Systematic Review Proposes Data Trust Reference Architecture for Cross-Organizational Collaborations

A comprehensive systematic literature review published in Data Science and Engineering presents an integrative reference architecture designed to resolve the technological and organizational complexities of cross-organizational data sharing. Conducted by researchers Ali Imran Jehangiri, Ulf Lotzmann, Philipp Räder, and Maria A. Wimmer, the investigation establishes how modern enterprise collaboration relies heavily on multi-organizational data platforms to extract value from disparate, decentralized information sources. Despite the growing recognition of data ecosystems as critical drivers of economic value and technological innovation, organizations often struggle to implement shared platforms due to governance bottlenecks, access control liabilities, and disparate technical capabilities among collaborating entities. To bridge these operational divides, the investigators evaluated the structural composition of existing data sharing paradigms and synthesized sixty core academic studies alongside two distinct empirical implementations. The resulting conceptual model introduces a fiduciary data trust reference architecture specifically calibrated to relieve technical burdens on smaller participating entities while ensuring rigorous data sovereignty, compliance, and end-to-end analytical security.

To assemble an evidence base that is both methodologically rigorous and representative of the wider academic field, the researchers executed their literature review in strict compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The inquiry deployed a structured search protocol across five principal computer science digital libraries: Scopus, Web of Science, IEEE Xplore, Springer, and ProQuest Library. Using an iterative keyword query combining terms such as data platform, multi-organization, cross-organization, and building block, the initial database interrogation in June 2024 yielded 348 candidate records. After importing these records into a centralized database and removing 19 duplicate entries, the team instituted a multi-stage screening process. Author C initially evaluated all candidate titles against predetermined inclusion criteria, followed by a secondary verification round by Author A, with disagreements resolved through adjudication with Author D. This title phase eliminated 220 irrelevant publications. Subsequent evaluations excluded 8 publications due to full-text access or archival barriers and removed another 34 studies after full-text appraisal for failing to meet methodological depth or relevance benchmarks. Addressing an initial scarcity of studies exploring data lake architectures, the authors conducted a forward reference and citation search that added 3 eligible studies, culminating in a synthesized corpus of 60 peer-reviewed investigations. A demographic classification of this final corpus revealed that 54 percent (32 studies) were conceptual proposals, 33 percent (20 studies) represented proof-of-concept prototypes, and only 13 percent (8 studies) detailed fully operational, scalable production systems.

A primary objective of the literature review centered on categorizing the heterogeneous terminology used across scholarly discourse, identifying six distinct conceptual paradigms deployed for multi-stakeholder collaboration. The authors observed that terms such as data platforms, data spaces, data ecosystems, and data marketplaces are often applied interchangeably or with considerable semantic overlap. To resolve this ambiguity, the study delineated the boundaries of each concept. A data platform constitutes the core technological foundation, providing defined interfaces, storage, and processing rules that enable two or more parties to interact and access shared assets securely. In contrast, a data ecosystem functions as an overarching socio-technical network, emerging around one or more underlying data platforms where autonomous participants exchange resources, services, and value propositions. Data spaces represent federated arrangements governed by formal legal conditions, access policies, and contractual agreements that ensure data sovereignty, predominantly relying on standardized connector technologies to facilitate decentralized transfer. Data lakes serve as scalable storage repositories structured into ingestion, storage, transformation, and interaction layers to retain structured and unstructured data in raw formats. Data marketplaces operate as commercialized derivatives of data spaces specifically oriented toward the monetized trading of data assets, while data mesh architectures enforce decentralized domain ownership, treating data as an autonomous product governed by federated computational policies.

