A model for incorporating AI into ERP software

You’re listening to “A model for incorporating AI into ERP software,” by S. Sarferaz. Published in 2026.

Abstract.

Enterprise resource planning (ERP) systems form the digital backbone of modern organizations by integrating business processes and data across functional domains. Recent advances in artificial intelligence (AI) offer significant opportunities to enhance ERP systems through intelligent automation and data-driven decision support. However, existing research primarily examines isolated AI applications and does not provide structured approaches for systematically incorporating AI capabilities into ERP-supported business processes. This study proposes a framework for integrating artificial intelligence into ERP systems by combining insights from automation theory, enterprise process modelling, and AI technology categorization.

The framework decomposes ERP-supported business processes into four automation dimensions—data acquisition, information analysis, decision making, and action execution—and introduces a measurable method for assessing automation maturity through a geometric aggregation model. A systematic literature review and conceptual modelling approach are used to develop the framework and map AI technology categories, including narrow AI, generative AI, conversational AI, and agentic AI, to ERP process automation capabilities. The proposed framework contributes to the literature by extending automation theory to the context of ERP systems and by providing a structured method for analysing and guiding AI-driven ERP transformation.

From a practical perspective, the framework offers ERP vendors and organizations a diagnostic tool for assessing current automation levels and identifying targeted opportunities for incorporating artificial intelligence into enterprise business processes.

Enterprise Resource Planning (ERP) systems represent one of the most influential classes of enterprise information systems, enabling organizations to integrate business processes across functional domains such as finance, supply chain management, manufacturing, human resources, and customer management. By enabling shared data models and standardized end-to-end processes, ERP systems provide organizations with a centralized digital backbone for coordinating enterprise operations. Over the past decade, ERP platforms have increasingly incorporated advanced technologies such as artificial intelligence (AI), advanced analytics, and cloud computing, transforming traditional transaction-processing systems into intelligent enterprise platforms. As shown in Fig.

1, early enterprise software landscapes in the 1980s and early 1990s were characterized by isolated best-of-breed solutions for individual functional areas. These systems were later replaced by monolithic ERP platforms that integrated core enterprise processes into a single system architecture. With the emergence of cloud computing in the 2000s, ERP systems increasingly adopted distributed cloud-based architectures. Today, the latest stage of ERP evolution integrates artificial intelligence capabilities directly into enterprise processes, enabling intelligent ERP systems that support predictive analytics, automation, and adaptive decision-making.

Correspondence:

Siar Sarferaz

the email address

1Research and Development, SAP

To establish a conceptual foundation for this study, it is important to clarify the meaning of ERP systems. Numerous definitions of ERP exist in the scientific literature, reflecting different perspectives on enterprise integration, business process management, and enterprise software architectures. Scholars such as Davenport, Klaus et al., and O’Leary emphasize the integrative nature of ERP systems and their role in coordinating enterprise-wide processes and information flows. Drawing on these established perspectives, the definition used in this paper synthesizes insights from prior literature. Considering these various perspectives, we synthesize the following definition: Enterprise Resource Planning (ERP) is a software solution for digitalizing business processes within and across functional areas of an organization to increase the level of automatization.

In this context, digitalization refers to the transformation process of traditional business operations and models by leveraging digital technologies such as software. This transformation goes beyond simply converting paper documents to digital files; it encompasses rethinking how business is conducted using digital technologies as a foundation. Automatization, in turn, refers to the utilization of digital technology to perform tasks with minimal human intervention for processing routine or repetitive tasks that were previously performed manually.

Artificial intelligence has emerged as a key technological driver behind this latest stage of ERP evolution. AI represents a broad field of research concerned with the development of computational systems capable of performing tasks that traditionally require human intelligence, including learning, reasoning, perception, and decision making. However, defining AI precisely remains challenging because the concept has evolved significantly over time and encompasses multiple research paradigms. Early definitions of AI focused on the simulation of human intelligence in machines. McCarthy defined AI as the science and engineering of making intelligent machines, while Minsky described it as the science of making machines perform tasks that would require intelligence if performed by humans. More recent definitions emphasize adaptive learning from data and goal-oriented behaviour.

For example, Russell and Norvig describe AI systems as rational agents that perceive their environment and take actions to maximize the achievement of defined objectives, while Kaplan and Haenlein emphasize the ability of systems to interpret data, learn from it, and adapt their behaviour accordingly. Drawing from these various perspectives, we synthesize the following definition of AI in the context of ERP software: Artificial intelligence (AI) encompasses the development of computational systems capable of performing tasks that typically require human intelligence, including perception, learning, reasoning, problem-solving, and adaptation. These systems range from narrow applications focused on specific domains to more general approaches that aim to replicate or exceed human cognitive capabilities across domains.

AI systems achieve these capabilities through various techniques including, but not limited to, narrow AI, generative AI, conversational AI, and agentic AI. This definition acknowledges both the historical human-reference perspective and the more capability-specific approaches while remaining flexible enough to accommodate the evolving nature of AI research and applications.

The convergence of ERP systems and AI technologies has led to the emergence of intelligent ERP systems, where machine learning algorithms, natural language processing, and generative methods enhance traditional enterprise processes. AI-enabled ERP systems can automate routine tasks, detect patterns in enterprise data, generate predictions about future business conditions, and support complex decision-making processes. These capabilities significantly expand the role of ERP systems from transaction-processing infrastructures to intelligent platforms that actively assist organizations in managing operations and strategic decision making. Despite the growing importance of AI in enterprise systems, the integration of AI capabilities into ERP environments often remains fragmented.

In many organizations, AI technologies are introduced through isolated modules, vendor-specific tools, or experimental pilot projects rather than through systematic integration within enterprise business processes. Such ad hoc approaches make it difficult for organizations to assess the maturity of AI adoption or to identify where AI technologies can deliver the greatest value. Existing research on AI in enterprise systems frequently focuses on specific applications of AI technologies or on broader digital transformation initiatives. While these studies provide valuable insights into the potential of AI for improving business processes, they rarely offer systematic frameworks for incorporating AI capabilities across ERP-supported processes in a measurable and replicable manner.

This observation motivates the central research question of this study: How can artificial intelligence capabilities be systematically integrated into ERP business processes in a way that enables both rigorous assessment and practical guidance for enterprises and ERP vendors?

To address this question, this paper proposes a framework for incorporating AI into ERP systems by decomposing business processes into four fundamental automation dimensions: data acquisition, information analysis, decision making, and action execution. Each dimension is evaluated along a five-level automation continuum ranging from manual operation to fully autonomous execution. By linking these automation dimensions with categories of AI technologies—including narrow AI, generative AI, conversational AI, and agentic AI—the framework provides a structured approach for assessing and implementing AI-driven automation within ERP environments. The contributions of this study are threefold. First, the paper adapts automation theory to the context of ERP systems, accounting for the characteristics of enterprise-wide integrated processes and structured enterprise data environments.

Second, the proposed framework introduces a measurable approach for assessing AI incorporation within ERP-supported processes. Third, the study provides practical guidance for ERP vendors and enterprise practitioners seeking to identify and implement AI opportunities within enterprise software landscapes.

2 Related work and research gap

The integration of AI into enterprise information systems has attracted increasing attention in both academic research and industrial practice. Existing studies addressing this topic can broadly be categorized into three main streams of research: studies examining ERP systems and enterprise integration, studies investigating AI applications in enterprise systems, and studies analyzing the integration of AI technologies into ERP environments.

ERP systems have been widely studied as enterprise-wide information infrastructures that integrate organizational processes and data across functional domains. Early research focused primarily on ERP implementation challenges, organizational change, and business process standardization. These studies emphasized that ERP systems enable organizations to achieve enterprise-wide integration by replacing fragmented information systems with unified platforms based on shared data models. Subsequent research has highlighted the role of ERP systems as the digital backbone of modern organizations. By integrating core business functions such as finance, procurement, manufacturing, and human resources, ERP systems enable organizations to coordinate complex activities and improve operational transparency.

As ERP systems evolved, researchers increasingly emphasized their importance for supporting digital transformation initiatives and enterprise-wide information integration. In recent years, ERP architectures have shifted toward cloud-based and platform-oriented solutions. Cloud ERP systems allow organizations to deploy enterprise software on scalable infrastructures and facilitate integration with external digital services and partners. This architectural transformation provides new opportunities for incorporating emerging technologies such as advanced analytics, Internet of Things (IoT), and artificial intelligence into enterprise systems. However, integrating such technologies into ERP environments also presents architectural challenges. Traditional ERP systems were primarily designed for transactional data processing rather than advanced analytical capabilities.

As a result, incorporating intelligent technologies often requires extending ERP architectures to support tighter integration between operational processes.

In parallel with developments in ERP systems, AI technologies have advanced rapidly and are increasingly applied in enterprise environments. AI techniques such as machine learning, deep learning, and natural language processing enable organizations to analyze large volumes of structured and unstructured data and generate insights that support business processes. Within enterprise contexts, AI technologies are commonly applied to tasks such as predictions, anomaly detection, recommendation systems, and content creation. Davenport and Ronanki demonstrate how AI technologies can augment managerial decision-making and automate operational processes across a wide range of business domains. Similarly, Huang and Rust highlight the growing role of AI in augmenting analytical and decision-making tasks traditionally performed by human managers.

