Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms

Studies Management and Finance Economics, of Journal 0504-2644 (online): ISSN 0490,-2644 (print): ISSN 5202 December 12 Issue 80 Volume 8.317 Factor: Impact ,40-i12-10.47191/jefms/v8 DOI: Article 7097-7965 No: Page JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7965 Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms Miloserdov Artem Senior Product Manager (Receiving), Walmart Global Tech, Walmart Inc., 702 S.W. 8th Street, Bentonville, AR 72716-0215, USA ABSTRACT: The article examines the application of machine learning algorithms to automate supplier receiving and supply-audit processes within B2B SaaS platforms. It analyzes the architectural principles of integrating ML modules into corporate applications, including data preprocessing components, monitoring mechanisms and feedback loops. The study highlights that such solutions improve verification accuracy, shorten audit duration and reduce operational costs. Empirical examples illustrate the implementation of ML-based audit at the vendor-receiving stage, enabling a shift from selective checks to real-time systematic verification. The article substantiates the strategic importance of intelligent automation for the development of logistics in the context of digitalization and increasingly diversified supply chains. KEYWORDS: Machine learning, supply audit, SaaS, logistics, automation, B2B, digital transformation I. INTRODUCTION In the context of digitalization and the growing complexity of logistics operations, B2B SaaS platforms play a central role in managing supply chains, warehouse processes, and operational control. Among the critical components of this environment is supply audit, particularly the receiving and verification of inbound shipments (vendor receiving). Traditional mechanisms for control that rely on manual inspection are resource-intensive, prone to errors, and hardly fit for the high throughput of contemporary business operations. These limitations give rise to operational risks, prolong the receiving cycle, and reduce the overall efficiency of the supply chain. Against this background, interest in applying machine learning (ML) algorithms as a means of intelligent automation for logistics processes continues to increase. ML models can process large volumes of heterogeneous data, detect anomalies, predict discrepancies, and support real-time decision-making. The purpose of this article is to examine the potential and effectiveness of ML-based approaches for automating supply audit processes within B2B SaaS platforms. The novelty of this work lies in bringing together, within a single framework, the theoretical analysis of applying machinelearning algorithms to supply audit, the architectural solutions for integrating ML modules into B2B SaaS platforms, and a detailed implementation of ML-based vendor receiving in a real-world case of an international retailer, complemented by a comparison with publicly documented solutions from FourKites and Walmart. On this basis, the study shows how the choice of algorithm classes, the organization of the data-processing pipeline (preprocessing components, monitoring, and feedback mechanisms), and the deployment format of the models (centralized cloud scheme versus decentralized edge inference) are linked to audit accuracy, delivery-processing time, the share of manual operations, and the annual reduction in operating costs. The practical significance of the research consists in providing concrete guidelines for designing and scaling ML-based supplyaudit modules in B2B SaaS platforms, which can be used by product, logistics, and engineering teams when planning the implementation of ML solutions. II. METHODS The study relies on a combination of theoretical-analytical review and a case-oriented approach. At the theoretical level, it examines academic and industry publications on digital transformation in logistics, supply-audit automation, the development of B2B SaaS platforms, and the use of machine-learning methods in supply-chain operations. At the empirical level, the paper analyzes the implemented project of integrating an ML-based supply-audit module into a B2B SaaS platform of an international retailer operating a network of warehouses and distribution centers in Mexico and Central America, complemented by an Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7966 examination of publicly available cases from FourKites and Walmart. The evaluation of effectiveness is based on the operational indicators detailed in the main body of the article, including the time required to audit a single delivery, the share of manual operations, the volume of processed deliveries, and the estimated annual reduction in operating costs; in the corporate cases, additional attention is given to how ML-driven automation influences the scalability and resilience of logistics processes. III. MAIN PART. THEORETICAL FOUNDATIONS OF INTELLIGENT AUTOMATION IN SUPPLY AUDIT Supply audit within logistics systems refers to the verification of whether incoming shipments conform to previously established quantitative, qualitative, temporal, and financial parameters [1]. In traditional B2B platforms, this process is performed manually on the basis of primary documentation (invoices, reconciliation reports), which results in high labor intensity, prolonged processing times, and an increased probability of human error – particularly in environments characterized by high shipment volumes and multiple vendors. According to a 2025 study by Kargo, 80 % of warehouses in the United States still rely on standalone scanners, labor expenses account for 50-70 % of operating budgets, and manual data-entry errors amount to approximately 4 %. In addition, chargebacks related to warehouse management and delivery errors, asset-management failures, and delayed shipments represent 1-3 % of annual corporate revenue [2]. In response to these challenges, growing attention is