Three experiments find AI that explains itself lifts customer satisfaction after small failures

The benefit depended on how bad the failure was. When the service failure was low in severity, transparency worked through both warmth and competence. When severity was high, those indirect effects were not significant, which the authors identify as a boundary condition for transparency.

The work treats transparency as a kind of anthropomorphism, a cue that makes the AI's communication feel more human, and applies social cognitive theory to AI-mediated service recovery. The findings come from experimental scenarios, not from observed customer interactions.

The research was built in three steps. The first experiment, with 98 participants, tested whether AI transparency affects recovery satisfaction directly. The second, with 230, tested whether perceived warmth and perceived competence each carry part of the effect. The third, with 284, added service failure severity as a moderator.

Results show transparency improving recovery satisfaction indirectly through warmth and competence. Severity moderated the effects of transparency on both mediators, so the pathways held under low-severity failures and were not significant under high-severity ones.

The authors frame transparency as an anthropomorphic cue that exposes the AI's thought process, extending social cognitive theory to AI-mediated service recovery.