Strategic Management of AI Investments: A Dynamic Capabilities and Business Model Innovation Perspective
Artificial Intelligence (AI) promises substantial returns in organisations, motivating firms to cons ...
Artificial Intelligence (AI) promises substantial returns in organisations, motivating firms to cons ... ider AI investments to improve their businesses. The investments offer new opportunities for transforming business processes, which can increase the company’s value. Despite the promise of AI to drive competitive advantage, organisations struggle to realise its value, often because of fragmented initiatives and strategic misalignment rather than technological failures. Current research often overlooks the strategic initiatives required to convert dynamic technological potential into sustained performance. This study addresses this gap by integrating Dynamic Capabilities (DC) and Business Model Innovation (BMI) theories to deconstruct the realisation of AI value. Employing a mixed-methods design, we triangulated a systematic literature review with exploratory practitioner interviews to contrast theoretical ideals with operational realities, aiming to benefit from AI investments. The findings reveal that the primary barriers to AI value realisation, strategic misalignment, organisational inertia, and technical/data limitations, are interconnected and mutually reinforcing, creating a negative feedback loop that prevents scaled impact. This systemic failure is driven by critical operational frictions, including data governance silos, leadership literacy gaps, and the inability to quantify AI value, which collectively disrupt the link between strategic intent and execution. Also, findings showed the potential of strategically interlinking Dynamic Capabilities (DC) and Business Model Innovation (BMI) for business value, where DC senses and mobilizes AI opportunities and BMI turns over adaptive capacities into value creation and capture, thus contributing to AI value realization in organisations.
This paper presents a probabilistic forecasting approach tailored for low voltage (LV) substations, ... offering short-term predictions for three crucial variables: voltage, reactive power, and active power. These parameters play a vital role in the resilience of distribution systems, especially in the presence of Distributed Energy Resources (DERs).
Power demand forecasting is becoming a crucial tool for the planning and operation of Low Voltage (L ... V) distribution systems. Most importantly, the high penetration of Photovoltaics (PV) power generation as part of Distributed Energy Resource (DER)s has transformed the power demand forecasting problem at the distribution level into net-load forecasting. This paper introduces a novel and scalable approach to probabilistic forecasting at LV substations with PV generation. It presents a multi-variates probabilistic forecasting approach, leveraging Quantile Regression (QR). The proposed architecture uses a computationally efficient feed-forward neural net to capture the complex interaction between the historical load demands and covariate variables such as solar irradiance. It is empirically demonstrated that the proposed method can efficiently produce well-calibrated forecasts, both auto-regressively or in a single forward pass. Furthermore, a benchmark against four state-of-the-art forecasting approaches show that the proposed approach offers a desirable trade-off between forecasting accuracies, calibrated uncertainty, and computation complexity.
This paper presents a probabilistic forecasting approach tailored for low voltage (LV) substations, ... offering short-term predictions for three crucial variables: voltage, reactive power, and active power. These parameters play a vital role in the resilience of distribution systems, especially in the presence of Distributed Energy Resources (DERs).
Artificial intelligence adoption has accelerated across industries, yet many organisations struggle ... to translate AI investments into sustained business value. This paper identifies the strategic and organisational factors that differentiate high-value AI implementations from those that underdeliver. Drawing on practitioner frameworks and case analysis, we propose a multi-layer model for AI value creation encompassing data readiness, model governance, human-AI collaboration design, and executive alignment. We examine common failure modes — including misaligned incentives, poor problem scoping, and inadequate change management — and provide practical prescriptions for each. The resulting framework is designed to guide AI practitioners, product owners, and senior leaders in structuring AI programmes that deliver measurable organisational impact.
Industrial non-intrusive load monitoring (NILM) presents unique challenges compared to residential s ... ettings, including complex three-phase power systems, diverse load types, and high measurement noise. This paper proposes applying the symmetrical component transform (SCT) — a classical power systems technique — as a feature extraction step for industrial appliance classification in NILM. The SCT decomposes three-phase current and voltage signals into positive, negative, and zero sequence components, which encode distinct physical properties of each load type. We integrate SCT-derived features with a convolutional neural network classifier and evaluate the approach on the LILACD industrial dataset. Experimental results demonstrate that SCT features significantly improve classification accuracy over raw signal approaches, particularly for distinguishing motor-driven loads and non-linear power electronics. Our work bridges classical power engineering knowledge with modern deep learning for more effective industrial NILM.
