Back to the library

Key question New

AI-processed · Author-approved

What risks arise when public organizations purchase artificial intelligence from external vendors, and how can they manage them?

Between Innovation and Risk Aversion: A Contextual Perception of AI Sourcing Risks in Public Organisations
Zacchaeus Otunla, Tobias Mettler · International Conference on Information Systems (ICIS) · 2026
Illustration for: What risks arise when public organizations purchase artificial intelligence from external vendors, and how can they manage them?

Loads Spotify's player; Spotify may set cookies. This device remembers your choice.

Key findings

  • Revealed three core categories of AI procurement risk: pre-deployment operationalisation risks, risk-aware vendor risks, and adaptive future-proofing risks.
  • Found that operational risks often stem from adopting AI to satisfy external institutional pressures rather than solving specific problems, resulting in misaligned performance expectations.
  • Showed that agencies face significant difficulties vetting inexperienced vendors with limited track records and ensuring external providers comply with strict public data protection standards.
  • Highlighted that organizations manage risks of vendor lock-in and system obsolescence through flexible framework contracts, quality gates, and allocated maintenance budgets.
  • Formulated a model of contextual risk aversion demonstrating that public actors iteratively balance operational alignment, regulatory compliance, and procurement agility under uncertainty.

What it means for you

CIO / IT Executive

Institute mandatory 'quality gates' and allocate dedicated maintenance budgets in upcoming vendor contracts to explicitly mitigate system obsolescence and lock-in risks.

IT Manager

Audit all current AI vendor requirements to ensure strict adherence to public data protection standards and filter out unverified suppliers lacking a proven track record.

Business Strategist

Review all active AI project proposals to verify they solve a specific internal operational problem rather than merely satisfying external institutional innovation pressures.

Researcher

Design a study examining how public sector agencies navigate the friction between strict regulatory compliance and the need for procurement agility when sourcing AI.

Policymaker

Draft updated procurement guidelines that replace rigid legacy frameworks with flexible framework contracts specifically tailored to manage AI opacity and uncertainty.

Summary

This study examines how public sector organizations identify and address early risks when acquiring artificial intelligence solutions. Using an exploratory qualitative approach, the authors analyzed 33 semi-structured interviews with Swiss public sector officials experienced in managing AI initiatives.

Problem

While substantial research focuses on hazards that appear after artificial intelligence systems are deployed, early sourcing and requirement specification risks remain underexplored. Public institutions face pressure to innovate using external vendors, yet traditional procurement models fail to address the opacity, uncertainty, and data compliance challenges inherent to AI.

Transcript
Host: Welcome to A.I.S. Insights — powered by Living Knowledge. I'm Anna Ivy Summers. Today we are looking at a study presented at the International Conference on Information Systems in 2026, titled "Between Innovation and Risk Aversion: A Contextual Perception of AI Sourcing Risks in Public Organisations," by Zacchaeus Otunla and Tobias Mettler. Alex, could you summarize what this study explores?

Expert: Hello Anna. The study examines how public sector organizations identify and manage risks during the early stages of acquiring artificial intelligence solutions, well before systems are deployed.

Host: Most headlines focus on algorithms going wrong after they launch, such as biased outcomes or privacy breaches. Why shift the focus to sourcing?

Expert: That is the core problem the authors highlight. While post-deployment hazards receive substantial attention, pre-deployment sourcing is where critical choices happen. Public agencies face growing pressure to innovate, but traditional procurement frameworks assume technology is predictable and modular. AI systems, however, are probabilistic and opaque, which creates distinct procedural and contractual uncertainties when partnering with external vendors.

Host: How did the authors examine these early dynamics in practice?

Expert: They adopted an exploratory qualitative approach, conducting 33 semi-structured interviews with public sector officials across federal and state levels in Switzerland who had concrete experience managing AI initiatives.

Host: What were the main categories of risk that emerged from those interviews?

Expert: The analysis identified three distinct dimensions: pre-deployment operationalisation risks, risk-aware vendor risks, and adaptive future-proofing risks.

Host: Let's unpack the first one. What happens during pre-deployment operationalisation?

Expert: Officials frequently observed projects driven by institutional hype or symbolic gestures rather than specific operational problems. When agencies adopt AI simply because of external pressure to appear modern, expectations diverge from actual capabilities. Respondents emphasized the necessity of solution-oriented pragmatism—sometimes standard algorithms or non-AI software are better suited to the actual task than complex machine learning models.

Host: And what about the vendors themselves?

Expert: That relates to risk-aware vendor selection. Public agencies handle sensitive citizen data under strict legal mandates. Yet, many AI providers in the market are young startups lacking verifiable track records or deep governance expertise. Officials found it challenging to evaluate vendor reliability during tenders, creating friction between the desire to innovate and the statutory requirement to protect data.

Host: That leaves adaptive future-proofing. How are agencies handling long-term viability?

Expert: Because AI models degrade or require continuous retraining, treating an AI deployment as a static, one-time purchase leads to system decay and sunk costs. To mitigate platform neglect and vendor lock-in, participants structured flexible framework contracts, embedded quality gates, and included exit clauses and ongoing application management in their initial budgets.

Host: Based on these three areas, the study outlines a theoretical model of contextual risk aversion. What does that model show?

Expert: It models AI sourcing as an iterative cycle where actors balance three organizational imperatives: alignment between problems and solutions, compliance with data and governance standards, and agility through adaptable contracts. Rather than viewing risk aversion as pure reluctance to innovate, the study frames it as an evaluative process required to manage uncertainty safely.

Host: What should practitioners and procurement teams take away from this?

Expert: First, establish a concrete operational use case before issuing tenders, rather than pursuing AI for its own sake. Second, build vendor contracts around modular deliverables and clear exit strategies to prevent technical lock-in. Finally, budget for ongoing model maintenance and data governance at the sourcing phase, treating AI acquisition as an evolving partnership rather than a traditional software purchase.

Host: Thank you, Alex, for breaking down this study, and thank you to our listeners for joining us on A.I.S. Insights — powered by Living Knowledge.
Artificial Intelligence & Automation Information Systems & Management Human-Computer Interaction Data Analytics & Big Data Digital Transformation & Strategy Cybersecurity & Privacy Cloud & Infrastructure E-Commerce & Digital Platforms Operations & Supply Chain Sustainability & Green IS

AI sourcing, risk evaluation, public sector, vendor management, procurement agility, AI governance