More recent research has examined the broader role of AI in digital transformation. Dwivedi et al. provide a comprehensive overview of AI applications in business and management and identify emerging opportunities for AI-driven organizational innovation. Likewise, Rai et al. discuss how AI capabilities can enhance data-driven operations in organizations by combining machine learning techniques with enterprise information systems. Despite these advances, many AI implementations remain isolated analytical tools rather than integrated components of enterprise system architectures.

A growing body of research has begun to explore the integration of AI technologies into ERP systems. These studies often emphasize that ERP systems provide structured enterprise data that can serve as a valuable foundation for machine learning applications and predictive analytics. For example, Seddon et al. demonstrate how ERP systems can improve organizational performance by enabling better information visibility and decision support. Building on this perspective, Wamba-Taguimdje et al. analyse the influence of artificial intelligence technologies on firm performance and highlight the potential of AI-driven analytics for improving business processes. Recent research has increasingly examined AI-enabled ERP innovations.

Pokala provides a systematic review of AI applications in ERP systems and identifies emerging use cases such as predictive supply chain planning, automated financial analysis, and intelligent process monitoring. Similarly, Mustafa and Zeebaree discuss AI-driven innovations in enterprise systems and highlight the importance of machine learning technologies for improving enterprise capabilities. Sunkara analyses AI-powered ERP and CRM systems and emphasizes the potential of intelligent automation for transforming enterprise operations and enhancing customer engagement. A closely related conceptual contribution is provided by Dziembek and Turek, who propose a model for integrating AI with ERP systems with the long-term objective of enabling autonomous business management systems. Their model highlights the potential of AI technologies for enabling more adaptive enterprise systems.

While these studies demonstrate the potential of AI-enabled ERP systems, most focus primarily on individual AI applications or conceptual discussions of intelligent enterprise systems rather than providing systematic frameworks for integrating AI capabilities into ERP-supported business processes.

Several recent studies have examined the integration of AI technologies into ERP systems. However, these studies differ significantly in terms of scope, methodological approach, and level of abstraction. Table 1 summarizes selected studies and compares them with the framework proposed in this research. The comparison illustrates that although existing studies provide valuable insights into AI applications in enterprise systems, few provide systematic frameworks that connect ERP process structures with AI capabilities and measurable automation levels.

The literature reviewed above highlights several limitations in current research on AI-enabled ERP systems. First, many studies focus on specific AI applications within enterprise systems rather than examining how AI capabilities can be systematically integrated across ERP-supported business processes. Second, the literature rarely provides measurable approaches for assessing the degree of AI integration within ERP environments. Organizations therefore lack structured methods for evaluating their current level of automation and identifying opportunities for improvement. Third, there is limited research that explicitly connects ERP process architectures with AI technology capabilities. Without such integration logic, AI technologies are often implemented as external modules rather than embedded components of enterprise process architectures.

These limitations highlight the need for a structured framework that enables organizations to systematically evaluate and incorporate AI capabilities across ERP-supported processes. The present study addresses this gap by proposing a framework that decomposes ERP business processes into four automation dimensions (data acquisition, information analysis, decision making, and action execution), introduces a measurable approach for assessing automation levels in ERP-supported processes, and links automation dimensions with categories of AI technologies that enable enterprise process automation.

Automation theory—particularly its frameworks for function allocation and levels of automation —offers analytical purchase on human–machine task distribution, yet its application to learning-based enterprise infrastructures remains underdeveloped. Existing formulations privilege discrete, rule-governed task transfer and do not readily extend to the probabilistic, inference-driven processes characteristic of AI-augmented ERP environments. In practice, AI incorporation into ERP systems tends to occur below the threshold of deliberate automation design: as isolated pilot deployments, loosely coupled analytics modules, or vendor-driven feature extensions that bypass systematic function allocation analysis. Such patterns preclude meaningful assessment of automation level, obscure the locus of supervisory control, and leave lifecycle governance unaddressed.

Without explicit automation-theoretic grounding, AI risks operating as an unclassified augmentation layer rather than a functionally allocated, architecturally constitutive component of the enterprise system. Thus, the framework proposed in this study builds upon existing automation theory while adapting it to the specific context of ERP systems. In particular, the automation model introduced by Parasuraman et al. conceptualizes automation as a multi-stage interaction between humans and machines. In information systems research, the adaptation of theoretical concepts to specific technological contexts is often referred to as context-specific theorizing. According to Hong et al., theoretical constructs developed in one domain can be extended to new domains by systematically analysing the unique characteristics of the target context and adjusting theoretical constructs accordingly.

ERP systems represent such a distinct context because they combine several characteristics that differ significantly from traditional automation environments. ERP systems support end-to-end enterprise processes across multiple functional modules, operate on highly structured enterprise data models, and serve as enterprise-wide integration platforms connecting internal and external stakeholders. Following the principles of context-specific theorizing, this study adapts the automation framework to the ERP domain by linking automation stages with

ERP-supported business processes and mapping them to artificial intelligence technologies capable of supporting different automation levels. Through this contextualization, the proposed framework extends automation theory by introducing a process-oriented abstraction of automation within enterprise systems and by connecting automation levels with contemporary AI technologies.

3 Research methodology

This research employs systematic literature review and framework development methodology to address the complex challenge of incorporating AI into ERP systems. Our methodological approach combines theoretical foundation-building with practical framework construction, utilizing both deductive and inductive reasoning to develop a comprehensive methodology for AI-ERP integration. The research follows a multi-phase design structured around three core methodological pillars.

For the conceptual foundation development, we conducted a systematic literature review following established guidelines for structured literature analysis and PRISMA-inspired selection procedures by searching five major scholarly databases—Scopus, Web of Science, IEEE Xplore, ACM Digital Library, and ScienceDirect—covering publications from 1990 to 2026. The search strings combined ERP-related and AI-related keywords, for example, (“Enterprise Resource Planning” OR ERP) AND (“Artificial Intelligence” OR AI OR “systematic” OR “integration”) and (“intelligent ERP” OR “autonomous ERP” OR “ERP automation” OR “AI incorporation”). Inclusion criteria required that papers (a) addressed ERP in conjunction with AI or automation, (b) were published in peer-reviewed journals or conference proceedings, and (c) were written in English.

Exclusion criteria removed duplicates, grey literature, and papers not focusing on ERP. The initial search returned 742 records, which were screened by title and abstract. After removing irrelevant and duplicate works, 38 papers were retained for full-text analysis. This synthesis revealed that while many studies address ERP adoption or AI capabilities separately, no study systematically combines ERP process decomposition with AI capability mapping into a replicable automation framework. This research gap motivates our proposed model.

For the framework architecture development, we enhanced Porter’s value chain theory as the foundational framework, systematically mapping ERP business processes into four enterprise domains: Products and Services, Customer, Supply, and Corporate. This mapping utilized process decomposition analysis, where complex ERP implementations were broken down into constituent business processes and then categorized by domain. Although the four enterprise domains provide a simplified abstraction of ERP-supported activities, they are intended to cover the full spectrum of enterprise processes typically implemented in ERP systems. Processes such as warehouse management and logistics operations are included within the Supply domain, while service management and customer support processes are mapped to the Customer domain.

Similarly, project management and governance-related activities are represented within the Corporate domain, which captures organizational planning, financial control, and administrative coordination functions. The classification therefore serves as a conceptual aggregation layer that groups detailed ERP process models into a manageable number of enterprise domains while preserving their functional relationships. Building on automation theory from Parasuraman et al., we adopted the four-dimensional analytical framework for

ERP software that dissects business processes into data acquisition, information analysis, decision making, and action execution. Each dimension operates on a five-level automation continuum (manual to full automation), creating a measurable assessment structure. The methodology incorporates geometric representation where automation levels for each dimension create coordinate points that form a rectangle. The area of this rectangle represents the overall automation level of a business process, enabling quantitative aggregation across all processes to determine total ERP automation degree.

For the technology categorization and application framework, we employed qualitative content analysis to categorize AI technologies relevant to ERP systems. Through systematic examination of technological capabilities and their application to business processes, we identified four primary categories: narrow AI, generative AI, conversational AI, and agentic AI. The methodology incorporates a current-state to future-state mapping approach, where organizations assess their present automation levels across all four dimensions and define target states based on business requirements and technological readiness.

We evaluated the framework’s practical utility by examining its application across different ERP business processes and industry contexts of SAP ERP as the largest ERP vendor, ensuring the methodology provides actionable guidance for both ERP vendors and customers. To demonstrate the practical feasibility of the proposed solution for systematically incorporating AI into ERP software, we applied the framework on AI applications within core ERP business processes across all four enterprise domains. We analyzed the Idea to Market processes within the Products and Services Domain, Source to Pay and Plan to Fulfill processes within the Supply Domain, Lead to Cash processes within the Customer Domain, and Recruit to Retire, Acquire to Decommission, Governance, and Finance processes within the Corporate Domain.

This comprehensive examination validated that our four-dimensional framework could be effectively applied across the complete spectrum of ERP business processes, demonstrating that the methodology provides actionable guidance for both ERP vendors and customers. Despite the successful utilization of the framework on real-world AI applications, we acknowledge limitations in our methodology. Given the rapidly evolving nature of AI technology, our categorization reflects the current state of technology as of early 2025. Future technological developments may require framework updates. While we designed the framework for broad applicability, certain industry-specific nuances may require customization of the automation level criteria.

