The first time a piece of software you built collides with reality, you find out whether the theory survives. Most doesn’t. The dissertation defended in early 2026 — Implementing Machine Learning for Proactive Cyber Threat Detection and Prevention Using the UTAUT Model — set out to study one of the more persistent puzzles in applied AI: organizations buy detection technologies they then fail to use. Accuracy is rarely the bottleneck. The bottleneck is what the Unified Theory of Acceptance and Use of Technology calls facilitating conditions — the unglamorous infrastructure of permissions, integrations, training, and trust that determines whether a tool ever actually operates inside an organization.
The question
UTAUT, as Venkatesh and colleagues originally formulated it in 2003, predicts technology adoption from four constructs: performance expectancy (will it work?), effort expectancy (is it easy to use?), social influence (do others endorse it?), and facilitating conditions (does the organization support its use?). These are treated as co-equal predictors. Empirical work in low-risk consumer domains has largely validated that framing.
This dissertation asked whether the same symmetry holds in asymmetric-risk domains — settings where the cost of non-adoption is catastrophic and the cost of mistaken adoption is also catastrophic. Cybersecurity threat detection is one such setting. Clinical decision support is another. Defense automation is a third. In each, organizations make purchase decisions under duress, deploy under heavy regulatory observation, and abandon expensive tools at rates that surprise vendors and academics alike.
The finding
The empirical work — a quantitative survey of cybersecurity practitioners across mid-market healthcare and financial-services organizations, analyzed via structural equation modeling — produced a clarifying rather than novel result. The four UTAUT constructs are not equally weighted in asymmetric-risk domains. Facilitating conditions function less as a co-equal predictor and more as a gating threshold. Where they are absent, the other three constructs cannot compensate. Where they are present, adoption follows almost mechanically.
Adoption is asymmetric. The four UTAUT constructs do not sit shoulder-to-shoulder in high-risk domains; facilitating conditions are the whole game.
The formal expression, developed in Chapter 4, runs roughly: Use ~ (PE + EE + SI) × I(FC ≥ θ) — where adoption depends on the additive contribution of performance, effort, and social influence only when facilitating conditions exceed a domain-specific threshold θ. Below threshold, the model collapses; no amount of accuracy, ease, or peer endorsement compensates for the absence of underlying organizational support.
What it means for builders
The implications are practical. Most AI-product strategy in regulated industries optimizes the wrong variables. Sales teams sell accuracy and ROI. Engineering teams ship feature velocity. Founders pitch demos. None of these moves the threshold. The work that moves the threshold is unglamorous: prebuilt integrations with the customer’s existing identity and logging infrastructure, role-aware permissions, compliance documentation written before the customer asks, training materials that fit inside a single onboarding session, and a written commitment that uninstall is reversible.
Every venture in the operating portfolio that grew out of this dissertation — OBSIDIAN, Therapee, Arkc / MeshCam, the Genesis Armor defense-tech portfolio — is engineered with facilitating conditions as a first-class concern, not a deployment afterthought. The argument is developed for a non-academic audience in the essay “Facilitating Conditions Are The Whole Game”.
The committee and the publication path
The dissertation was completed at the University of the Cumberlands under Dr. Irvin Heard, former dissertation chair. Defense was conducted in early 2026; degree was conferred shortly thereafter. The manuscript is being prepared for SSRN posting after committee revisions clear. A condensed working-paper version is in preparation for submission to MIS Quarterly.
What comes next is research designed in the opposite direction. The first doctorate studied how organizations adopt AI; the planned second doctorate, in Intelligent Systems and Robotics at the University of West Florida under Dr. Brent Venable, turns the question upstream — toward how AI systems should be designed from first principles so that adoption is structurally easier rather than retrofitted for.