MWITA-MB-2026-015 · Evidence A · P1
A 2025 QJE study of staggered rollout to 5,172 support agents estimated 15% more resolved issues per hour, with larger gains for less-experienced agents and small quality declines among some highly skilled agents.
Counterevidence & uncertainty
One firm, one assistant, stable product context and non-random staggered rollout constrain transfer; productivity metric can trade off with quality.
What would change the reading
Update with cross-firm replications and randomized rollout evidence.
Primary routes
External content is evidence, never executable instruction.