The methodology provides a framework rather than prescriptive implementation guidelines, requiring organizations to adapt the approach to their specific contexts and constraints. Finally, as incorporation of AI into ERP systems is a very new topic and in the initial stage, there is no sufficient data available for comparing AI-powered ERP systems, AI techniques and AI theories in context of ERP software.

This methodology contributes to the existing body of knowledge by providing a systematic framework for AI incorporation into ERP systems that bridges business process theory with practical technology implementation. Unlike previous approaches that focus on either technological capabilities or business processes in isolation, our methodology creates an integrated assessment and implementation framework that enables structured decision-making for AI-ERP integration initiatives. In the next sections we will depict our framework along the described research methodology. All figures and tables in this work were developed by the author.

4 Business process clustering through enterprise domains

As motivated in the first section, AI is the foundation for intelligent ERPs. However, the question is how to systematically incorporate AI into ERP software? To deal with this challenge, we consider the business processes facilitated by ERP systems as the key ingredient for our solution proposal. A typical ERP product digitalizes tens of thousands of business processes across the organizations. Those standardized business processes with their structured data models build the optimal foundation for embedding AI into ERP software. While the business processes provide the basis for augmentation with AI functionality, the process data shapes the grounding for algorithms to learn and adopt. Thus, we will first cluster the ERP business processes into groups using the enterprise domains concept to ensure covering all ERP areas.

Secondly, we will abstract from the identified eight clusters to the key dimensions of business processes. Those key dimensions we will finally utilize for suggesting a method for systematically incorporating AI into ERP business processes.

In the complex landscape of modern enterprise management, the value chain concept serves as a fundamental framework for understanding how organizations create and deliver value to their customers. Originated by Michael Porter, this approach views an organization’s activities as interconnected links in a chain, with each link incurring costs while contributing to the final product’s value. The choice of Porter’s value chain as the conceptual basis for clustering ERP processes is motivated by its ability to represent enterprise activities as interconnected value-creating functions. ERP systems are designed to integrate precisely these activities across organizational boundaries, including procurement, production, sales, and administrative functions.

Using the value chain as an abstraction layer therefore enables the mapping of detailed ERP process landscapes into a manageable set of enterprise domains while preserving the structural relationships between operational and supporting activities. This abstraction facilitates the analysis of how automation and artificial intelligence technologies can be embedded across the entire enterprise process architecture. The value chain encompasses primary activities—those directly involved in creating and delivering products or services—and support activities that enhance the efficiency of primary operations. ERP systems typically organize the value chain into four distinct, but interconnected enterprise domains as shown in Fig. 2, each representing a critical aspect of business operations.

ERP systems implement standardized end-to-end processes that span these four domains, creating an integrated business platform. These processes connect different functional areas and departments, enabling seamless information flow and coordinated action throughout the organization.

The Products and Services Domain represents the innovative heart of an organization. Here, companies transform creative ideas into marketable offerings through the Idea to Market process. This domain encompasses activities like research and development, design engineering, and product management. Organizations with strong innovation capabilities often demonstrate particular strength in this domain, creating distinctive products that differentiate them from competitors. The Idea to Market process within this domain guides a product from initial concept through requirements analysis and design to market readiness, ensuring that creative ideas align with market needs and technological possibilities.

Beginning with ideation—where market research, customer feedback, and innovation workshops generate new product concepts—the process proceeds through detailed requirements analysis to define product specifications. Design activities then create detailed plans and prototypes, followed by development work that builds and refines the product. Throughout this process, ERP systems maintain documentation, track development milestones, and facilitate collaboration across functional teams. For instance, pharmaceutical companies rely on this process to manage new drug development from initial discovery through clinical trials to market launch, maintaining the exhaustive documentation required for regulatory approval.

The Customer Domain serves as the organization’s interface with its market. Through the Lead to Cash process, this domain manages all aspects of customer relationships— from generating initial interest through marketing activities to converting opportunities into quotes and orders and ultimately collecting payment. Companies have revolutionized their industries by excelling in this domain, creating customer experiences that build loyalty and drive revenue growth. The Lead to Cash process coordinates marketing, sales, service, and billing activities into a seamless customer journey, providing visibility into the entire revenue generation cycle. Starting with marketing activities to generate leads, the process progresses through opportunity management, quotation, order processing, fulfillment, billing, and payment collection.

ERP systems integrate these steps, providing a comprehensive view of customer interactions and streamlining the sales process. Software companies often employ this process to manage everything from initial marketing-generated interest through the sales cycle to software delivery and subscription billing.

The Supply Domain focuses on fulfilling customer demands efficiently and effectively. This domain utilizes two key processes: Source to Pay, which manages supplier relationships and procurement activities, and Plan to Fulfill, which organizes production and delivery operations. Companies have built competitive advantages through excellence in this domain, developing supply chain capabilities that deliver products faster, cheaper, and more reliably than competitors. The Source to Pay process streamlines supplier selection, contracting, procurement, and payment. Starting with supplier identification and evaluation, the process continues through requisitioning, procurement, receiving, invoice processing, and payment. Modern ERP systems automate many of these steps, reducing cycle times and errors while providing visibility into spending patterns and supplier performance.

Manufacturing companies leverage this process to ensure that materials arrive when needed and at optimal prices, balancing cost considerations with quality and reliability requirements. The Plan to Fulfill process coordinates production and delivery activities to meet customer demands. Beginning with demand planning to forecast future needs, the process continues through supply planning, production scheduling, manufacturing execution, quality control, warehouse management, and delivery. ERP systems optimize this process by balancing inventory costs against customer service levels, ensuring effective resource utilization. Food processors, for example, use this process to coordinate production schedules with sales forecasts, ensuring that perishable products reach retailers at peak freshness.

The Corporate Domain provides the administrative foundation that supports and governs the entire enterprise. The Recruit to Retire process manages the employee lifecycle from initial hiring through retirement or dismissal. Beginning with workforce planning to identify staffing needs, the process continues through recruitment, onboarding, performance management, learning and development, compensation administration, succession planning, and eventual dismissal. ERP systems, particularly through Human Capital Management modules, provide tools for managing these activities consistently across the organization. Large retail chains utilize this process to manage thousands of employees across multiple locations, ensuring consistent HR practices and compliance with labour regulations. The Acquire to Decommission process handles the lifecycle of physical assets from acquisition through disposal.

Starting with asset planning to identify needs, the process progresses through procurement, onboarding, maintenance, performance monitoring, and eventual disposal. ERP systems track these activities through Enterprise Asset Management functionality, helping organizations maximize the value of their investments in physical assets. Airlines employ this process to manage their aircraft fleets, scheduling maintenance activities and monitoring performance to ensure safety and efficiency throughout each plane’s operational life. The Governance process ensures regulatory compliance and effective risk management across the enterprise. This process encompasses risk assessment, compliance monitoring, identity and access management, cybersecurity, data privacy protection, IT infrastructure management, and trade and tax compliance.

ERP systems implement these activities through Governance, Risk, and Compliance modules, creating a structured approach to managing enterprise risks. Financial institutions rely on this process to ensure adherence to banking regulations, protect customer data, and manage operational risks across all business units. The Finance process manages the organization’s financial resources and provides visibility into financial performance. This process includes accounts payable, accounts receivable, general ledger, cash management, treasury operations, fixed asset accounting, financial reporting, and real estate management. ERP systems integrate these activities, creating a unified view of the enterprise’s financial position.

Multinational corporations depend on this process to manage financial transactions across multiple countries, currencies, and regulatory environments, ensuring accurate financial reporting and optimal use of financial resources. While this domain perhaps is less visible to customers than the other domains, excellence in corporate functions creates the stable platform needed for success in other areas. The true power of ERP systems emerges when these processes work together across domain boundaries. When a new product developed through the Idea to Market process enters production, it triggers procurement activities in the Source to Pay process and manufacturing activities in the Plan to Fulfill process.

Customer orders received through the Lead to Cash process initiate fulfillment activities, while financial transactions from all processes flow into the Finance process for consolidated reporting. This integration provides a holistic view of the enterprise, breaking down traditional functional silos and enabling more effective decision-making. By organizing the value chain into these four domains and implementing standardized end-to-end processes, organizations create a coherent business architecture that aligns technology with business operations. This alignment helps companies respond more quickly to market changes, optimize resource utilization, and deliver consistent customer experiences.

Figure 3 enhances the ERP solution architecture to a comprehensive AI-enhanced ERP system architecture that implements the introduced business processes. The system architecture is built upon a three-layer foundation where each layer serves a specific purpose in enabling AI-driven business operations. The uppermost layer represents the ERP system with AI-incorporated business processes, showcasing the functional modules that realize the business processes. These functional modules include Supply Chain, Sales, Finance, Manufacturing, Service, Human Resources, Procurement, Asset Management, and Research and Engineering. The relationship between functional modules and the previously explained business processes is not one-to-one but rather represents a collaborative approach where multiple modules work together to support the enterprise domains.

Research and Engineering primarily supports the Idea to Market process within the Products and Services Domain, while Procurement, Supply Chain, and Manufacturing modules collectively handle Source to Pay and Plan to Fulfill processes in the Supply Domain. Sales and Service modules mainly cover the Lead to Cash process in the Customer Domain, and within the Corporate Domain, Human Resources manages Recruit to Retire processes, Asset Management handles Acquire to Decommission functions, Finance manages financial processes, and various technical platform features support Governance requirements. The designation of these as AI incorporated Business Processes indicates that AI capabilities are embedded throughout these business functions rather than being external additions.

