A Multi-Layer Business Intelligence Framework for Real-Time KPI Monitoring in Distribution Operations

2.1 Business Intelligence and KPI monitoring

BI has evolved from static reporting tools into an integrated set of technologies and processes that transform heterogeneous raw data into actionable insights to support strategic and operational decision making [16, 17]. Contemporary BI platforms typically encompass data warehousing, online analytical processing, dashboard visualization, and analytical services that enable organizations to monitor performance through quantifiable indicators, commonly formalized as KPIs [13, 18]. KPIs function as a mechanism for translating organizational objectives into measurable targets, thereby supporting continuous performance monitoring, benchmarking, and alignment across organizational units [18, 19].

In distribution and logistics-intensive sectors, BI-driven KPI monitoring plays a particularly critical role due to high demand variability, multi-echelon inventory structures, and geographically dispersed operations [20]. Prior studies indicate that BI-supported KPI systems can reduce reporting latency, improve visibility into operational bottlenecks, and enhance responsiveness to disruptions across supply and distribution networks [21]. Nevertheless, many organizations continue to rely on spreadsheet-based reporting practices, which remain susceptible to data inconsistencies, manual entry errors, version control conflicts, and limited support for near real-time analytics [22-24]. These limitations constrain managers’ ability to identify anomalies early and to coordinate corrective actions across multiple divisions.

Although the literature broadly acknowledges the value of BI in enabling KPI-driven management, most existing studies conceptualize KPI dashboards primarily as end products rather than as components of an integrated analytical architecture. KPI computation logic is often embedded directly within extract, transform, and load routines or dashboard configurations, thereby reducing transparency, complicating maintenance, and limiting cross-divisional reuse. As a result, BI implementations frequently emphasize visualization outputs while underspecifying the architectural mechanisms that govern KPI definition, calculation consistency, and rule enforcement across organizational contexts [21, 25, 26].

This architectural limitation is particularly problematic in multi-divisional distribution environments, where performance indicators differ across sales, inventory, logistics, and human resource functions but must still adhere to a shared performance logic. Without explicit architectural support for standardized KPI computation and governance, inconsistencies in metric interpretation can emerge, undermining organizational alignment and decision reliability. Consequently, there remains a need to reconceptualize BI systems not merely as reporting tools, but as structured architectures that explicitly integrate data sources, transformation logic, KPI computation rules, and visualization layers into a coherent decision support ecosystem.

2.2 Business Intelligence frameworks and multi-layer architectures

To manage increasing data volume and analytical complexity, both researchers and practitioners have proposed various multi-layer BI architectures. These architectures are commonly structured into distinct layers that separate data acquisition, integration, semantic processing, and presentation functions [27]. Traditional data warehouse architectures, for example, distinguish between the data source layer where operational data are generated, the data integration or extract, transform, load layer that handles data preprocessing, the semantic or warehouse layer that stores cleansed and integrated data, and the presentation layer that delivers analytical results to end users through reporting and visualization tools [28, 29]. Several studies have extended these foundational architectures by introducing additional layers or components, such as metadata management, business rules, or analytical services, in order to support more advanced decision-making capabilities [30]. In performance management contexts, BI frameworks are often augmented with strategy maps, balanced scorecards, or KPI catalogs that aim to align operational indicators with higher-level organizational objectives [31, 32]. While these extensions enhance strategic alignment, many proposed frameworks remain conceptual and do not explicitly formalize the KPI logic layer where business rules, threshold definitions, and anomaly detection mechanisms are executed. In many existing BI implementations, KPI computation rules are embedded within ad hoc SQL scripts, extract, transform, load procedures, or dashboard configuration logic. This tightly coupled approach reduces transparency and limits reusability, as changes to KPI definitions often require modifications across multiple system components. Furthermore, such implementations hinder auditability and make it difficult to enforce consistent interpretation of KPIs across organizational units. As a result, KPI logic is frequently treated as an implementation detail rather than as a core architectural concern. These limitations become more pronounced in multi-divisional distribution firms, where performance indicators vary significantly across sales, inventory management, logistics operations, and human resources functions [33, 34]. In these environments, each division may require different aggregation levels, alert thresholds, and visualization formats while still contributing to a unified organizational performance narrative. However, most existing BI architectures either adopt sector-neutral designs or focus on domain-specific applications in areas such as healthcare, manufacturing, or finance [35, 36]. Consequently, there is limited guidance on how to architect BI systems that explicitly accommodate the coordination and governance challenges inherent in distribution-oriented organizations. This gap highlights the need for a multi-layer BI framework that explicitly incorporates a dedicated KPI logic and rules engine, semantic aggregation mechanisms tailored to multiple organizational roles, and near real-time visualization capabilities. Such an architecture is necessary to bridge the gap between high-level BI concepts and their practical implementation in operationally intensive distribution contexts.