Nonintrusive load monitoring (NILM) techniques are increasingly becoming a key instrument for identi ... fying the power consumption of individual appliances based on a single metering point. Particularly, deep learning (DL) models are gaining interest in this regard. However, the challenges brought by the NILM datasets and the nonavailability of common experimental guidelines tend to compromise comparison, research transparency, and replicability. The limited adoption of efficient research instruments and lack of best practices guidelines contribute in huge part to this problem, where no features, encouraging standardized formats for benchmarking, and results sharing are offered. To address these issues, we first present a brief overview of recent best practices for DL and highlight how deep NILM research can benefit from these practices. Furthermore, we suggest a novel open-source toolkit leveraging these practices, i.e., Deep-NILMTK. The proposed toolkit offers a common testing bed for NILM algorithms independently of the underlying deep learning framework with a modular NILM pipeline that can easily be customized. Furthermore, Deep-NILMTK introduces the concept of experiment templating to offer predesigned experiments allowing to enhancing research efficiency. Leveraging this concept and DL best practices, we present a case study of creating an online NILM benchmark repository1 considering eight of the most popular deep NILM algorithms. All sources relative to the tool are made publicly available on Github2 along with the corresponding documentation.
Appliance recognition in non-intrusive load monitoring (NILM) is typically framed as either a single ... -label or multi-label classification problem. Single-label approaches predict one appliance per activation event, while multi-label approaches predict the set of simultaneously active appliances. Both paradigms have complementary strengths: single-label methods excel at isolating dominant loads with clear event signatures, whereas multi-label methods handle concurrent appliance operation. This paper proposes a combined approach that leverages both paradigms within a unified recognition pipeline. We train a single-label classifier for isolated events and a multi-label classifier for overlapping load periods, then fuse their outputs via a learned gating mechanism. Evaluation on the UK-DALE dataset shows that the combined approach outperforms either method in isolation, achieving higher F1 scores across all evaluated appliances.
The increased penetration of Renewable Energy Sources (RES) as part of a decentralized and distribut ... ed power system makes net-load forecasting a critical component in the planning and operation of power systems. However, compared to the transmission level, producing accurate short-term net-load forecasts at the distribution level is complex due to the small number of consumers. Moreover, owing to the stochastic nature of RES, it is necessary to quantify the uncertainty of the forecasted net-load at any given time, which is critical for the real-world decision process. This work presents parameterized deep quantile regression for short-term probabilistic net-load forecasting at the distribution level. To be precise, we use a Deep Neural Network (DNN) to learn both the quantile fractions and quantile values of the quantile function. Furthermore, we propose a scoring metric that reflects the trade-off between predictive uncertainty performance and forecast accuracy. We evaluate the proposed techniques on historical real-world data from a low-voltage distribution substation and further assess its robustness when applied in real-time. The experiment’s outcomes show that the resulting forecasts from our approach are well-calibrated and provide a desirable trade-off between forecasting accuracies and predictive uncertainty performance that are very robust even when applied in real-time.
To this day, hyperparameter tuning remains a cumbersome task in Non-Intrusive Load Monitoring (NILM) ... research, as researchers and practitioners are forced to invest a considerable amount of time in this task. This paper proposes adaptive weighted recurrence graph blocks (AWRG) for appliance feature representation in event-based NILM. An AWRG block can be combined with traditional deep neural network architectures such as Convolutional Neural Networks for appliance recognition. Our approach transforms one cycle per activation current into an weighted recurrence graph and treats the associated hyper-parameters as learn-able parameters. We evaluate our technique on two energy datasets, the industrial dataset LILACD and the residential PLAID dataset. The outcome of our experiments shows that transforming current waveforms into weighted recurrence graphs provides a better feature representation and thus, improved classification results. It is concluded that our approach can guarantee uniqueness of appliance features, leading to enhanced generalisation abilities when compared to the widely researched V-I image features. Furthermore, we show that the initialisation parameters of the AWRG's have a significant impact on the performance and training convergence.