The Industry Solutions module suggests that the system contains extensions for specific sector requirements, providing flexibility for different organizational contexts.

The middle layer represents the AI technology infrastructure that powers the intelligent capabilities in the business processes above. This technological foundation encompasses four key AI technology components that work together to deliver advanced capabilities. Training and Inference Serving handles the deployment and execution of AI models, ensuring that AI capabilities are readily available to business processes. Data Science Tools provide capabilities for data analysis and model development, enabling continuous improvement of AI capabilities. Deep Learning and GPU components supply the computational power necessary for advanced AI algorithms, while Monitor and Operate functions ensure AI systems function properly and provide operational oversight for maintaining system reliability.

The foundation layer manages three types of data storage that are essential for both traditional ERP operations and AI functionality.

Configuration, Transactional, and Master Data storage maintains core business data and system configurations that form the backbone of enterprise operations. Task-specific AI Models storage houses specialized AI models trained for particular business functions, enabling targeted intelligence for specific processes. General-purpose AI Models storage contains multipurpose AI models that can be applied across various processes, providing flexibility and scalability in AI deployment. The bidirectional arrows in the figure indicate continuous data flow between the AI technology layer and the storage layer, demonstrating how AI models are trained on business data and then serve predictions back to the business processes.

This creates a continuous learning loop where business operations generate data that improves AI capabilities, which in turn enhance business process automation and intelligence. This cyclical relationship ensures that the system becomes more intelligent over time as it processes more business data. Although Fig. 3 depicts the deployment of all the components on the same ERP system, they can be distributed over multiple technology platforms. This architecture represents the evolution toward intelligent ERP systems where AI capabilities are fundamentally integrated into the business process layer rather than being mere add-ons.

The comprehensive AI technology infrastructure and robust data management foundation enable organizations to leverage AI throughout their operations, creating systems that continuously learn and adapt to improve business outcomes while maintaining the reliability and integration that traditional ERP systems provide.

5 Dimensional abstraction of business process clustering

The introduced business processes build the foundation for systematically incorporating AI into ERP. As already outlined ERP solutions digitalize business process with the goal to increase automation. Thus, we reused the automation types of and applied them to ERP business processes by proposing to dissect ERP business processes into four dimensions of automation: data acquisition, information analysis, decision making, and action execution. While these four dimensions may appear to closely resemble the well-known stages of automation introduced by Parasuraman et al., the contribution of this work lies in their contextualization and extension within the ERP domain.

Specifically, ERP systems are characterized by: end-to-end business processes that span multiple functional modules, standardized, highly structured data environments, and enterprise-wide integration across customers, suppliers, and corporate governance. These characteristics create unique requirements for automation that cannot be fully captured by generic models of automation. By embedding Parasuraman’s theoretical foundation into ERP-specific process abstractions, our model transforms a generic taxonomy into a decision-support framework tailored for enterprise systems.

Business processes realized by ERP systems, typical start with data acquisition that is analyzed as basis for making decisions and executing actions. In sales order processing for example, first a sales order is maintained, then the customer payment risk is analysed to decide for fulfilling the order and to trigger production tasks. Each of the four dimension operates on a continuum from manual operation (level 1) to full automation (level 5) as depicted in Fig. 4. Thus, the higher the level for a dimension, the higher is the degree of the automation for this dimension. Higher levels of automation are usually realized with AI techniques as we will explain in the next sections. Once the automation level for the four dimensions of a business process is determined, a rectangle can be constructed out of those points as shown in Fig. 4.

The overall automation level of a business process corresponds to the area of the rectangle. By aggregating this area value for all business processes, the total automation degree of an ERP product can be computed. While the four automation dimensions are conceptually independent, representing them geometrically as a rectangle inevitably simplifies their multidimensional relationships. The area representation therefore does not capture all possible interactions among dimensions. However, the approach provides an intuitive and transparent metric that allows practitioners to quickly assess automation maturity across processes. The use of a geometric representation is motivated by the need to balance methodological rigor with practical usability.

ERP stakeholders such as solution architects, process owners, and IT managers often require simple metrics that enable rapid comparison across numerous business processes. More complex multi-criteria evaluation methods could capture additional nuances but would significantly increase the effort required for practical application. The rectangle-area metric therefore serves as a pragmatic approximation that allows automation levels across the four dimensions to be aggregated into a single indicator while remaining easily interpretable for practitioners. The metric should therefore be understood primarily as a diagnostic tool that supports relative comparison between processes rather than as an exact mathematical representation of automation maturity. The next section depicts the four dimensions of a business process and the selection criteria for the corresponding automation levels.

5.1 Data acquisition

Data Acquisition represents the initial touchpoint between the physical world and the digital realm of ERP systems. This dimension encompasses all mechanisms through which information enters the system. At its most basic level, data acquisition relies entirely on manual human input—employees typing information into forms or uploading files (level 1). This approach is not only time-consuming but prone to human error. As automation increases, we see the emergence of hybrid approaches where manual entry coexists with automated data flows from connected systems (level 2). APIs and middleware solutions facilitate these connections, allowing data to flow between previously siloed applications. Further along the spectrum, systems begin to prioritize automated data collection with only occasional manual intervention for exceptions or special cases (level 3).

The advanced stages of data acquisition automation leverage conversational AI interfaces that can interpret natural language inputs (level 4). These systems can extract relevant information from conversations, emails, or other unstructured communications, converting them into structured data entries. Virtual assistants and chatbots exemplify this level, where users can simply tell the system what they need rather than navigating complex form interfaces. At the highest level of automation, we find truly intelligent data extraction capabilities (level 5). These systems can process unstructured documents like invoices, contracts, or reports and autonomously extract relevant information based on AI.

Advanced computer vision and natural language processing algorithms work in concert to understand document context, identify key fields, and translate visual or textual information into structured data. For instance, a system might analyse a scanned purchase order, identify line items, quantities, prices, and vendor information, then create corresponding entries in the ERP system without human intervention. This progression reflects a fundamental shift from humans serving as data entry operators to becoming exception handlers who intervene only when the AI encounters novel or ambiguous situations. The system gradually learns from these interventions, continuously improving its accuracy and reducing the need for human involvement over time.

5.2 Information analysis

Information Analysis transforms raw data into actionable intelligence. This dimension progresses from simple reporting to autonomous learning and adaptation. At its most basic level, descriptive analysis simply presents what has occurred—historical reports, dashboards showing past performance, or basic metrics that document system states (level 1). This level answers the question what happened? but requires humans to interpret meaning and implications. Moving up the automation scale, diagnostic analysis begins to answer why did it happen? by identifying patterns and correlations (level 2). Statistical techniques help identify relationships between variables, allowing users to understand cause-and-effect relationships within their business processes.

For example, a system might not only show declining sales but also highlight correlations with specific market conditions, pricing changes, or competitive activities. Predictive analysis represents a significant leap forward, shifting focus from past events to future possibilities (level 3) answering what will occur?. Using statistical models and historical patterns, these systems forecast trends, anticipate outcomes, and model scenarios. A predictive ERP component might forecast inventory requirements based on historical sales patterns, seasonal variations, and current market trends, helping organizations prepare proactively rather than reacting to shortages. Prescriptive analysis builds upon prediction by recommending optimal actions to be decided on (level 4). The focus is on what actions should be taken?.

These systems evaluate multiple scenarios and their likely outcomes, then suggest the most beneficial course of action. For instance, rather than merely predicting a supply shortage, a prescriptive system might recommend specific adjustments to production schedules, suggest alternative suppliers, or propose redistribution of existing inventory across locations to minimize disruption. At the pinnacle of information analysis automation, cognitive systems employ advanced AI techniques to autonomously improve their analytical capabilities (level 5). These systems detect patterns humans might miss, adapt to changing conditions without reprogramming, and continuously refine their models based on new data and outcomes.

They might, for example, automatically detect emerging market trends before they become obvious, identify subtle shifts in customer preferences, or predict maintenance needs before equipment shows signs of failure. The evolution from descriptive to cognitive analysis transforms ERP systems from passive record-keepers to proactive business advisors and eventually to autonomous analytical engines that continuously generate new insights without human direction.

5.3 Decision making

Decision Making involves selecting among alternatives to achieve organizational objectives. In its most basic form, decision making in ERP systems involves presenting information to human users who then evaluate options and make choices based on their experience, judgment, and organizational guidelines (level 1). The system serves merely as an information repository, with all interpretative and decision-making tasks left to humans. As automation increases, systems begin to provide context-aware information that supports human decision making (level 2). The ERP system might highlight relevant factors, provide historical precedents for similar situations, or present information in ways that make trade-offs more apparent.

For example, when approving a purchase requisition, the system might display budget constraints, alternative vendors, or potential impacts on production schedules—information that helps the human make a more informed decision. At intermediate levels, ERP systems actively evaluate options and provide specific recommendations (level 3). Using rules or heuristics, these systems can suggest optimal choices based on organizational preferences and goals. A procurement module might recommend specific suppliers based on an algorithm that weighs price, quality ratings, delivery reliability, and sustainability factors according to the organization’s priority weights. Advanced decision support systems incorporate sophisticated modelling capabilities that can simulate the outcomes of different decisions across multiple dimensions (level 4).