2.3 Real-time KPI monitoring and decision support systems

Real-time or near-real-time BI has gained increasing attention as organizations seek to shorten the latency between data generation, analysis, and decision-making [37]. In contrast to traditional batch reporting, real-time BI enables continuous monitoring of KPIs, immediate identification of deviations from expected performance, and timely intervention in dynamic environments. In distribution networks, real-time visibility into sales, stock levels, delivery performance, and workforce activity is critical for mitigating stockouts, capacity bottlenecks, and service-level violations [25, 38-40].

Decision Support Systems (DSS) historically focused on model-driven or data-driven approaches to aid managerial decision-making [30, 41]. With the advent of BI and analytics platforms, contemporary DSS increasingly leverage dashboards and KPI-driven interfaces as primary decision-support channels [11]. Recent studies highlight that situational awareness and early anomaly detection are key benefits of real-time KPI dashboards, particularly in domains such as power systems, supply chain operations, and emergency management [42, 43]. However, many reported systems rely on static thresholds, lack explicit integration with multi-layer BI architectures, or provide limited support for role-based decision support, where different user groups require different views of the same underlying data [44, 45].

Furthermore, although some works have introduced rule-based or machine-learning-based mechanisms for anomaly detection within KPI streams, these methods are often applied in isolated analytic pipelines rather than embedded within a holistic BI framework that spans from data capture to decision support [13, 46]. As a result, there remains a need for integrated approaches that combine real-time KPI monitoring, rule-based anomaly detection, and role-specific decision support within a unified architectural model, particularly in complex distribution operations.

2.4 Usability and human–computer interaction in Business Intelligence systems

Although the technical capabilities of BI systems have advanced significantly, their practical effectiveness remains highly dependent on usability and user experience. BI systems typically serve a diverse group of users, including executives, middle managers, analysts, and operational staff, each with different levels of technical expertise and distinct information needs. Poorly designed dashboards can lead to misinterpretation of performance indicators, underutilization of analytical features, and resistance to organizational adoption, even when the underlying data infrastructure is robust [47].

HCI research has long emphasized usability as a critical determinant of system effectiveness. Among the most widely adopted evaluation frameworks are Nielsen’s Ten Usability Heuristics, which provide structured criteria for assessing interface quality during early and iterative design stages [48, 49]. These heuristics include principles such as system status visibility, consistency, error prevention, recognition rather than recall, and support for user control. When applied systematically, heuristic evaluation enables the identification of usability issues that may impede efficient task execution and comprehension.

Several studies have applied heuristic evaluation methods and standardized usability instruments, such as the System Usability Scale, Questionnaire for User Interaction Satisfaction, and User Experience Questionnaire, to assess dashboards and analytical applications [50]. Findings from this body of work consistently indicate that usability problems often stem from unclear feedback, excessive information density, inconsistent navigation structures, and insufficient user guidance. However, compared with domains such as web applications and general information systems, the application of usability-driven evaluation in BI research remains limited.

