AI-Label vs. Human-Label: Perceived Usefulness of Process Innovations

Topic

This thesis examines whether disclosing the source of a process innovation, namely whether it is attributed to an Artificial Intelligence or a human expert, influences how useful that innovation is perceived to be. Set within Business Process Management, it investigates this labeling effect across two contrasting application domains: a utilitarian context (used car purchase) and a hedonic one (home caregiving). Drawing on the Technology Acceptance Model and cognitive heuristics research, it tests whether source attribution alone shapes evaluations independently of actual content quality, and whether this effect depends on the business domain in which the innovation is assessed.

Relevance

Organizations increasingly deploy Generative AI to design and optimize their workflows, yet adoption frequently fails for psychological rather than technical reasons. Evidence shows that simply labeling output as AI-generated can trigger systematic devaluation, regardless of objective quality. For practitioners introducing AI-driven process innovations, this means that resistance may stem not from the innovation itself, but from how its origin is communicated. Understanding when and why source disclosure undermines acceptance allows managers to design communication and rollout strategies that reduce evaluative resistance that is disproportionate to actual process quality, particularly in high-stakes domains where evaluator resistance is most consequential.

Results

AI-labeled process innovations were rated as less useful than human-labeled ones, but this direct effect narrowly missed conventional significance (p = .055, d = 0.39), likely due to limited statistical power and a ceiling effect. Contrary to theory, the AI-label penalty was larger in the utilitarian domain than in the hedonic one, reversing the expected pattern. The clearest finding was mediation: human-labeled innovations were perceived as requiring more effort, and this perceived effort significantly raised perceived usefulness, accounting for roughly 40% of the total effect. Source attribution thus shapes evaluations primarily through the effort heuristic.

Implications for practitioners

• The present findings suggest that the AI-label penalty on perceived usefulness may vary across process domains in ways not fully predicted by existing theory, warranting domain-sensitive caution during introduction.
• Make the development effort behind AI-generated processes visible, including validation steps and quality checks, as perceived effort is the primary pathway through which source attribution shapes perceived usefulness.
• Adapt source disclosure strategies to the specific application domain rather than relying on a uniform, one-size-fits-all rollout approach.

Methods

The study employed a quantitative 2×2 between-subjects factorial design, manipulating source label (AI vs. human expert) and domain type (utilitarian vs. hedonic) across four conditions. Working professionals in the DACH region (N = 102) were recruited online and randomly assigned via Qualtrics to evaluate a process innovation vignette in which only the source label varied. Perceived usefulness, perceived effort, and two covariates were measured on six-point scales. Hypotheses were tested in R using an independent samples t-test (H1), a two-factorial ANOVA and ANCOVA (H2), and a bootstrapped simple mediation analysis with 5,000 samples (H3). All scales exceeded the reliability threshold of α ≥ .70.