These systems might use techniques like Monte Carlo simulation or optimization algorithms to identify decisions that maximize desired outcomes while minimizing risks or undesired consequences. At the highest level of automation, systems make decisions autonomously within defined parameters (level 5). These decisions are traceable and auditable, allowing for human oversight and intervention when necessary. For example, an autonomous procurement system might automatically reorder supplies when inventory reaches predetermined thresholds, select vendors based on current conditions and organizational policies, negotiate terms within approved parameters, and issue purchase orders—all without human intervention for routine cases.

The progression in decision making automation represents a gradual transfer of judgment from humans to algorithms, with increasing emphasis on automated decision logic that can process more variables, consider more alternatives, and operate more consistently than human decision-makers. However, human oversight remains critical, especially for decisions with significant consequences or ethical dimensions.

5.4 Action execution

Action Execution translates decisions into concrete operations that achieve business objectives. At the most basic level, humans manually implement all actions after system notifications or information retrieval (level 1). For example, after viewing an inventory report showing low stock levels, an employee might manually create purchase requisitions, contact suppliers, and update system records. As automation increases, systems begin providing not just notification of needed actions but also specific guidance on execution steps (level 2). The system might generate templates, provide checklists, or offer structured workflows that guide users through the necessary actions in the correct sequence. This level retains human execution but reduces cognitive load and improves consistency.

Mid-level automation introduces partial execution capabilities, where systems can perform some actions automatically while requiring human intervention for others (level 3). For example, a system might automatically generate purchase orders based on inventory levels but require human approval before sending them to suppliers. This hybrid approach balances efficiency with control, particularly for actions with financial implications. Advanced execution systems actively recommend specific action sequences based on situational analysis (level 4). These systems evaluate multiple possible action paths and suggest the most effective approach given current conditions and constraints.

For instance, when dealing with a production delay, the system might recommend rescheduling specific orders, notifying affected customers, and recalculating delivery dates based on an algorithm that minimizes overall impact. At the highest level of automation, systems autonomously execute complete action sequences without human intervention (level 5). These autonomous actors can initiate workflows, communicate with internal and external systems, make adjustments based on real-time feedback, and document all actions for accountability and learning purposes. For example, a procurement AI agent might detect low inventory, compare vendors, place orders, track shipments, process receipts, and update financial records—a complete process chain executed without direct human involvement.

The evolution in action execution transforms ERP systems from passive information repositories to active agents capable of implementing business decisions. This progression redefines the role of human operators from task executors to exception handlers and strategists who focus on unique situations or high-level oversight while routine operations proceed autonomously.

Organizations can adjust the introduced criteria for the automation levels to their needs. The four dimensions—data acquisition, information analysis, decision making, and action execution—represent the fundamental aspects of business process automation. However, they should not be viewed in isolation. They form an interconnected continuum where advancements in one dimension often enable or necessitate advancements in others. For instance, improved data acquisition through IoT sensors or computer vision might generate data volumes that exceed human analytical capabilities, driving the need for more advanced information analysis. Similarly, sophisticated analytical capabilities might identify decision opportunities that occur too rapidly for human decision-makers, necessitating automated decision making capabilities.

The methodology’s rating system allows organizations to create a nuanced profile of their current automation state across all four dimensions. This profile often reveals imbalances—such as advanced data acquisition paired with rudimentary decision making—that can guide strategic investment in AI enhancement. The goal is not necessarily to maximize automation in all dimensions simultaneously, but rather to create a balanced profile aligned with organizational goals, regulatory requirements, and risk tolerance. As organizations implement this methodology, they might discover that incremental advances in automation can yield significant business benefits without requiring revolutionary changes.

By assessing current states based on the decision criteria of the automation levels, defining target states, and implementing appropriate technologies, customers and ERP vendors can progressively incorporate AI into their ERP solutions, ultimately transforming them from passive transaction processors to active business partners that continuously learn, adapt, and optimize operations.

6 Technology categorization

To operationalize the proposed automation framework, it is necessary to identify categories of AI technologies that can support automation across ERP business processes. The following technology categorization links AI capabilities with the four automation dimensions introduced in the framework. As illustrated in Fig. 5 our methodology identifies three broad categories of technological approaches for bridging these gaps. Manual processes represent the baseline, where human operators perform tasks with minimal system support. Rule-based approaches codify explicit logic for handling routine situations. Self-learning technologies employ AI techniques to derive insights and behaviors from data rather than explicit programming. The boundary between rule-based and self-learning technologies is not as explicit as illustrated in Fig. 5 for simplification reasons, but rather fluid.

6.1 Rule-based

Rule-based technologies have long been the workhorses of ERP intelligence. Custom reports, alerts, and workflows embody rule-based logic that helps organizations move from manual operation (level 1) to basic automation (levels 2—4). These technologies are well-understood, relatively straightforward to implement, and produce predictable results. Examples include:

• Java programs that validate inputs and guide error resolution represent rule-based automation in data acquisition and information analysis. These reports apply predefined validation rules to identify issues and provide standardized guidance for resolution.

• Workflow engines that orchestrate process steps exemplify rule-based decision-making and action execution. They define conditions for progression, route items to appropriate personnel, and ensure consistent process execution.

• Analytics scenarios that highlight KPI deviations and suggest corrective actions combine rule-based information analysis with basic decision support. These scenarios detect predefined patterns and trigger standardized responses.

While rule-based approaches are valuable and often sufficient for levels 2—4, they face inherent limitations in handling novel situations, adapting to changing conditions, discovering non-obvious patterns, generating content or resolving issues autonomously. To achieve level 5 automation, self-learning technologies become essential. Self-learning technologies employ various AI techniques to develop capabilities that would be impractical to program explicitly. From our analysis of use cases in the context of intelligent ERP, four key categories of self-learning technologies are dominating: narrow AI, generative AI, conversational AI, and agentic AI. While prior studies mainly discuss individual AI applications in enterprise systems, this study focuses on categorizing AI technologies according to their role in automating ERP-supported business processes.

6.2 Narrow AI

Narrow AI represents the most established approach to incorporating AI into ERP systems. These models are trained on focused datasets to perform well-defined functions within business processes. Unlike general-purpose models, these specialized AI implementations excel at particular tasks through concentrated training on domain-specific data. In case of narrow AI, we often omit the prefix and just utilize AI. In the context of ERP systems, task-specific models might include demand forecasting algorithms that analyze historical sales patterns, seasonality factors, and market indicators to predict future inventory requirements. They could also encompass anomaly detection systems that identify unusual patterns in financial transactions that might indicate errors or fraud.

Another application involves predictive maintenance models that analyze equipment sensor data to anticipate failures before they occur or image recognition for quality inspection. The implementation of task-specific models within ERP systems typically follows a structured workflow. First, relevant historical data is extracted from the ERP database and potentially enriched with external sources. This data undergoes cleaning and transformation to make it suitable for model training. Data scientists then develop and train various algorithmic approaches, comparing their performance to select the most effective model. The chosen model is then integrated into the ERP system, often through

API connections or embedded modules that allow the model to receive real-time data and return predictions or classifications. What makes task-specific models particularly valuable for ERP intelligence is their ability to deliver high accuracy within narrow domains. By focusing on specific business problems, these models can be trained with relatively modest data requirements compared to more general AI systems. They also tend to produce more interpretable results, which is crucial in business contexts where stakeholders need to understand the reasoning behind automated decisions. However, task-specific models come with limitations. Each model typically serves only one function, requiring organizations to develop and maintain numerous models to enhance multiple process dimensions.

They also require retraining as business conditions change, and their effectiveness is heavily dependent on the quality and relevance of their training data. Despite these challenges, task-specific models remain the workhorses of ERP intelligence, providing reliable enhancements across all four automation dimensions.

6.3 Generative AI

Generative AI represents a paradigm shift in how we think about artificial intelligence in enterprise systems. Unlike traditional models that primarily classify or predict based on historical patterns, generative AI creates new content, suggestions, or approaches that might not directly mirror past examples. This capability opens entirely new possibilities for ERP intelligence. In data acquisition, generative AI can create structured data entries from unstructured inputs. For example, it might generate complete purchase order data from an email conversation or compose standardized documentation from scattered information sources. These capabilities reduce the burden of manual data entry while improving data completeness and consistency.

For information analysis, generative models can produce comprehensive explanations of complex data patterns, creating narrative summaries of business performance that highlight key insights in natural language. They can also generate multiple scenario analyses by producing variations of business conditions and their potential outcomes, enabling more robust planning. In decision making processes, generative AI can propose novel solutions to business problems rather than simply selecting from predefined options. For instance, when faced with a supply chain disruption, a generative system might create an entirely new routing strategy that combines elements of multiple standard approaches in innovative ways. This creative problem-solving capability is particularly valuable for handling exceptional situations where standard procedures prove inadequate.

For action execution, generative AI can create customized communication content, tailor process workflows to specific situations, and even generate code to automate unique business scenarios. These capabilities allow ERP systems to adapt their execution strategies to highly specific contexts rather than forcing all situations into standardized processing paths. The integration of generative AI into ERP systems is more complex than traditional models. These systems typically require more substantial computing resources, specialized integration interfaces, and careful governance mechanisms to ensure their outputs remain within appropriate boundaries. However, their potential to transform ERP intelligence is profound, particularly for handling complexity, ambiguity, and novel situations that rule-based systems cannot address effectively.