Moreover, many BI-related studies focus on technical performance (e.g., query latency, scalability, data accuracy) while treating usability evaluation as a secondary or optional activity. Only a limited number of works explicitly integrate usability evaluation into the BI development lifecycle, for example, by using heuristic evaluation to refine dashboard design iteratively or by combining scenario-based tasks with role-specific usability assessment [47, 51]. This gap is particularly pronounced in industrial case studies, where BI deployments often evolve through informal feedback and localized adjustments rather than through systematic, theory-informed usability evaluation.

Although the studies summarized in Table 1 provide foundational models for BI implementation, a critical analysis reveals three persistent architectural limitations that constrain their applicability in distribution operations.

Table 1. Summary of prior Business Intelligence (BI) framework studies and identified research gaps

First, regarding KPI Logic, existing frameworks such as [52, 53] typically embed KPI computation rules directly within SQL scripts or visualization widgets. This tightly coupled approach limits auditability and reusability, as modifying KPI definitions often requires changes across multiple system components. In contrast, the proposed framework explicitly decouples KPI computation into a dedicated, reusable KPI logic layer, enabling consistent rule management across divisions.

Second, regarding Real-Time Capability, prior studies, including, predominantly rely on batch-oriented ETL processes that are suitable for periodic reporting but are insufficient for the continuous monitoring demands of logistics and distribution operations. These architectures often reduce reporting delays without explicitly addressing end-to-end latency, particularly in environments still dependent on spreadsheet-based data aggregation.

Third, regarding Domain Adaptability, many reviewed architectures focus on generic SME contexts or on sector-specific applications, such as facilities management. As a result, they do not explicitly accommodate the multi-divisional coordination challenges inherent in distribution enterprises, where Sales, Logistics, Inventory, and HR KPIs must be synchronized under a unified performance logic. This study addresses this gap by introducing a role-based semantic aggregation layer tailored to the operational structure of distribution firms.

2.5 Synthesis of research gaps

Based on the reviewed literature, several interrelated gaps can be identified:

1. Architectural Gap

Existing BI frameworks typically describe generic multi-layer architectures but rarely model a distinct KPI logic layer that encodes business rules, alert thresholds, and anomaly-detection mechanisms in a reusable, maintainable way. The limited attention to KPI-centric architectures is particularly evident in multi-divisional distribution settings.

2. Domain-Specific Gap (Distribution Sector)

Few studies provide an integrated BI framework specifically designed for building materials distributors or similar multi-division distribution firms, in which KPI structures differ across sales, inventory, logistics, and HR divisions yet must still be governed by a shared performance logic.

3. Real-Time Decision Support Gap

Although real-time KPI dashboards are increasingly discussed, many implementations focus on visualization and do not situate KPI monitoring within a broader decision-support architecture that explicitly connects data sources, ETL processes, rule-based anomaly detection, and role-based visualization.

4. Usability Evaluation Gap

While usability is widely recognized as a critical factor in BI adoption, systematic usability-driven evaluation, particularly through established Human Computer Interaction methods such as Nielsen’s heuristics, combined with scenario-based task evaluation, remains underrepresented in the BI literature. Few studies position usability evaluation as a core component of BI framework validation.

This study responds to these gaps by:

1. Proposing a multi-layer BI framework that explicitly incorporates a KPI logic and rules engine, semantic aggregation, and role-specific visualization into a unified architectural model for real-time KPI monitoring.
2. Applying and reporting a usability-driven evaluation that uses heuristic assessment and scenario-based testing with multiple organizational roles in an industrial distribution case study.

With this foundation, the next section presents the proposed multi-layer BI framework. It describes how it operationalizes these concepts into an implementable architectural model for real-time KPI monitoring and decision support.

This study adopts a mixed-methods design–evaluation approach that integrates systems engineering principles for framework development with HCI-oriented evaluation techniques to assess dashboard usability. The methodological sequence consists of four major phases: research design, case study context and data collection, system development through an iterative BI lifecycle, and usability-driven evaluation. This structured approach ensures both technical rigor and practical relevance, allowing the proposed BI solution to be examined in a realistic organizational environment.

