327-4410 or by e-mail to james.malackowski@jsheld.com.
Eric T. Carnick is an experienced expert who has testified in federal court and arbitration matters. He is a Managing Director in the Intellectual Property Disputes Financial Expert Testimony practice with Ocean Tomo, a part of J.S. Held. Mr. Carnick’s over 15 years of consulting experience includes the analysis and quantification of economic damages arising from patent, trademark, trade secrets, copyright infringement, and breach of contract in more than 100 matters. He has a vast knowledge base of financial issues and theories related to intellectual property and breach of contract litigation from discovery to trial.
Mr. Carnick may be contacted at (312) 377-4860 or by e-mail to eric.carnick@jsheld.com.
David Ngo is a Senior Analyst in the Intellectual Property Disputes Financial Expert Testimony practice at Ocean Tomo, a part of J.S. Held, working out of the San Francisco office. The practice quantifies economic damages arising from intellectual property disputes and provides general litigation support. Mr. Ngo assists in the valuation of intellectual properties for companies across multiple industries by applying his knowledge to financial statement analysis, market research, and consultation to develop complex models for clients.
Mr. Ngo may be contacted at (415) 946-2565 or by e-mail to david.ngo@jsheld.com.
References
[1] https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai, https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap, https://cmr.berkeley.edu/2025/08/adoption-of-ai-and-agentic-systems-value-challenges-and-pathways/
[2] https://news.bloombergtax.com/financial-accounting/accounting-groups-differ-on-tracking-intangible-assets-in-ai-era, https://www.datastudios.org/post/how-to-classify-and-measure-intangible-assets-under-ifrs-and-us-gaap, https://www.spglobal.com/market-intelligence/en/news-insights/research/sme-it-spending-strategies-for-2025-driven-by-ai-adoption
[3] https://profwurzer.com/navigating-the-labyrinth-ip-management-in-smes/, https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0323249#pone.0323249.ref009
[4] https://oceantomo.com/insights/ai-as-ip-a-framework-for-boards-executives-and-investors/, https://www.ifrs.org/issued-standards/list-of-standards/ias-38-intangible-assets/, https://kpmg.com/mt/en/home/insights/2021/05/capitalisation-of-internally-generated-intangible-assets.html
[6] https://www.hitechdigital.com/blog/accurate-ai-training-data-for-machine-learning
[7] https://www.forbes.com/councils/forbestechcouncil/2025/06/03/the-massive-implications-of-data-becoming-a-commodity/, https://www.ainewsinternational.com/the-data-moat-nobody-talks-about-why-proprietary-curation-is-the-new-competitive-advantage/, https://research.aimultiple.com/data-curation/, https://www.ibm.com/think/topics/data-curation
[8] https://www.ibm.com/think/topics/neural-networks, https://www.ibm.com/think/topics/embedding
[9] https://www.ibm.com/think/insights/proprietary-data-gen-ai-competitive-edge
[10] https://ieeexplore.ieee.org/document/9036442, https://learn.microsoft.com/en-us/azure/architecture/ai-ml/guide/ai-agent-design-patterns, https://www.conductor.com/academy/ai-optimization/
[11] https://www.nvidia.com/en-gb/glossary/ai-infrastructure/, https://www.ibm.com/think/topics/ai-infrastructure
[12] https://www.nvidia.com/en-gb/glossary/ai-infrastructure/, https://www.ibm.com/think/topics/ai-infrastructure
[13] https://www.ifrs.org/issued-standards/list-of-standards/ias-38-intangible-assets/, https://bridgeway.com/perspectives/investing-in-the-age-of-ai-why-intangible-assets-matter-more-than-ever/, https://kpmg.com/mt/en/home/insights/2021/05/capitalisation-of-internally-generated-intangible-assets.html, https://www.pkf-l.com/insights/capitalising-ai-tools-accounting-ias-38/
[14] https://aign.global/wp-content/uploads/2025/08/AIGN-The-Supervisory-AI-Governance-Framework.pdf; https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/services/consulting/2024/ai-governance-roadmap-accessible.pdf
[15] https://www.accenture.com/us-en/blogs/intelligent-operations-blog/operations-maturity, https://www.simon-kucher.com/en/insights/private-equity-operational-era-value-creation-accelerates
[16] https://www.truefoundry.com/blog/what-is-ai-model-deployment
