Who adopts open source? Wrong question. Ask who the adopter is.¶

A plain-language companion to the draft working paper "Adopting Open Source in 2026: A UTAUT Model Grounded in the FLOSS User Typology" (v0.4, July 2026). The paper carries the model, the instrument, and the analysis plan; this text carries the ideas. This is a research design: the survey has not been fielded, and no result is claimed.

Read the full draft working paper (PDF)

The short version. Ask why organizations adopt open source and you get a list of factors: cost, quality, community, security, independence. Twenty years of surveys have produced such lists, and the lists wobble from study to study, partly because they average over organizations for which the question means different things. An organization that writes code weighs open source differently from one that cannot read it. The paper builds the survey that takes this seriously: a standard technology-acceptance model whose respondents are sorted, before any data arrive, by what they can actually do with source code. The design states which drivers should matter for whom, commits to its tests in advance, and is candid about the sample it needs. The survey itself is the next phase.

Three kinds of adopters¶

The segmentation comes from the economics of open source rather than from marketing demographics. The theory of Jullien, Viseur and Zimmermann (2025) carries a classification of adopters, descending from von Hippel's work on user innovation, by two capacities: can the organization translate its needs into technical requirements, and can it develop code? Both capacities make an Innovative user, which can serve itself. Specification without development makes a Frontier user, which knows what it wants and must buy the building. Neither makes a No-skill user, which must buy even the knowing. A second axis sorts what the organization needs: standard and cheap, personalized, durable and lock-in-proof, or personalized and durable at once.

The theory's punchline is self-provision. Whatever a segment can produce in-house is cheap for it, so external supports for that thing carry little decision weight; whatever it cannot produce must come from the environment (a project's community, a vendor, internal support machinery), so perceptions of that environment decide the adoption. The same logic, read from the other side, explains what open-source companies sell, which is why this survey is one half of a pair with a supply-side companion paper.

What should matter where¶

The measurement apparatus is UTAUT, the standard workhorse of technology-acceptance research: performance expectancy, effort expectancy, social influence, and facilitating conditions, kept deliberately lean against the model's documented tendency to sprout constructs. Three open-source-specific constructs are added, each with an economic channel the core does not carry: community support (assistance the adopter did not buy), transparency and auditability (verification the adopter controls), and strategic alignment (the independence and sovereignty motive). Two 2026 institutional forces enter as moderators: perceived regulatory risk from the Cyber Resilience Act and the AI Act, mitigated by commercial assurance, and sovereignty, which for European public buyers now carries an institutional mandate on top of its option value.

The central hypothesis crosses the constructs with the segments: effort and support-side drivers should bind hardest for No-skill and Frontier adopters, while performance and strategic alignment should dominate for Innovative ones. A capable organization barely notices the effort cost of adoption; an organization that must buy every adaptation lives on ease, help, and assurance. A companion methods paper derives this prediction from a cost model and disciplines its test: cross-segment comparisons must run on ratios of effects or at matched baselines, because naive comparisons can flip the answer.

A survey that shows its cards¶

Design papers are cheap to write and easy to oversell, so this one binds itself in advance. Every construct is measured by named items with stated provenance. The classification into segments uses concrete capacities (could your team fix a defect in source? write an implementable specification?) rather than self-flattery, and it is validated against a hard behavioral fact: whether the team has had a contribution accepted by an open-source project it does not lead. The analysis plan is pre-registered and frozen before data collection, with exclusion rules and fallbacks stated.

The candor extends to statistical power, where the news is mixed and the paper says so. The computed power analysis shows the headline three-segment comparison would need 650 to 800 respondents, while the realistic recruitment channels (European open-source associations) are expected to deliver 150 to 300. The design responds by registering two feasible co-primary tests (a two-group contrast and a continuous-capability moderation), setting a floor of 100 completes per segment with a target of 450 or more, and committing to report the sensitivity actually achieved whenever the sample falls short. If recruitment lands at the low end, the paper reports what the data can support and defers the strong segment claims to a second wave.

The small print¶

The sample will be a specialty sample: organizations reached through open-source associations, which selects for engagement with everything the survey measures. The design accepts this openly, because the population of interest cannot be reached by representative sampling at any feasible budget: all claims are within-sample and structural (which drivers bind where), no population share will ever be reported, and comparisons to representative benchmarks are labeled as approximate context. The outcome is stated intention, anchored in current practice but still self-reported. The segmentation belongs to the cited literature; the paper's contribution is the instrument, the registered analysis plan, and the discipline around the central test. Until the survey runs, the model asserts no coefficient, and the paper says so on its first page.