4.1 Research design

This study employs a single-case embedded design, an approach widely recognized in information systems research for investigating complex organizational phenomena where context and multi-role interactions are critical. The research design integrates two complementary methodological strands: a Design Science Research (DSR) approach for system development and a Human-Computer Interaction (HCI) oriented evaluation. As illustrated in Figure 5, the research workflow consists of four primary phases: diagnosis and data collection, iterative system development, and a usability-driven evaluation. This methodological integration ensures that the resulting BI framework is not only architecturally robust but also cognitively aligned with the practical needs of decision-makers in a resource-constrained environment.

Figure 5. Research workflow / methodological framework

4.2 Case study context

The empirical setting for this study is a multi-divisional building materials distribution firm operating across sales, logistics, warehouse management, human resources (HR), and finance. The organization generates substantial volumes of data from high-frequency sales transactions, multi-location warehouse operations, and daily delivery logistics. This context is ideal for evaluating BI integration because the firm historically relied on decentralized spreadsheet-based reporting and manual data aggregation. These legacy practices resulted in significant data latency, inconsistent KPI interpretations, and delayed decision-making, common challenges in Small and Medium Enterprises (SMEs) in developing economies.

4.3 Data collection procedures

Data collection used a triangulation approach with three main sources to ensure accuracy. First, operational data were pulled from raw datasets, including daily sales reports, inventory movement logs, delivery schedules, and attendance records. These datasets provided the basis for building the ETL pipeline and calculating KPI formulas. Second, semi-structured interviews were held with key stakeholders, such as sales supervisors, warehouse managers, logistics coordinators, and division heads. These interviews focused on defining core KPIs, identifying operational challenges, and setting dashboard requirements. Third, ongoing user feedback was collected during the iterative prototyping stage to improve dashboard layouts, color coding, and drill-down interaction paths.

4.4 System development process

The BI system was developed through an iterative lifecycle that integrated ETL modeling, KPI rule formulation, and role-based visualization design. The process began with Requirements Analysis, in which functional requirements were derived from process mapping and stakeholder interviews. Non-functional requirements, such as real-time accuracy and low cognitive load, were also established.

Subsequently, the ETL Pipeline Construction phase focused on harmonizing heterogeneous datasets. This process involved cleaning key fields, standardizing formats, and applying business logic to generate consolidated datasets for the semantic layer. The specific transformation rules applied to ensure data integrity are detailed in Table 2 (Section 3).

Following data preparation, the KPI Rule Engineering phase focused on encoding the performance logic. Crucially, the determination of specific alert thresholds and performance benchmarks, as summarized in Table 3, was not arbitrary. These parameters were established through a triangulation method combining: (1) the identification of logistics and distribution performance indicators with reference to established frameworks like the SCOR model, (2) the analysis of 12 months of historical operational data to determine typical performance ranges, and (3) expert consultations with managerial stakeholders to refine KPI definitions and threshold levels.

Based on this foundation, each performance indicator was computed using the standardized mathematical definitions and formulas presented in Table 4. To automate the identification of performance outliers, the strictly defined threshold rules (Green/Yellow/Red) were mapped to each KPI as detailed in the same table. These rules were subsequently encoded into transformation scripts and rigorously validated through cross-divisional checks to ensure calculation accuracy before deployment to the visualization layer.

Figure 6. Technical technology stack architecture

The final development stage was Dashboard Development, which focused on creating specialized views for three distinct user roles: Directors (strategic view), Supervisors (tactical view), and Operational Staff (execution view). The interface design adhered to principles of visual minimalism and consistency and was implemented in Google Looker Studio to ensure accessibility and cost-effectiveness. The user interface architecture and component interactions are depicted in Figure 4.

