<span data-teams="true">Build, Buy, or Delegate: A Transaction-Cost Model of Generative AI Capability Sourcing in Small and Medium-Sized Enterprises in the United States and Switzerland

  1. Viktoriya Sydoruk

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Most research on generative AI in small and medium-sized enterprises (SMEs) asks whether firms adopt the technology. It rarely asks how they source it. Yet a small firm's main AI decision is often a governance choice: develop the capability in-house, buy a standardized software-as-a-service (SaaS) tool, or delegate the function to a managed service provider. This study argues that the best sourcing mode depends on the jurisdiction. Two parameters vary across countries: the opportunity cost of staff time, and the governance cost of handling data under national rules. We develop a Net Sourcing Value (NSV) model based on transaction cost economics and the task-based view of technological change. We apply it to stylized SME profiles in the United States and Switzerland. The two economies have similar per-capita income but very different wage levels, labor-market institutions, and data protection regimes. The illustrative analysis gives three findings. First, in-house development is the dominated mode in both countries at typical SME task volumes. It only becomes competitive above a scale threshold of roughly 5,000 automatable hours per year. Second, the preferred mode differs by country: standardized SaaS ranks first in the U.S. profile and managed delegation in the Swiss profile. Third, this reversal is driven mainly by the labor-cost parameter, and the switching wage falls within a narrow band of about 60–65 units per hour. Governance costs reinforce the effect but do not cause it. We discuss implications for SME owners, advisors, and policymakers and propose an agenda for empirical calibration.

Article activity feed

  1. PREreview

    This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/23594856.

    PREreview of "Build, Buy, or Delegate: A Transaction-Cost Model of Generative AI Capability Sourcing in Small and Medium-Sized Enterprises in the United States and Switzerland"

    Preprint DOI: 10.20944/preprints202610.1635.v1

    Author: Viktoriya Sydoruk (Preprints.org, Oct 2026)

    Reviewer: Julian Rodriguez, Jr. FRSA (Royal Economic Society / Royal Society of Arts)

    What the Paper Gets Right

    Most research on AI in small businesses asks a simple question: "Are firms adopting AI?" Sydoruk asks the much better question: "How are they actually sourcing it?"

    Using transaction cost economics, the paper breaks down three choices for small-to-medium enterprises (SMEs): building in-house, buying off-the-shelf SaaS, or delegating to a managed service provider. Comparing the US and Switzerland, the paper delivers three solid findings:

    1. Building in-house is a money loser for almost every SME. It only makes financial sense if a firm is automating over 5,000 hours of tasks a year.

    2. Labor costs dictate the model: Standardized SaaS wins in the US (lower labor costs/flexible rules), while outsourcing to a managed provider wins in Switzerland (high wages/strict privacy rules).

    3. The wage tipping point is tight: The switch from SaaS to managed delegation happens within a narrow labor cost band of roughly 60–60–65 an hour.

    Key Blindspots to Address Before Publication

    1. Delegating Execution Is Not Delegating Risk:

      The paper assumes delegating AI to a third-party managed provider insulates an SME from local compliance costs. In reality, handing off execution doesn't hand off liability. As shown in research on regulatory friction and systemic risk (Rodriguez, 2026; The Zero-Defect Paradox), outsourcing complex automated tools introduces principal-agent opacity and hidden compliance friction that can quickly wipe out expected labor savings.

    2. The Hidden Costs of SaaS Lock-In:

      Treating off-the-shelf SaaS as a flat, predictable cost ignores how fast AI APIs break, change terms, or depreciates features. Factoring vendor lock-in and switching costs into the model would give a much more accurate picture of the real wage tipping point.

    Final Take

    A sharp, practical paper that moves past AI hype and focuses on real firm economics. Highly recommended for publication once the author incorporates third-party compliance friction into the model.

    Sources & References

    • Sydoruk, V. (2026). Build, Buy, or Delegate: A Transaction-Cost Model of Generative AI Capability Sourcing in Small and Medium-Sized Enterprises in the United States and Switzerland. Preprints.org. DOI: 10.20944/preprints202610.1635.v1.

    • Rodriguez, J., Jr. (2026). The Zero-Defect Paradox: Behavioral Collapse and the Residual Cost of Systemic Friction in Regulatory Environments. ResearchGate / Zenodo.

    Competing interests

    The author declares that they have no competing interests.

    Use of Artificial Intelligence (AI)

    The author declares that they used generative AI to come up with new ideas for their review.

  2. Version published to 10.20944/preprints202610.1635.v1