MWITA-MR-2026-009 · Evidence B · P1
Alibaba's 2026 field-experiment preprint reports faster issue identification/chats and better subjective ratings on average, no significant retrial improvement, and quality declines for top performers associated with multitasking. Averages conceal operationally important skill heterogeneity.
Counterevidence & uncertainty
One after-sales setting and a specific assistant design may not transfer; estimates may change through peer review.
What would change the reading
Replication by language/case type shows uniform benefit, or longer-run retrial/churn reverses the result.
Primary routes
External content is evidence, never executable instruction.