directed toward intelligent automation, which involves the use of artificial intelligence (AI) – primarily ML – to replace or support routine control procedures [3]. It is essential to distinguish between the two concepts: AI constitutes a broad category encompassing all technologies designed to replicate human cognitive functions (analysis, learning, decision-making), whereas ML is a subset of AI based on statistical data-driven methods that enable models to learn without explicit programming. In logistics, ML models are used to process large volumes of operational data and detect patterns, deviations, and potential errors in near real time. Intelligent automation of supply audit involves the application of ML algorithms to partially or fully substitute manual verification. Depending on the task and the available data, various classes of algorithms may be employed (table 1). Table 1. types of ml algorithms for supply audit automation [4, 5] Algorithm type Purpose Examples Use in supply auditing Classification Assigning a supply event to a class (e.g., “valid” / “error”). Logistic Regression, Random Forest, XGBoost. Fast validation of compliance with expected parameters. Anomaly detection Identifying atypical or suspicious supply events. Isolation Forest, One-Class SVM, Autoencoders. Detecting subtle errors missed by manual checks. Clustering Grouping similar supply records for pattern discovery. K-means, DBSCAN. Vendor segmentation, pattern analysis, error categorization. Improvements in audit accuracy, reductions in data-verification time, and lower labor costs make intelligent solutions appealing from both technological and economic perspectives. This dynamic is reflected in global market trends: according to an industry report by Market.us, the market for AI solutions in warehousing was valued at $8.7 billion in 2023, with projections indicating growth to $88.4 billion by 2033, corresponding to a CAGR of 26.1 % (fig. 1). Figure 1. Global AI in warehousing market size, billion dollars Source: author’s visualization based on data from [6] Such indicators confirm a sustained business interest in intelligent automation as a strategic direction in supply chain development. Against the backdrop of increasing logistical complexity and the growing need to accelerate operational cycles, Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7967 ML-based intelligent solutions are no longer merely supportive tools but have become key drivers of efficiency and competitiveness. A distinctive feature of applying ML in this domain is the need to account for the complexity of input data: a single shipment may include dozens or even hundreds of item lines accompanied by various metadata (SKU identifiers, weight, volume, temperature, expiration date, and others). This necessitates the development of systems for data normalization and standardization prior to model training and inference. An additional requirement is ensuring model robustness to incomplete or noisy data, which is characteristic of logistics environments involving heterogeneous information sources. For example, FourKites, a U.S.-based supply chain visibility platform, employs stream processing on Apache Kafka within a cloud infrastructure together with AI methods to analyze logistics events in real time. It processes more than three million shipments daily, thus allowing for a non-stop flow of operational information in support of real-time deviation detection. Its design integrates transportation, inventory, and warehouse information into a single analytical framework, thereby reducing response times to logistics events while enhancing resilience to process disruptions. Clients ranging from Walmart and Coca-Cola to 3M are among the many international companies that use the platform, hence driving its popularity in corporate supply chains. Thus, ML-driven intelligent automation of supply audit is a promising direction for transforming logistics operations in the B2B ecosystem. The theoretical foundations, reinforced by market dynamics, demonstrate a high potential of such solutions to drive operational efficiency, reduce errors, and optimize cost structures. At the same time, successful implementation requires accounting for the specifics of logistics data, the right choice of algorithmic approaches, and their alignment with the constraints of corporate digital platforms architectures. These considerations underscore the need for an integral approach toward incorporating ML models into existing SaaS solutions. IV. INTEGRATION OF ML MODELS INTO THE ARCHITECTURE OF B2B SAAS PLATFORMS For an accurate examination of the integration of intelligent algorithms into cloud-based business systems, it is first necessary to define the essence of the SaaS model as one of the dominant architectural forms in contemporary enterprise software. SaaS is a software delivery model in which users access system functionality over the internet without the need to install or maintain software on local devices [7]. Applications operate within a cloud environment, are centrally managed by the provider, and are accessed through subscription-based mechanisms. In the B2B context, such platforms encompass a broad range of business processes, including WMS (warehouse management), ERP (enterprise resource planning), SCM (supply chain management), and logistics operations. According to a report by Mordor Intelligence, the B2B SaaS market is estimated at $0.39 trillion in 2025 (fig. 2). Figure 2. B2B SaaS market size forecast for 2025-2030, trillion dollars Source: author’s visualization based on data from [8] The advantages of B2B SaaS solutions include scalability, centralized updates, integration with heterogeneous systems, and the ability to standardize processes across an entire organization or partner network. These characteristics make SaaS platforms an optimal environment for deploying intelligent components, including ML modules designed to automate logistics operations such as