Is Analytics Decentralization a Boon or a Bane?

With parallel development in the availability of data across different sectors and mainstreaming of data science approaches to decision-making, predictive and prescriptive analytics techniques have become desirable toolkits for administrators and managers to gain valuable insights about the environment to make informed decisions. In this regard, although obtaining integrated analytics solutions encompassing various departments in an organization and various stakeholders in a supply chain has the merits of providing optimal and efficient solutions at the systemic level, approaching the problems in a decentralized manner for each stakeholder within the system does have its own share of merits despite it falling short of attaining systemwide optimality, leading to suboptimal solutions. We present the case for why analytics decentralization is useful, particularly in the presence of multiple stakeholders in the system taking administrative, technological and environmental factors into consideration.

Administrative Considerations

In large enterprises or public sector operations, systems that have evolved over decades have modularized their respective work into several connected components that manifested into massive departments or verticals over time. Any changes in the current workflow requires coordination among several such stakeholders and reevaluation of long-standing procedures and processes. It might be true that maximization of profit or public good is the objective of any enterprise; however, in practice, mammoth institutions have lesser desire to restructure their long-standing processes for minor improvements because of several reasons:

1. Stakeholders at similar levels in the power structure of any organization show lesser desire to adopt better practices by improving coordination levels and sharing operational data. This behavior occurs because of the lack of incentive to coordinate in day-to-day operations unless it is externally thrust on them from a senior executive.
2. End-to-end systemic level improvements through analytics also comes with a high capital cost, and organizations may not have the risk appetite to invest (in hopes of cost savings because senior management understands the turf war that arises in toppling the apple cart of settled business processes and the energy, effort and expenditure required for such change).

Therefore, there exists a need to understand the practical realities of business administration and public administration in adopting analytics in a decentralized fashion for every department so that there can be a bottom-up approach to develop analytics-based decision-making at the higher levels. If a decentralized analytics system considering stakeholder autonomy and localized requirements is constructed, every stakeholder shall be empowered to implement optimal and efficient solutions taking aid from the data science engines.

Technological Considerations

Technological constraints such as the inability to implement computationally efficient hardware across all stakeholders, leading to centralized cloud computing architectures and the parallel counter emergence of edge computing through a decentralized architecture, have necessitated the need for lightweight analytics toolkits [1]. Therefore, a decentralized approach to decision-making can be useful to the concerned stakeholders given the infrastructure. Stakeholders do not require sophisticated hardware resources because problems at the decentralized level tend to be smaller in size, leading to the efficient usage of existing resources to solve the problems at hand. Considering the case of a certain class of prescriptive analytics optimization problems that are computationally expensive, or training periods in the predictive analytics setting, the size of the problem or data set can quickly lead to the demand for high-end computational resources, particularly in the presence of an integrated solution approach that tends to deal with large datasets or problems, unlike the decentralized approach.

Environmental Considerations

Another important point of consideration in this age of sustainable development is that solutions with environmental focus are given increased importance over other alternate solutions. Therefore, it becomes imperative to acknowledge how environmentally friendly the analytics algorithms are, let alone the solutions. It is pertinent to note that the cost of training certain predictive algorithms can lead to extraordinary emissions in proportion to model size on top of extraordinary computational resource requirements [2]. Given this, it is only prudent to consider decentralized analytics approaches that deal with small-scale problems in comparison with integrated, large-scale problems, especially when the implementation of these solutions is not straightforward.

Overall, rather than putting forth an integrated solution, which will be a nonstarter, a decentralized solution approach empowering the stakeholder can enable the organization to leverage the power of analytics and data science in practice. This also leads to positive effects in terms of technological constraints and sustainability considerations.

References

1. Hoffpauir, K., Simmons, J., Schmidt, N., Pittala, R., Briggs, I., Makani, S., & Jararweh, Y., 2023, “A Survey on Edge Intelligence and Lightweight Machine Learning Support for Future Applications and Services,” ACM Journal of Data and Information Quality, Vol. 15, No. 2, pp. 1-30.
2. Hao, K., 2019, “Training a single AI model can emit as much carbon as five cars in their lifetimes,” MIT Technology Review, June 6, https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/.