Supporting Managerial Decision-Making for Federated Machine Learning - Design of a Technology Selection Tool

Aug 2, 2026

Upload by Ivy

Insufficient and high-quality training data remains a persistent bottleneck for production-ready machine learning. Data silos limit collaboration, yet federated machine learning promises model-to-data training on decentralized, potentially siloed datasets. Despite this potential, many projects never move beyond prototypes due to unrealistic expectations and weak alignment with the use case. The work proposes a decision-support tool to evaluate suitability and complexity for FedML initiatives, enabling more grounded technology selection, reducing hype-driven adoption, and improving project success.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Motivation: training data scarcity and data silos

Federated Machine Learning adoption gap

Research aim and contributions

Research Questions

RQ1: supporting technology selection for FedML

RQ2: designing a decision-support tool

Proceedings of the 57th Hawaii International Conference on System Sciences | 2024 Supporting Managerial Decision-Making for Federated Machine Learning: Design of a Technology Selection Tool Milena Zahn Technical University of Munich and SAP SE [milena.zahn@sap.com](milena.zahn@sap.com) Tobias M¨uller Technical University of Munich and SAP SE [tobias.mueller15@sap.com](tobias.mueller15@sap.com) Florian Matthes Technical University of Munich [matthes@tum.de](matthes@tum.de) Abstract The insufficient amount of training data is a persisting bottleneck of Machine Learning systems. A large portion of the world’s data is scattered and locked in data silos. Breaking up these data silos could alleviate this problem. Federated Machine Learning is a novel model-to-data approach that enables the training of Machine Learning models, on decentralized, potentially siloed data. Despite its promising potential, most Federated Machine Learning projects never leave the prototype stage. This can be attributed to exaggerated expectations and an inappropriate fit between the technology and the use case. Current literature does not offer guidance for assessing the fit between Federated Machine Learning and their use case. Against this backdrop, we design a decision-support tool to aid decision-makers in the suitability and complexity assessment of FedML projects. Thereby, we aim to facilitate the technology selection process, avoid exaggerated expectations and consequently facilitate the success of Federated Machine Learning projects. Keywords: Federated Machine Learning, Technology Adoption, Design Science Research 1. Introduction The lack of sufficient and high-quality training data is a persisting challenge in engineering production-ready Machine Learning (ML) systems. Especially small and medium-sized enterprises (SMEs) suffer from an insufficient amount of training data for the development of data-demanding ML systems (Bauer et al., 2020) . Despite the growing wealth of digitized data, a considerable amount is still unavailable, scattered, and locked up in data silos. Especially SMEs could facilitate the lack of training data by breaking down these data silos through sharing data and collaborating. However, the companies’ willingness to share data is low due to privacy concerns and potential loss of intellectual property (Schomakers et al., 2020) . FedML is a novel ML paradigm that allows the joint training of an ML model on distributed data without the direct need for data sharing. Through its model-to-data approach, FedML allows organizations to collaborate on ML projects without having to disclose their data to other organizations. As pointed out by the World Economic Forum (2020), this capability to collaborate and share data will become increasingly important as”true masters of digitalization” not only leverage their own data but also improve existing applications or create new ones with data collaboration. Despite its promise to foster collaboration and enable the usage of currently untapped data, the adoption of FedML in production-ready systems remains limited (Lo et al., 2021) . The lack of operationalized FedML systems can be attributed to a manifold of factors, such as the technical complexity of engineering non-deterministic ML systems (Giray, 2021) or the difficulties of managing collaborative projects (M¨uller et al., 2023) . Additionally, the complexity and number of emerging technologies make it increasingly complicated for practitioners and decision-makers to get a solid understanding of the technology that is needed to determine the appropriate fit of a technology for their use case (Shen et al., 2010) . This contributes to the observation of Maghazeiet al. (2022), that decision-makers tend to use emerging technologies only based on the hype surrounding the technology without examining their actual business benefits. However, a well-grounded fit between the URI: [https://hdl.handle.net/1012

FAQs

Why is federated machine learning adoption limited in production systems?

Most FedML efforts stall in prototypes because expectations are exaggerated and the technology often does not fit the specific use case, alongside engineering complexity and collaboration management challenges.

What research gap does this work address?

Current literature lacks guidance and decision-support tools that help practitioners assess both the suitability and complexity of FedML for their use cases.

What is the purpose of the proposed decision-support tool?

The tool supports decision-makers in evaluating feasibility and complexity of FedML projects, helping avoid hype-based choices and increasing the likelihood of successful implementation and value creation.

Aug 2, 2026

Upload by Ivy