4.4.1 Technology stack and implementation details

To ensure replicability and minimize capital expenditure (CAPEX) for the target SME, the system utilizes a cloud-native, low-code technology stack. The specific technological components selected for each architectural layer are detailed below and illustrated in Figure 6.

1. Data Storage Layer (Database)

Google Sheets was selected as the primary data warehouse due to its native integration with operational inputs (CSV/Excel) and zero licensing cost. For scalability and historical archiving, the system connects to Google BigQuery.

2. ETL & Logic Layer

Google Apps Script functions as the middleware. It executes the ETL routines (cleaning and validation) and hosts the KPI Rules Engine scripts described in Section 3.3. This serverless environment enables automated, trigger-based execution (e.g., time-driven triggers at 15-minute intervals) without dedicated infrastructure.

3. Visualization Layer

Google Looker Studio serves as the front-end interface, chosen for its drag-and-drop flexibility and ability to render real-time data from the Google ecosystem without connector latency.

This selection is justified by the resource-constrained nature of the case study environment, in which high-maintenance enterprise ERP systems are often financially infeasible.

4.5 Usability-driven evaluation

To validate the system's effectiveness, a two-phase usability assessment was conducted. The first phase was a Heuristic Evaluation performed by three trained evaluators using Nielsen’s Ten Usability Heuristics. Each identified issue was assigned a severity score (0–3) to prioritize interface refinements.

The second phase involved a scenario-based usability evaluation conducted with 15 participants selected through purposive sampling to represent the system's intended user strata. The participant pool consisted of two Directors evaluating the strategic dashboard, five Divisional Managers evaluating the tactical dashboard, and eight Operational Staff evaluating the operational dashboard.

Participants were asked to complete realistic, role-specific usage scenarios that reflected routine decision-making and monitoring tasks. The evaluation focused on task completion, clarity of information presentation, ease of navigation, and qualitative user feedback. Consistent with the formative nature of the study, the objective was to identify usability issues and assess practical suitability rather than to perform inferential statistical analysis.

4.6 Data analysis techniques

Data analysis employed both quantitative and qualitative methods. Quantitative analysis focused on descriptive statistics of task performance metrics and on aggregating heuristic severity scores. Qualitative analysis involved thematic coding of evaluator comments and user feedback. The triangulation of these results provides robust evidence regarding the system’s usability and its impact on decision-making efficiency.

4.7 Methodological rigor and validity

To ensure research rigor, this study addresses construct validity by using standardized KPI formulas and established heuristic guidelines. Internal validity is supported by the consistency of findings across different user roles, while reliability is maintained through replicable ETL processes and rule definitions. Although the single-case design limits broad generalization, the detailed description of the architectural implementation offers a transferrable blueprint for similar distribution enterprises.

This section interprets the study’s findings in relation to existing literature, managerial practice, and the theoretical foundations of BI. By integrating architectural design, KPI-centric logic, and usability-driven assessment, the proposed framework contributes to three key domains: BI architectural theory, real-time KPI analytics, and user-centered decision support within industrial environments.

6.1 Theoretical implications

6.1.1 Advancing BI architectural theory through a KPI-centric model

The findings demonstrate that a BI architecture explicitly incorporating a KPI Logic & Rules Engine Layer offers distinct theoretical value beyond traditional BI models. While prior studies often treat KPI calculations as technical details embedded within ETL routines or dashboard scripts, this study elevates KPI logic to a dedicated architectural layer. This separation formalizes KPI computation as a central analytical component, clarifies the translation of strategic objectives into operational metrics, and reduces the semantic ambiguity that frequently arises in multi-divisional contexts. This refinement aligns with systems engineering principles and enriches existing BI frameworks, which typically emphasize data warehousing or visualization but under-specify the transformation of raw data into decision-relevant intelligence. Consequently, this study proposes that KPI logic should be conceptualized as a first-class, reusable, and auditable component of modern BI architectures.