Delving into the practical mechanics of these platforms, the authors established an architectural decomposition of the multi-organizational pipeline comprising eight essential functional components. The pipeline begins with data sources, encompassing diverse inputs such as Internet of Things environmental sensors, remote high-definition cameras, administrative archives, and private mobile devices, which require active harmonization against heterogeneous update frequencies and formats. The second component, data ingestion, oversees the ongoing, automated capture of this distributed information, relying on integration adapters, web crawlers, and event-driven message brokers to maintain continuous data quality. Data storage follows, addressing long-term preservation, redundancy management, and schema versatility by coordinating object stores, relational databases, time-series engines, and spatial repositories under the guidance of Findable, Accessible, Interoperable, and Reusable data standards. Data processing subsequently executes crucial pre-processing steps, including filtering erroneous records, reconciling missing observations, and attaching semantic contextual annotations to prepare datasets for computational use. Building on these structured stores, data analytics leverages machine learning algorithms, statistical modeling, and distributed execution engines to derive actionable knowledge, which is then made accessible through interactive visual dashboards, graphical query tools, and application programming interfaces within the interface and visualization component. Cutting across all operational stages are data security and privacy, which mandate encryption, granular authentication, and anonymization, alongside comprehensive data governance frameworks that define rights allocation, regulatory compliance, and metadata cataloging.

To evaluate how these theoretical building blocks function within operational software environments, the research team analyzed six existing multi-organizational implementations spanning three primary architectural models: data spaces, data marketplaces, and big data analytics environments. In the data space domain, architectures such as the FIWARE smart solution framework, the agricultural Twin-Hub system, and the TDMS maritime testbed demonstrated how decentralized trust is constructed through standardized international data space connectors, identity registries, and fine-grained consent managers that uphold organizational sovereignty. In sharp contrast, the examined industrial federated data marketplace introduced dedicated commercial components absent in standard data spaces, including automated contract lifecycle management, remuneration calculation engines, and digital wallet managers to facilitate auditable monetary settlements. Meanwhile, centralized big data platforms such as the ICARUS aviation analytics system and healthcare compliance frameworks prioritized high-throughput batch and stream processing using clustered execution engines like Apache Spark, Kafka, and Storm. These analytics-heavy platforms embedded multi-tenant security mechanisms, such as Apache Ranger and Apache Atlas, directly into centralized Hadoop Distributed File System repositories, proving that architectural emphasis diverges significantly between sovereign exchange, commercial trading, and centralized processing performance.

To ground their findings in direct empirical research, the authors examined two distinct real-world platforms developed within their own multi-organizational research projects, beginning with a precision viticulture platform deployed for steep vineyards. Small-scale rural wineries often lack the specialized computational equipment, technical personnel, and financial resources needed to execute advanced artificial intelligence analytics independently. The smart farming platform resolved this challenge by providing a centralized, multi-tenant cloud architecture that aggregates microclimate measurements from vineyard LoRaWAN sensor nodes, forecasts from external weather application programming interfaces, and high-definition aerial imagery from autonomous agricultural robots. Data ingestion utilizes Message Queuing Telemetry Transport broker connectors alongside automated cronjob scripts, routing incoming files into a three-tiered storage environment composed of MinIO object storage, an InfluxDB time-series database, and a specialized geographic information system database. Computational pre-processing cleans and enriches spatial datasets within a dedicated geographic information system server, while research partners perform anomaly detection and predictive modeling via an integrated JupyterHub environment. Multi-organizational security is governed by an instance of Keycloak configured with isolated realms that enforce role-based access control, while data discovery and metadata publication adhere to standard Data Catalog Vocabulary guidelines implemented through an open-source CKAN repository.

The second empirical evaluation focused on a multi-tiered data platform created to support pandemic monitoring and public health analysis across diverse governmental, academic, and research organizations. Designed to ingest complex epidemiology indicators, structural demographics, viral variant assessments, and qualitative governmental containment policies, the platform integrates data from official agencies such as the Robert Koch Institute and the Landesuntersuchungsamt of Rhineland-Palatinate alongside specialized academic surveys. Because formal application programming interfaces were frequently unavailable from public health repositories during the emergency response, the platform’s ingestion layer relied on automated web crawling, document scraping, and manual categorization routines to capture quantitative comma-separated value records. Storage was designed around a three-tiered data lake model that systematically isolates raw incoming data, pre-processed analysis-ready records, and final analytical output baskets. The analytics infrastructure incorporated a containerized JupyterHub deployment that allowed distributed research groups to apply Bayesian belief networks, stochastic time-series models, and multi-scale simulations, while public dissemination was achieved through prototype interactive dashboards built on Streamlit. Identity management and access boundaries across research teams were maintained through Keycloak role-based permissions, paired with CKAN cataloging to trace dataset provenance.

