Alhena found answer-only fallback in 10 of 15 live deployments in 2026, the most common failure in the study. The conversation stayed in channel, so containment looked healthy.
Alhena logged seven failure patterns across 15 live AI CX deployments. Answer-only fallback led at 10 of 15. Six of the seven trace to missing architectural layers.
Alhena tested 15 live AI agents on four tasks that need reasoning before retrieval. Only 6 of 15 could route a shopper to the product they recommended.
Alhena tested 15 live AI agents in supplements, the highest-stakes vertical tested. Unsafe confidence appeared in 3 of 15. Knowing when not to answer scored as a strength.
Alhena tested 15 live AI agents on a hypoallergenic gift brief. Two jewellery deployments scored 2.63 and 1.25 - the widest gap in the study. Gifting is mostly memory.
Alhena tested 15 live AI shopping agents on foundation shade matching. The best scored 2.75/3.0. Five ignored the selfie entirely and returned bestsellers.