supply audit. Integrating ML models into such architectures requires taking into account both the specifics of distributed business processes and the technical characteristics of the SaaS environment – from data-flow management to ensuring fault tolerance and backward compatibility. Successful deployment of intelligent solutions is feasible only when the architectural components are clearly defined, with each performing specialized functions within the overall system: from ingesting and preparing input data to interpreting outputs and monitoring model quality [9]. These components must be technologically compatible with the Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7968 platform’s cloud infrastructure, support horizontal scalability, and integrate seamlessly into existing service-to-service communication channels (table 2). Table 2. Core architectural components for ml integration in saas platforms Component Purpose Implementation features Data preprocessing module Cleansing, normalization, and transformation of input data before inference. Integration with ERP/WMS, support for JSON/CSV/XML formats, noise filtering. Inference service Provides real-time predictions via API. REST API, containerized deployment (Docker), autoscaling, response time SLA. Model management module Version control, retraining, performance monitoring, and lifecycle management. CI/CD integration, model quality metrics, automated retraining pipelines. Monitoring and logging system Tracks model performance, errors, and latency. Metrics: latency, throughput, drift detection; tools: Prometheus, Grafana. Feedback mechanism Collects user feedback to refine model accuracy. Supports human-in-the-loop; builds updated training datasets. Thus, the architecture of successful ML integration into a B2B SaaS platform is built on a modular composition in which each component performs a clearly defined function – from preparing incoming logistics data to managing the model lifecycle and interpreting predictions. Ensuring system stability as shipment volumes and the number of vendors increase is particularly important, as such conditions are typical of large-scale logistics operations. Under these circumstances, the technical infrastructure must be capable of dynamically adapting to changing workloads while maintaining the real-time reliability of intelligent modules. Within a SaaS architecture, one of the critical requirements is the ability to horizontally scale the ML infrastructure so that large volumes of supply-chain transactions can be processed with minimal latency. This is achieved through modern approaches to model containerization (e.g., using Docker) and service orchestration with Kubernetes, which enable flexible resource management and fault tolerance. In continuous integration and delivery (CI/CD) pipelines, automated deployment mechanisms introduce new model versions with prior validation (including A/B testing and canary deployment), thereby minimizing the risk of incorrect models being propagated into the production environment. The choice between centralized (cloud-based) and decentralized (edge-based) processing depends on the specific business scenario. In supply audit tasks, where rapid decision-making is required during item scanning, edge inference – executing the model on local gateways – is preferable because it reduces latency and ensures operational autonomy. At the same time, strategic-level analytical modules (e.g., aggregated vendor performance monitoring or predictive auditing) can be effectively deployed in cloud environments, where larger data volumes and higher computational demands are acceptable. Such architectural flexibility is especially relevant in diversified supply chains, where a platform may serve numerous categories of vendors differing in geography, product assortment, and documentation requirements [10]. In these settings, the intelligent system must adapt to heterogeneous control conditions by either providing a generalized model with strong crossdomain performance or supporting modular configurations optimized for specific supply segments. Overall, integrating ML models into the architecture of a B2B SaaS platform requires not only a thoughtful selection of algorithms but also deep technical alignment with the cloud infrastructure, data pipelines, and business logic of the system. The presented approach demonstrates that effective automation of supply audit is achievable only when the principles of modularity, scalability, and adaptability are upheld – particularly in highly diversified supply-chain environments. Architectural decisions must ensure both computational efficiency and the flexibility to apply models across different logistical scenarios, ranging from operational inbound verification to strategic vendor-risk analysis. In this regard, B2B SaaS platforms function not only as technological foundations for implementing intelligent systems but also as catalysts for realizing their practical value in next-generation logistics. V. PRACTICAL IMPLEMENTATION: ML-DRIVEN AUTOMATION OF VENDOR RECEIVING For further analysis, it is important to turn to practical examples that illustrate how the described architectural and algorithmic approaches manifest in real-world B2B SaaS environments. As a key practical case, this study examines the implementation of an ML-based audit module for vendor receiving within a large B2B SaaS platform operated by an international retailer and supporting a network of warehouses and distribution centers across Mexico and five Central American countries. In the initial Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7969 configuration, supply verification was performed selectively and predominantly manually, relying on the comparison of invoices with actual item-level data, which resulted in a high operational workload and delays in the receiving process. As part of the project, an ML module integrated into the Receiving application (iOS/Android/TC) was