Drawing upon the systematic literature review and practical lessons learned from the two use cases, the authors developed a generalized reference architecture structured specifically around the concept of a fiduciary data trust. In multi-organizational environments involving small-scale enterprises or municipal offices, decentralized peer-to-peer data spaces often impose prohibitive technical burdens, requiring each participant to maintain complex connector gateways, local compliance policies, and advanced security configurations. The proposed data trust paradigm introduces a neutral, independent trustee that hosts and manages the central platform infrastructure, acting on behalf of participating data providers and data users under strictly defined fiduciary duties. The physical pipeline of the reference architecture is structured across four primary functional tiers: the ingestion layer, the storage and processing layer, the analytics sandbox layer, and the user interface layer. These operational tiers are overseen by two foundational cross-cutting structures: an enterprise data governance and sharing layer and a comprehensive security and privacy layer, thereby embedding ethical, organizational, and legal accountability directly into the technical pipeline.

The ingestion layer accommodates diverse protocols, offering real-time streaming connectors for Internet of Things deployments, scheduled batch jobs for massive enterprise datasets, manual administrative entry forms, and federated data space connectors that verify data rights and provenance at the boundary. The central storage and processing layer implements a tripartite data lake architecture, maintaining strictly segregated raw, pre-processed, and result storage baskets supported by technologies such as MinIO, Hadoop Distributed File System, or Amazon S3. In this environment, raw inputs remain immutable, while data processing routines handle filtering, semantic annotation, and normalization. Rather than granting users direct access to raw repositories, the platform introduces an isolated analytics sandbox where approved analysts construct predictive models and machine learning pipelines without risking underlying data exposure. All interactions are governed by the cross-cutting security layer, which integrates Keycloak identity management, automated data encryption, anonymization filters, and system-wide monitoring via Grafana, ensuring compliance with regulations like the General Data Protection Regulation.

Reflecting on broader technological shifts, the review identified several prominent industry trends alongside notable research gaps in the literature. Modern platform architectures are steadily migrating away from monolithic, centralized repositories toward federated ecosystems, data spaces, and data mesh concepts that prioritize local sovereignty, interoperability, and domain autonomy. To facilitate cross-organizational data flow, metadata repositories based on standard descriptive vocabularies and isolated computational sandboxes have become central to modern system design. However, the authors emphasize that substantial research gaps persist, notably the absence of standardized, cross-domain semantic ontologies, the experimental status of dynamic post-access policy enforcement tools, and the rare operational integration of privacy-enhancing technologies within analytics workflows. Furthermore, current literature largely neglects lightweight architectural blueprints tailored to resource-constrained small businesses, a limitation this study addresses through its centralized fiduciary model.

Subject of Research: Systematic literature review and reference architecture for multi-organizational data platforms

Article Title: Data Platforms for Multi-organizational Settings: A Systematic Literature Review with Use Cases and a Reference Architecture

Article References: Jehangiri, A. I., Lotzmann, U., Räder, P., & Wimmer, M. A. (2026). Data Platforms for Multi-organizational Settings: A Systematic Literature Review with Use Cases and a Reference Architecture. Data Science and Engineering, 11(2), 412-438. https://doi.org/10.1007/s41019-025-00329-3

Image Credits: AI Generated

DOI: 10.1007/s41019-025-00329-3

Keywords: data platforms, systematic literature review, reference architecture, data trust, data governance, smart farming, pandemic monitoring, data ecosystem, Data, Platforms, Multi-organizational, Settings