The key advantage of smart meters over traditional metering devices is their ability to transfer con ... sumption information to remote data processing systems. Besides enabling the automated collection of a customer’s electricity consumption for billing purposes, the data collected by these devices makes the realization of many novel use cases possible. However, the large majority of such services are tailored to improve the power grid’s operation as a whole. For example, forecasts of household energy consumption or photovoltaic production allow for improved power plant generation scheduling. Similarly, the detection of anomalous consumption patterns can indicate electricity theft and serve as a trigger for corresponding investigations. Even though customers can directly influence their electrical energy consumption, the range of use cases to the users’ benefit remains much smaller than those that benefit the grid in general. In this work, we thus review the range of services tailored to the needs of end-customers. By briefly discussing their technological foundations and their potential impact on future developments, we highlight the great potentials of utilizing smart meter data from a user-centric perspective. Several open research challenges in this domain, arising from the shortcomings of state-of-the-art data communication and processing methods, are furthermore given. We expect their investigation to lead to significant advancements in data processing services and ultimately raise the customer experience of operating smart meters.
Appliance recognition is one of the vital sub-tasks of NILM in which a machine learning classier is ... used to detect and recognize active appliances from power measurements. The performance of the appliance classifier highly depends on the signal features used to characterize the loads. Recently, different appliance features derived from the voltage–current (V–I) waveforms have been extensively used to describe appliances. However, the performance of V–I-based approaches is still unsatisfactory as it is still not distinctive enough to recognize devices that fall into the same category. Instead, we propose an appliance recognition method utilizing the recurrence graph (RG) technique and convolutional neural networks (CNNs). We introduce the weighted recurrent graph (WRG) generation that, given one-cycle current and voltage, produces an image-like representation with more values than the binary output created by RG. Experimental results on three different sub-metered datasets show that the proposed WRG-based image representation provides superior feature representation and, therefore, improves classification performance compared to V–I-based features.
The advance in energy-sensing and smart-meter technologies have motivated the use of a Non-Intrusive ... Load Monitoring (NILM), a data-driven technique that recognizes active end-use appliances by analyzing the data streams coming from these devices. NILM offers an electricity consumption pattern of individual loads at consumer premises, which is crucial in the design of energy efficiency and energy demand management strategies in buildings. Appliance classification, also known as load identification is an essential sub-task for identifying the type and status of an unknown load from appliance features extracted from the aggregate power signal. Most of the existing work for appliance recognition in NILM uses a single-label learning strategy which, assumes only one appliance is active at a time. This assumption ignores the fact that multiple devices can be active simultaneously and requires a perfect event detector to recognize the appliance. In this paper proposes the Convolutional Neural Network (CNN)-based multi-label learning approach, which links multiple loads to an observed aggregate current signal. Our approach applies the Fryze power theory to decompose the current features into active and non-active components and use the Euclidean distance similarity function to transform the decomposed current into an image-like representation which, is used as input to the CNN. Experimental results suggest that the proposed approach is sufficient for recognizing multiple appliances from aggregated measurements.
Non-Intrusive Load Monitoring (NILM) is the task of decomposing aggregate power consumption into ind ... ividual appliance-level signals. Existing deep learning approaches for NILM typically address either appliance state detection or power estimation as separate tasks, limiting their practical utility. This paper proposes UNet-NILM, a multi-task deep neural network architecture inspired by the U-Net encoder-decoder design for semantic segmentation. UNet-NILM simultaneously detects appliance ON/OFF states and estimates per-appliance active power consumption from aggregate mains readings. We evaluate the approach on the UK-DALE and REFIT datasets across five target appliances. Results show that UNet-NILM achieves state-of-the-art performance on both tasks simultaneously, demonstrating that joint optimisation of detection and estimation objectives is mutually beneficial. The proposed architecture offers a practical step toward deployable NILM systems that provide both appliance status and fine-grained energy consumption data.