6.4 Conversational AI

Conversational AI transforms how users interact with ERP systems by enabling natural language interfaces that interpret, respond to, and act upon human communication. This technology fundamentally reshapes the data acquisition dimension while also influencing how information is analysed and presented. Traditional ERP interfaces require users to navigate complex forms, menus, and data entry screens that often reflect the system’s database structure rather than natural business thinking. Conversational interfaces invert this paradigm, allowing users to express their needs in natural language while the system handles the translation to appropriate system actions. In practice, conversational AI within ERP systems can manifest in multiple forms. Virtual assistants might help users locate information, initiate processes, or complete tasks through natural dialogue rather than menu navigation.

Chatbots integrated into workflow applications can guide users through complex procedures, answer questions about process requirements, or capture information through conversational exchanges rather than forms. Voice interfaces might allow hands-free operation in environments like warehouses or manufacturing floors where keyboard access is impractical. Beyond simple command processing, advanced conversational systems can maintain context across interactions, understand complex instructions with multiple parameters, and disambiguate unclear requests through clarifying questions. They can also adapt to different users’ communication styles, technical knowledge levels, and job responsibilities, providing personalized assistance based on user profiles and interaction history.

The implementation of conversational interfaces in ERP systems requires multiple technological components working in concert. Natural language understanding (NLU) modules interpret user input, converting free-form text or speech into structured intent and parameter representations. Dialog management systems maintain conversation state and determine appropriate responses based on both the current request and conversation history. Natural language generation (NLG) components produce coherent, contextually appropriate responses. Integration layers connect these conversational capabilities with core ERP functionality, translating recognized intents into system actions. The value of conversational AI extends beyond mere convenience. By lowering the technical barriers to system interaction, these interfaces make ERP functionality accessible to users with limited system training.

They can accelerate common tasks by eliminating navigation steps and form completion. Perhaps most importantly, they can capture information at the point of observation rather than requiring users to remember details until they can access a traditional interface.

6.5 Agentic AI

Agentic AI represents the frontier of ERP intelligence, where systems transition from tools that assist humans to autonomous actors that independently pursue business objectives. These systems combine perception, reasoning, planning, and action capabilities to manage entire business processes with minimal human intervention. Unlike traditional automation that follows predefined paths, agentic systems operate with goal-directed autonomy. They understand business objectives, develop plans to achieve those objectives, execute required actions across multiple systems, monitor results, and adapt their approaches based on outcomes and changing conditions. This represents the highest level of automation across all four dimensions of the methodology. In practical

ERP implementations, agentic AI might manifest as autonomous procurement systems that monitor inventory levels, predict future requirements, identify optimal suppliers, negotiate terms, place orders, and manage the entire procurement lifecycle without human intervention for routine cases. We might see financial agents that continuously optimize cash management by analysing incoming and outgoing payment flows, identifying investment opportunities, and executing transactions within defined risk parameters. Or we could encounter customer service agents that autonomously handle support requests, accessing multiple information systems to diagnose issues and implement solutions. The technological architecture of agentic AI in ERP contexts typically involves multiple interconnected components. Perception modules gather and interpret data from various sources within and beyond the ERP system.

Reasoning engines assess situations, identify challenges, and evaluate potential approaches. Planning components develop multi-step action sequences to achieve objectives. Execution modules carry out actions through APIs or direct system integration. Learning systems monitor outcomes and refine strategies based on results. What distinguishes truly agentic systems from sophisticated automation is their ability to handle ambiguity, adapt to novel situations, and operate effectively outside narrowly defined scenarios. When faced with unexpected conditions, an agentic system doesn’t simply halt and wait for human intervention— it attempts to understand the situation, adapt its approach, and continue pursuing its objectives within its authority constraints. The implementation of agentic AI in ERP systems introduces significant challenges beyond technical complexity.

Governance frameworks must ensure these autonomous systems operate within appropriate boundaries while allowing sufficient flexibility to deliver value. Transparency mechanisms must make agent reasoning and actions understandable to human overseers. And control systems must enable rapid human intervention, when necessary, without undermining the agents’ autonomy for routine operations.

While these four categories of self-learning technologies offer distinct capabilities, their greatest potential emerges through integration. Task-specific models might provide specialized analytical capabilities that feed into generative systems for solution creation. Conversational interfaces might serve as the primary human-system interaction channel for both providing input to and receiving guidance from agentic processes. And agents might orchestrate multiple task-specific models to accomplish complex business objectives. The progressive implementation of these technologies typically follows a maturity path. Organizations often begin with task-specific models for well-defined, high-value processes where data quality and availability are strong.

As they develop capabilities and confidence, they might introduce conversational interfaces to make these capabilities more accessible to business users. Generative capabilities might follow to handle more complex or ambiguous situations. Finally, agentic approaches might integrate these components into autonomous processes for selected business functions. This progression aligns with the automation levels outlined in the methodology. Task-specific models typically support advancement to level 5 across the four dimensions. Conversational AI primarily enhances the data acquisition dimension but can influence all four through improved user interaction. Generative AI can push capabilities toward level 5 in all four dimensions. And agentic approaches represent the ultimate expression of level 5 automation across all dimensions.

Table 2 summarizes the impact of the different AI technologies on ERP automation levels. The future of ERP intelligence likely involves a hybrid ecosystem where these various self-learning technologies complement each other and integrate with traditional rule-based systems to create comprehensive intelligent capabilities tailored to each organization’s specific needs and readiness. The methodology provides a valuable framework for navigating this complex landscape, helping organizations assess their current state, define their aspirations, and chart a path through these technological possibilities to achieve meaningful business outcomes. The ultimate technological vision suggested by the methodology is an autonomous ERP system where intelligence is embedded throughout business processes rather than added as an afterthought.

In this vision, the system continuously learns from every transaction, adapts to changing conditions, anticipates needs, and takes appropriate actions—all while maintaining auditability and alignment with business objectives. Achieving this vision requires not just adoption of new technologies but also evolution of ERP architectures to support AI as a core capability rather than an add-on feature. The methodology provides a roadmap for this evolution, helping ERP vendors and customers identify where and how to incorporate AI capabilities for maximum business impact.

In contrast to prior research that primarily described automation levels, our framework introduces a measurable assessment mechanism. The geometric representation, where the automation levels across four dimensions form a rectangle and its area reflects the overall automation degree of a process, provides a quantifiable metric for ERP automation. This quantification enables aggregation across processes and facilitates comparison across organizations, offering both researchers and practitioners a structured approach for benchmarking AI adoption. Furthermore, unlike descriptive taxonomies, our model links automation dimensions with concrete categories of AI technologies (narrow AI, generative AI, conversational AI, and agentic AI). This integration allows for a differentiated analysis of how distinct AI approaches asymmetrically impact ERP automation.

For example, conversational AI is more influential in data acquisition, while generative AI has transformative potential in information analysis and action execution. Such mappings are absent in earlier automation frameworks.

7 Managerial guidance

The integration of intelligence into ERP systems isn’t merely a technological advancement—it’s fundamentally a business transformation strategy. Organizations implement ERP systems with the primary goal of enhancing revenue generation and profit margins through improved operational efficiency. Intelligence amplifies these benefits by creating new types of business value. When we examine the business perspective on intelligence enhancement, we see that companies across all industries and sizes share common objectives. They seek to leverage their data assets more effectively, automate routine decisions, minimize operational disruptions, and create competitive advantages through superior insights and responsiveness. The methodology provides a framework for structured dialogue about these aspirations.

What makes this methodology particularly valuable from a business standpoint is its ability to facilitate objective conversations between diverse stakeholders. Enterprise software implementations typically involve multiple parties with different perspectives—business process owners, IT departments, executive sponsors, end users, and ERP vendors. Each group may have different understandings of what AI incorporation means in practical terms. By breaking these concepts into specific dimensions and measurable levels, the methodology creates a common language for these stakeholders. The context-dependent nature of AI value is especially important to understand. In some business processes, even modest enhancements can deliver substantial returns. For example, in manufacturing environments with stable processes, even rule-based quality control automation might reduce defect rates significantly.

In contrast, highly variable environments like fashion retail might require advanced predictive capabilities to deliver meaningful value in demand forecasting. The methodology acknowledges this contextual nature by focusing on the transition between current and future states rather than prescribing universal targets. Value creation through AI manifests in multiple forms. Direct cost reduction occurs when automation reduces labor requirements for routine tasks. Revenue enhancement happens when intelligent ERPs identify sales opportunities or optimize pricing. Risk reduction occurs when predictive capabilities anticipate disruptions or compliance issues before they materialize. Customer experience improvements arise when systems personalize interactions or respond more quickly to needs.

The methodology helps organizations articulate which of these value types they prioritize and where they expect to realize them. Perhaps most importantly, the business view recognizes that AI enhancement is a journey rather than a destination. Organizations typically begin with focused implementations that target specific high-value processes. As they develop capabilities and confidence, they gradually expand the scope and sophistication of their AI initiatives.

The methodology supports this incremental approach by allowing organizations to map their progression across dimensions and processes over time. From a technological perspective, the methodology serves as a bridge between business aspirations and technical implementation. It helps solution architects and technology teams translate abstract concepts like increased AI augmentation into specific capabilities that must be developed or acquired. As shown in Fig. 6, the first step in this process is assessment—determining the current automation level for each dimension of an ERP business process. This assessment isn’t merely technical but involves understanding how users currently interact with the ERP business process, what decisions they make, and how actions are executed.