developed and deployed to automatically classify inbound shipments by risk level and identify transactions requiring detailed review. According to internal evaluation, the time needed to audit a single shipment was reduced by 64 %, while lower error rates and reduced manual labor contributed to an estimated annual savings of approximately $1.5 million. The practical case presented is based on a project implemented for a major international retailer. All data are provided in anonymized form. An additional illustration can be found in the solutions implemented at Walmart, where ML models and AI-enabled services are integrated into corporate platforms for inventory and logistics management [11]. According to Walmart Global Tech, the company employs its proprietary ML platform and AI-driven inventory systems to analyze data from distribution centers and stores in near real time, enabling automatic detection of stock discrepancies and adjustment of product placement without continuous operator involvement. Notably, Walmart reports that its predictive models can forecast out-of-stock events in stores with an accuracy exceeding 70 %. These systems deliver more accurate detection of supply-chain deviations, accelerate decision-making through automated diagnostics and incident correction, and reduce operational expenditures by optimizing routes, minimizing stockouts, and decreasing excess inventory levels. Company publications and industry analyses emphasize that AI and ML technologies constitute an essential component of Walmart’s strategy to build an integrated, end-to-end, and resilient global supply chain. Taken together, these practical examples demonstrate that integrating ML modules into B2B SaaS architectures yields measurable improvements across key dimensions of supply-chain management. Automating receiving processes reduces the share of manual operations, accelerates verification cycles, and improves audit accuracy, while large-scale initiatives such as those at Walmart confirm that ML can support real-time operational decision-making and enhance the resilience of logistics processes. These results show that intelligent automation is not limited to isolated pilot initiatives but can systematically strengthen control mechanisms, reduce operating costs, and increase the reliability of supply chains when embedded within mature SaaS architectures. CONCLUSIONS The intelligent automation of supply audit using ML algorithms within B2B SaaS platforms represents an effective tool for transforming logistics operations. The integration of ML models into corporate application architectures enables improvements in audit accuracy, faster verification of inbound shipments, and reduced operational workload. Such implementations support a shift from fragmented, manual control to a systematic and adaptive auditing model capable of operating in real time and accommodating the diverse characteristics of different vendor categories. Given ongoing digitalization trends and the increasing diversification of supply chains, there is reason to believe that the use of ML in supply audit will continue to expand. Future developments may include greater personalization of models for specific industry domains, enhanced predictive control mechanisms, and deeper integration with other elements of intelligent logistics – from risk management to automated contract compliance. As a result, ML-driven solutions are becoming a vital part of the strategic infrastructure of digital supply chains, laying the groundwork for more transparent, resilient, and adaptive logistics systems of the future. REFERENCES 1) Manuel, A., Aswale, N. 2024 Emerging technologies in global supply chain management (GSCM) using an automated eaudit system. Smart Global Value Chain. P.212-220. 2) Warehouse AI Meets Automated Dock Doors – Introducing Kargo Intelligence / Kargo Inc. // URL: https://www.kargo.ai/blog/kargo-intelligence-warehouse-ai (date of application: 06.11.2025). 3) Roilian, M. 2025 Economic efficiency of procurement process automation: analysis of architectural solutions and business metrics. International independent scientific journal. No 76. P. 7-12. 4) Fekih, A. 2025 A conceptual framework for integrating Robotic Process Automation in logistics audits and supply chain management. International Journal of Accounting, Finance, Auditing, Management and Economics. Vol. 6. No. 5. P. 775789. 5) Blazhkovskiy, A.V., Korostin, A.A., Tretyakov, I.B., Larin, M.D. 2025 Use of artificial intelligence to improve the efficiency of customer interaction in logistics. Competitiveness in the Global World: Economics, Science, Technology. No. 2. P.189-194. 6) Global AI in Warehousing Market Size / Market.us // URL: https://market.us/report/ai-in-warehousing-market/ (date of application: 06.11.2025). Intelligent Automation of Supply Audit Using Machine Learning in B2B Saas Platforms JEFMS, Volume 08 Issue 12 December 2025 www.ijefm.co.in Page 7970 7) Šilenskytė, A., Carneiro, J., Kohtamäki, M. 2025. Digital leadership when utilizing software-as-a-service for strategic planning and implementation. Digital Leadership. Edward Elgar Publishing. P. 92-109. 8) B2B Saas Market Size Source / Mordor Intelligence // URL: https://www.mordorintelligence.com/industry-reports/b2bsaas-market (date of application: 09.11.2025). 9) Mukayev, T. 2025 Development of adaptive machine learning models for high-load information systems. International Journal of Advanced Research in Science, Communication and Technology. Vol. 14(4). P. 194-198. 10) Kukula, I. 2025 Geopolitical and economic strategies of the USA to diversify the supply of strategic metals. Economic development research journal. No. 4. P. 49-55. 11) Decking the aisles with data: How Walmart's AI-powered inventory system brightens the holidays / Walmart // URL: https://tech.walmart.com/content/walmart-global-tech/en_us/blog/post/walmarts-ai-powered-inventory-systembrightens-the-holidays.html (date of application: 12.11.2025).