Energy communities — groups of households and businesses that share local renewable energy resources ... — are emerging as a key mechanism for the energy transition. Realising self-sufficiency within such communities requires intelligent coordination of generation, storage, and consumption. This paper investigates how machine learning can support this coordination by addressing three interrelated problems: short-term solar and demand forecasting, appliance-level load disaggregation via NILM, and community-scale demand response optimisation. We propose an integrated ML pipeline connecting these modules and evaluate it on a simulated energy community with real solar irradiance and consumption data. Results demonstrate that the combined ML approach improves self-sufficiency rates by 18% compared to rule-based scheduling, with forecasting accuracy being the dominant factor in overall community performance.
Non-intrusive load monitoring (NILM) research has produced a rich body of machine learning approache ... s for appliance state detection and energy disaggregation. However, the community lacks standardised metrics for evaluating how well models trained on one dataset transfer to another. This paper addresses this gap by analysing existing evaluation metrics, identifying their limitations for cross-dataset transfer, and proposing a framework of transferability metrics grounded in information-theoretic and statistical measures. We demonstrate the framework on multiple residential energy datasets and show how commonly used metrics such as F1 score can be misleading when assessing model transferability. Our findings provide guidance for researchers on selecting appropriate metrics when reporting NILM model performance in multi-dataset settings.
The rapid urbanization of developing countries coupled with explosion in construction of high rising ... buildings and the high power usage in them calls for conservation and efficient energy program. Such a program require monitoring of end-use appliances energy consumption in real-time. The worldwide recent adoption of smart-meter in smart-grid, has led to the rise of Non-Intrusive Load Monitoring (NILM); which enables estimation of appliance-specific power consumption from building's aggregate power consumption reading. NILM provides households with cost-effective real-time monitoring of end-use appliances to help them understand their consumption pattern and become part and parcel of energy conservation strategy. This paper presents an up to date overview of NILM system and its associated methods and techniques for energy disaggregation problem. This is followed by the review of the state-of-the art NILM algorithms. Furthermore, we review several performance metrics used by NILM researcher to evaluate NILM algorithms and discuss existing benchmarking framework for direct comparison of the state of the art NILM algorithms. Finally, the paper discuss potential NILM use-cases, presents an overview of the public available dataset and highlight challenges and future research directions.
Rural communities in developing regions continue to face significant barriers to cellular connectivi ... ty due to the high capital costs of proprietary telecommunications infrastructure. This paper investigates the feasibility of open source cellular technologies — including OpenBTS, OpenBSC, and Osmocom — as low-cost alternatives for rural area coverage. We evaluate deployment architectures, spectrum requirements, hardware costs, and community network governance models. A comparative analysis of open source and proprietary solutions is conducted with respect to total cost of ownership, maintenance complexity, and quality of service. Our findings suggest that open source cellular platforms can reduce infrastructure costs by up to 70% compared to conventional approaches while delivering acceptable service quality for voice and basic data services in rural deployments.
As the human population growth and industry pressure in most developing countries continue to increa ... se, effective water quality monitoring and evaluation has become critical for water resources management programs. This paper presents the ubiquitous mobile sensing system for water quality data collection and monitoring applications in developing countries. The system was designed based on the analysis of the existing solution. Open source hardware and software was used to develop the prototype of the system. Field testing of the system conducted in Nkokonjero, Uganda and Mwanza, Tanzania verified the functionalities of the system and its practical application in actual environment. Results show that proposed solution is able to collect and present data in a mobile environment.
The need for effective and efficient monitoring, evaluation and control of water quality in Lake Vic ... toria Basin (LVB) has become more demanding in this era of urbanization, population growth and climate change and variability. Traditional methods that rely on collecting water samples, testing and analyses in water laboratories are not only costly but also lack capability for real-time data capture, analyses and fast dissemination of information to relevant stakeholders for making timely and informed decisions. In this paper, a Water Sensor Network (WSN) system prototype developed for water quality monitoring in LVB is presented. The development was preceded by evaluation of prevailing environment including availability of cellular network coverage at the site of operation. The system consists of an Arduino microcontroller, water quality sensors, and a wireless network connection module. It detects water temperature, dissolved oxygen, pH, and electrical conductivity in real-time and disseminates the information in graphical and tabular formats to relevant stakeholders through a web-based portal and mobile phone platforms. The experimental results show that the system has great prospect and can be used to operate in real world environment for optimum control and protection of water resources by providing key actors with relevant and timely information to facilitate quick action taking.