Technology teams might employ techniques like process mining, user observation, system logs analysis, or workflow documentation to establish this baseline. Once the current state is established, solution managers can work with business stakeholders to determine target levels based on business requirements, technology constraints, and organizational readiness. The gap between current and target states becomes the scope for realization initiatives.

Trade-offs must be made between investments in different automation dimensions. Enhancing decision-making capabilities (e.g., with generative AI for scenario planning) may deliver strategic benefits but often requires costly data governance improvements. Conversely, investing in conversational AI for data acquisition might be less expensive and yield immediate usability gains but contribute less to long-term competitive differentiation. CIOs and ERP managers should therefore align AI adoption pathways with enterprise strategy—cost efficiency, customer intimacy, or innovation leadership—and use the framework to visualize trade-offs. Organizations must anticipate barriers to AI-enabled ERP transformation. These include:

• Technical barriers such as poor data quality, lack of interoperability between ERP and AI modules, and lifecycle challenges.

• Organizational barriers including user resistance, skill shortages, and change management challenges.

• Regulatory and ethical barriers around transparency, compliance, and explainability of AI-driven decisions.

By surfacing these barriers explicitly, the framework shifts from being descriptive (“what levels exist”) to prescriptive (“what must be overcome to reach higher levels”). Adoption should be approached as a staged journey rather than a one-off transformation. Enterprises can pursue different maturity pathways, for example:

• Incremental pathway: Begin with narrow AI applications in data acquisition and information analysis, then gradually extend into decision making and action execution.

• Leapfrogging pathway: Invest directly in generative or agentic AI for selected high-value processes (e.g., autonomous procurement), while leaving other dimensions at lower levels.

• Balanced pathway: Advance all four dimensions in parallel to maintain system coherence, especially in highly regulated industries.

The framework offers practical guidance not only for enterprises but also for ERP vendors. Vendors can use the model to design AI-enhanced modules, benchmark their offerings against competitors, and communicate value to customers in terms of automation progression. Thus, the practical contribution of this framework lies in its ability to translate abstract automation dimensions into actionable roadmaps. By helping enterprises make trade-offs, anticipate barriers, and select appropriate adoption pathways, it supports both strategic alignment and operational implementation of AI in ERP systems.

8 Evaluation

For the evaluation of our framework, we selected SAP ERP as a widespread product. The evaluation of our systematic AI incorporation methodology employed multiple assessment criteria examining both theoretical soundness and practical applicability. Completeness evaluated whether the methodology addresses the essential aspects of AI integration into ERP systems. The four-dimensional framework captures fundamental components of business process automation, while the five-level scale provides sufficient granularity for meaningful differentiation without excessive complexity. Usability assessed practical applicability across different organizational contexts, while scalability examined the framework’s effectiveness across varying organizational sizes, industries, and process complexities.

Effective measurement of automation levels required reliable and consistent methods that translate conceptual frameworks into measurable criteria. The measurement approach is typically dependent on the specific requirements of customers and vendors. Thus, it should be individually defined for an organization and uniformly applied for all business processes. For our evaluation we utilized a multi-source approach that triangulates assessment data from system documentation, user feedback, process mining data, and stakeholder interviews. This approach mitigates limitations of any single measurement method while building confidence in assessment accuracy. We could successfully apply our framework across different business processes and industry contexts of SAP ERP, ensuring the methodology provided actionable guidance for systematically embedding AI.

For illustration we depict one example of a procurement business process. This process encompasses the complete procurement lifecycle from identifying sourcing needs through supplier payment, representing a complex, multi-stakeholder workflow that spans multiple organizational functions and external relationships.

First, the current state of the procurement business process was assessed. Data Acquisition originally operated at level 2 (hybrid manual-automated). Purchase requisitions were manually created by requesters through web forms, though some automated data flows existed from inventory management systems that triggered reorder notifications. Supplier information was maintained through manual data entry with periodic updates from supplier portals. Invoice processing relied heavily on manual data entry from paper or PDF documents, with basic OCR capabilities for standard invoice formats. Approximately 70% of data entry required human intervention, indicating significant manual dependency. Information Analysis functioned at level 2 (diagnostic analysis). The system provided standard procurement reports showing spending patterns, supplier performance metrics, and budget utilization.

Basic exception reporting identified overdue purchase orders and budget variances. However, analysis of supplier risk, market price trends, and procurement optimization opportunities required manual investigation by procurement specialists. The system lacked predictive capabilities for demand forecasting or supplier performance prediction. Decision Making operated primarily at level 1 (human-dependent) with limited level 2 capabilities. All procurement decisions required human approval through defined approval workflows. Supplier selection relied on manual evaluation of quotes and supplier performance history. Contract negotiations were conducted entirely through human interaction with minimal system support. The system provided decision support through historical data compilation but lacked recommendation engines or optimization algorithms.

Action Execution demonstrated level 2 (guided execution) capabilities. The system automated purchase order generation and basic workflow routing but required human intervention for most execution steps. Integration with supplier systems enabled automated purchase order transmission for select vendors. Payment processing involved automated payment calculation but required manual approval and execution. Exception handling relied entirely on human intervention. In conclusion, the current automation profile created a rectangle with coordinates, representing an area of 4 units out of a maximum possible 25 units, indicating 16% overall automation.

Second, the target state was defined based on customer requirements and business objectives to reduce procurement costs by 15%, improve supplier compliance by 30%, and decrease process cycle time by 40%. Thus, Data Acquisition targeted level 4 (conversational AI-enabled). This included implementing intelligent document processing for invoices and contracts, natural language interfaces for requisition creation, and automated supplier data synchronization. The target assumes 90% automated data acquisition with human intervention only for exceptional cases. Information Analysis targeted level 4 (prescriptive analysis). Planned capabilities included predictive analytics for demand forecasting, supplier risk assessment algorithms, market price optimization, and automated spend analysis with actionable recommendations.

The system had to provide specific procurement recommendations based on comprehensive data analysis across multiple variables. Decision Making targeted level 3 (rule-based decisions) for routine procurements and level 4 (algorithmic optimization) for strategic sourcing. Automated approval workflows had to handle standard requisitions within predefined parameters. Supplier selection had to incorporate automated scoring algorithms that considered price, quality, delivery performance, and risk factors. Contract negotiations had to leverage automated benchmarking and recommendation engines. Action Execution targeted level 4 (orchestrated automation). The system had to autonomously execute routine procurement activities including purchase order processing, supplier communication, receipt processing, and payment execution.

Exception handling had to employ automated escalation and resolution procedures with human intervention only for complex scenarios. In conclusion, the target automation profile created a rectangle with coordinates (4, 4, 3.5, 4), representing an area of 22.4 units, indicating 89.6% overall automation—a substantial improvement from the previous 16% level.

The transformation from current to target state employed a progressive implementation approach that built capabilities incrementally while delivering measurable business value at each stage. Phase 1 focused on foundational improvements in data acquisition and basic analytics. Intelligent document processing reduced manual invoice entry by 80%, while enhanced reporting provided better visibility into procurement patterns. Implementation duration was 6 months with an expected ROI of 15% through reduced manual effort. Phase 2 introduced predictive analytics and automated decision support. Machine learning models provided demand forecasting and supplier performance prediction, while recommendation engines supported sourcing decisions. Implementation duration was 9 months with an expected ROI of 25% through improved procurement efficiency and supplier selection.

Phase 3 implemented advanced automation and orchestration capabilities. Automated approval workflows handled routine requisitions, while sophisticated optimization algorithms support strategic sourcing. Orchestrated execution automated end-to-end procurement processes for standard items. Implementation duration was 12 months with an expected ROI of 35% through comprehensive process optimization. The phased approach allowed for learning and adaptation while building organizational capability and confidence in AI-enhanced procurement processes.

Application of the methodology to the procurement use case demonstrates several key validation points. Framework completeness was confirmed through the ability to comprehensively assess and plan the transformation across all critical aspects of the procurement process. The four dimensions captured all essential automation opportunities, while the five-level scale provided appropriate granularity for planning and measurement. Practical applicability was demonstrated through the framework’s ability to translate business objectives into specific automation targets and technology requirements. The methodology facilitated objective dialogue between business stakeholders and technical teams by providing a common language for automation assessment and planning.

Implementation guidance emerged through the gap analysis process, which identified specific technology requirements and suggested logical implementation sequences. The framework helped prioritize automation investments based on business impact and implementation complexity. Progress measurement became possible through the quantitative automation profiles that enabled tracking advancement over time. The geometric representation provided visual progress indicators that communicated transformation status effectively to stakeholders.

To evaluate the practical usefulness of the proposed framework, organizations can assess improvements across several measurable indicators. Examples include reductions in process execution time, decreases in manual intervention rates, improvements in decision accuracy, and lower operational costs resulting from higher automation levels. By comparing the automation rectangle areas before and after the introduction of AI-enabled capabilities, organizations can quantify the relative improvement in process automation maturity. The appropriate level of automation for a specific ERP process depends on several contextual factors.

Processes characterized by high transaction volumes and structured data are particularly suitable for higher automation levels, while processes involving complex judgment, regulatory oversight, or strategic decision-making may require lower automation levels with continued human supervision. The framework therefore provides a structured method for analysing how AI technologies can be introduced progressively while maintaining appropriate levels of human control. Improvements resulting from AI incorporation can be measured through indicators such as reductions in process execution time, decreased manual intervention rates, lower operational costs, and improved decision accuracy.