Thus, the study contributes to BI theory by proposing that KPI logic should be conceptualized as a first-class, reusable, and auditable layer within BI architectures. This insight has not been adequately discussed in prior literature.

6.1.2 Integrating usability evaluation into BI research

Traditional BI research has predominantly focused on technical performance, scalability, and data accuracy, often treating usability as a secondary consideration. This study advances the discourse by systematically incorporating heuristic evaluation and scenario-based testing into the BI development lifecycle. The results confirm that even technically robust dashboards may fail to deliver decision support if usability barriers persist. By demonstrating that HCI techniques can be effectively applied to validate BI systems, this research strengthens the methodological rigor of BI evaluation. It reinforces the theoretical proposition that usability is a determining factor of analytical effectiveness.

6.2 Practical implications

6.2.1 Enhanced organizational visibility and cross-division alignment

The real-time KPI intelligence delivered by the framework proved effective in addressing the case company's long-standing reporting challenges. By consolidating and standardizing data across sales, inventory, logistics, and HR divisions, the framework enabled a synchronized understanding of performance and faster anomaly detection. Managers reported significantly improved clarity in interpreting indicators, whereas directors benefited from uniform cross-divisional comparisons that had previously been hindered by inconsistent legacy reporting. These outcomes highlight the practical value of structured BI architectures for organizations operating with fragmented data ecosystems.

6.2.2 Role-based decision support for multi-level management

The successful deployment of role-specific dashboards illustrates the importance of tailoring information granularity to user needs. The system effectively differentiated views—providing strategic overviews for directors, tactical insights for supervisors, and task-level alerts for operational staff. This differentiation reduced cognitive overload and improved decision relevance, addressing a common failure point in BI projects: "one-size-fits-all" dashboards that lead to user disengagement. The findings suggest that role-based customization is essential for enhancing adoption and managerial efficiency in multi-functional enterprises.

6.2.3 The viability of rule-based approaches for developing economies

A critical practical implication of this study is the validation of deterministic rule-based analytics as a viable alternative to high-cost AI solutions for SMEs in developing economies. While machine learning offers predictive capabilities, it requires substantial historical data, computational power, and specialized expertise—resources that are often scarce in these regions. The proposed framework demonstrates that a well-structured rule-based engine can deliver immediate operational stability, auditability, and sufficient anomaly-detection capabilities at a fraction of the cost. This finding offers a strategic pathway for SMEs to achieve digital maturity without prematurely overinvesting in complex technologies.

6.3 Comparison with prior studies

Compared with existing BI literature, which often emphasizes visualization techniques or back-end data integration in isolation, this study introduces a holistic, multi-layer architecture that integrates ETL, KPI logic, semantic models, and decision support into a unified system. Furthermore, unlike generic BI studies, this research provides a domain-specific adaptation for the distribution sector that addresses the unique complexities of multi-divisional logistics. Most notably, this study bridges the gap between systems engineering and HCI by providing empirical evidence on how usability evaluation can directly inform BI architectural refinement—an approach rarely detailed in prior industrial case studies.

6.4 Managerial insights

Three key lessons emerge for practitioners. First, standardization reduces organizational friction; harmonized data pipelines eliminate cross-division disputes over data validity and accelerate report generation. Second, real-time responsiveness is critical; the ability to detect sales drops or stock anomalies within minutes rather than days significantly strengthens operational resilience. Third, usability drives sustained adoption; the preference for the redesigned dashboards confirms that user experience quality is as critical as data accuracy. Managers are more likely to utilize analytics tools that minimize navigation effort and provide clear, actionable feedback.

6.5 Limitations

Despite its contributions, this study has limitations. First, the evaluation focused on a single organizational case, which may limit the generalizability of the findings to other industries. Second, the anomaly detection logic is rule-based rather than predictive; while effective for current operational needs, it does not forecast future trends. Third, the sample size for the usability evaluation was moderate. These limitations open opportunities for future research, such as incorporating machine learning for predictive analytics and expanding the review to a broader range of distribution firms.