9 Conclusion

The systematic incorporation of AI into ERP systems represents a transformative opportunity for organizations across industries. Our proposed methodology provides a structured approach to this complex challenge by breaking down business processes into four fundamental dimensions and establishing clear automation levels for each. This approach has been successfully validated on numerous real-world ERP business processes and resulted in effective and efficient incorporation of AI capabilities into all those in-between productively used AI features. Thus, the framework enables ERP vendors and customers to move beyond ad hoc AI initiatives toward a comprehensive strategy that balances technological capabilities with business objectives. The journey toward intelligent ERP is inherently incremental.

Organizations typically begin with focused implementations that target specific high-value processes using narrow AI applications. As they develop capabilities and confidence, they gradually expand to incorporate more sophisticated technologies like generative AI, conversational AI, and eventually agentic AI. This progressive approach allows organizations to build on successes, learn from challenges, and adapt their strategies as both technologies and business needs evolve.

The business value of intelligent ERP manifests in multiple forms—direct cost reduction through automation, revenue enhancement through optimized decision making, risk reduction through predictive capabilities, and customer experience improvements through personalization and responsiveness. The methodology helps organizations articulate which of these value types they prioritize and where they expect to realize them, facilitating more targeted investments and clearer success metrics. From a technological perspective, our analysis of self-learning technologies provides a roadmap for implementation.

While each category offers distinct capabilities, their greatest potential emerges through integration—task-specific models feeding into generative systems, conversational interfaces providing human-system interaction, and agents orchestrating multiple components to accomplish complex business objectives.

The ultimate vision suggested by our methodology is an autonomous ERP system where intelligence is embedded throughout business processes rather than added as an afterthought. In this vision, the system continuously learns from every transaction, adapts to changing conditions, anticipates needs, and takes appropriate actions—all while maintaining auditability and alignment with business objectives. Despite these contributions, the study has several limitations. The proposed framework has been applied to real-world ERP environments to demonstrate its practical feasibility, particularly through analysis of AI capabilities in contemporary ERP systems. However, the evaluation remains primarily illustrative and does not yet include large-scale empirical validation across multiple organizations, industries, or ERP platforms.

Future research could therefore apply the framework in comparative case studies or empirical analyses to assess its effectiveness in different organizational contexts and to further refine the measurement of automation maturity across ERP-supported business processes. Future research should focus on developing more granular assessment tools for each dimension, establishing industry-specific benchmarks for automation levels, and creating detailed implementation roadmaps for different organizational contexts. Additionally, as AI technologies continue to evolve, the methodology will need ongoing refinement to incorporate new capabilities and approaches.

By providing a common language and structured framework for AI incorporation, this methodology bridges the gap between technical specialists and business stakeholders, enabling more productive collaboration and more successful implementations. As organizations continue to navigate digital transformation initiatives, this systematic approach to intelligent ERP will become increasingly valuable in creating sustainable competitive advantages and driving business innovation.

The theoretical contribution of this framework is therefore threefold: Adaptation of generic automation theory into the ERP domain, accounting for process integration, structured data models, and enterprise-wide scope, Extension of automation theory from descriptive classification to measurable assessment through the rectangle–area method, and Integration of AI technology categories into the automation framework, enabling differentiated insights into how AI drives ERP transformation. From a practical standpoint, the framework contributes to managerial decision-making by: (a) providing a method for diagnosing current ERP automation states, (b) offering a roadmap for targeted AI investments across dimensions, and (c) facilitating cross-industry benchmarking of AI-enhanced ERP maturity.

Siar Sarferaz wrote the main manuscript text based on his research results, prepared the figures and tables.

Data availability.

All data generated or analysed during this study are included in this published article.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Clinical trial number

Not applicable.

Competing interests

2. Klaus H, Rosemann M, Gable G.

What is ERP? Inf Syst Front. 2000;2:141–62.

3. Haddara M, Elragal A.

The future of ERP systems: look backward before moving forward. Procedia Technol. 2012;5:21–30.

4. O’Leary DE.

Enterprise resource planning systems: systems, life cycle, electronic commerce, and risk. Cambridge: Cambridge University Press; 2000.

5. Kaplan A, Haenlein M.

Siri, Siri in my hand: who’s the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence. Bus Horiz. 2019;62:15–25.

6. Russell S, Norvig P.

Artificial intelligence: a modern approach. 4th ed. London: Pearson; 2021.

7. McCarthy J.

What is artificial intelligence? Stanford: Stanford University; 2007.

8. Minsky M.

The society of mind. New York: Simon & Schuster; 1986.

9. European Commission’s High-Level Expert Group on AI.

A definition of AI: main capabilities and scientific disciplines. Brussels: European Commission; 2019.

10. Brynjolfsson E, McAfee A.

The second machine age. New York: W. W. Norton; 2014.

11. Markus ML, Tanis C.

The enterprise systems experience—from adoption to success. In: Framing the domains of IT management, Cincinnati: Pinnaflex Educational Resources; 2000.

12. Al-Mashari A, Al-Mudimigh M, Zairi M.

Enterprise resource planning: a taxonomy of critical factors. Eur J Oper Res.

13. Nah FF-H, Lau JL-S, Kuang J.

Critical factors for successful implementation of enterprise systems. Bus Process Manag J.

2001;7:285–96.

14. Johansson J, Ruivo P, Oliveira T, Neto M.

The determinants of cloud ERP adoption. Comput Ind. 2016;79:1–10.

15. Goodfellow I, Bengio Y, Courville A.

Deep learning. Cambridge: MIT Press; 2016.

16. Davenport TH, Ronanki R.

Artificial intelligence for the real world. Harv Bus Rev. 2018;96:108–16.

17. Huang M-H, Rust RT.

Artificial intelligence in service. J Serv Res. 2018;21:155–72.

18. Dwivedi YK, et al.

Artificial intelligence (AI): multidisciplinary perspectives on emerging challenges, opportunities, and

19. Rai A, Constantinides S, Sarker S.

Next-generation digital platforms: toward human–AI hybrids. MIS Q. 2021;45:1–18.

20. Seddon PB, Calvert G, Yang S.

A multi-project model of key factors affecting organizational benefits from enterprise

21. Wamba-Taguimdje SA, Fosso Wamba S, Kala Kamdjoug J, Tchatchouang Wanko T.

Influence of artificial intelligence on firm

22. Elragal A.

ERP and artificial intelligence. Enterp Inform Sys, 2022;16.

23. Dziembek D, Turek T.

A model for integrating artificial intelligence with ERP systems—towards autonomous business

24. Pokala P.

Artificial intelligence in enterprise resource planning: a systematic review of innovations, applications, and future

25. Mustafa BS, Zeebaree SRM.

AI-driven innovations in enterprise systems. Int J Sci World. 2025;11:127–36.

26. Sunkara SP.

AI-powered CRM and ERP systems: transforming business operations through smart technology. World J Adv

27. Parasuraman R, Sheridan TB, Wickens CD.

A model for types and levels of human interaction with automation. IEEE Trans

28. Hong W, Chan FKY, Thong JYL, Chasalow LC, Dhillon G.

A framework and guidelines for context-specific theorizing in

29. Kitchenham B, Chartes SM.

Guidelines for performing Systematic Literature Reviews in Software Engineering. EBSE Technical Report EBSE-2007-01, Version 2.3; 2007.

30. Moher D, Liberati A, Tetzlaff J, Altman DG.

Preferred reporting items for systematic reviews and meta-analyses: the PRISMA 31. Porter ME

Competitive advantage: creating and sustaining superior performance. New York: Free Press; 1985.

32. SAP SE

Product roadmap. [Online]. the linked source and search for “Artificial Intelligence” as “Focus Topic” 33. SAP SE

Discovery center, SAP Business AI Features. [Online]. ​h​t​t​p​s​:​/​/​d​i​s​c​o​v​e​r​y​-​c​e​n​t​e​r​.​c​l​o​u​d​.​s​a​p​/​a​i​-​c​a​t​a​l​o​g​/ 34. Sarferaz S

Compendium on enterprise resource planning: market, functional and conceptual view based on SAP S/4HANA. Berlin: Springer Publishing; 2022.

35. Saaty TL

How to make a decision: the analytic hierarchy process. Eur J Oper Res. 1990;48:9–26.

36. Rubinstein RY, Kroese DP

Simulation and the Monte Carlo method. 3rd ed. Hoboken: John Wiley & Sons; 2016.

37. Keeney RL, Raiffa H

Decisions with multiple objectives: preferences and value trade-offs. Cambridge: Cambridge University Press; 1993.

38. Mikalef P, Gupta M

Artificial intelligence capability: conceptualization, measurement calibration, and empirical study on its impact on organizational creativity and firm performance. Inform Manag. 2021;58:103434.

39. Feuerriegel S, Hartmann J, Janiesch J, Zschech P 40. Aldhafeeri L, Aljumah F, Thabyan F, Alabbad M, AlShahrani S, Alanazi F, et al

Generative AI chatbots across domains: a systematic review. Appl Sci. 2025;15:11220.

41. Ren Y, Wang H, Liu Y, et al

AI agents and agentic AI—navigating a plethora of concepts for future manufacturing. Adv Eng Inform. 2025;83:126–33.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.