But we piloted speech analytics prompts with 24 agents for 6 weeks; AHT dropped 9% and average hold time fell 22 seconds, but QA alignment slipped because the prompts didn’t map cleanly to our scorecard behaviors. For those running RT assist at scale, how are you tying prompts to coaching cadence and calibration so supervisors aren’t fighting alert fatigue?
We fixed this by tagging each nudge to a scorecard ID and gating it: it only fires until ASR detects the behavior, then goes on a per-call cooldown. Using those IDs, we auto-roll misses into the weekly coaching queue so supervisors see just the top three repeat gaps per agent — fewer sirens, more seatbelts. @Priya, are you tracking nudge acceptance/overrides so you can tune thresholds by team?
Building on @marcus_lee27: gate by coaching focus — only surface nudges for the two scorecard behaviors on each rep’s current plan, suppress the rest, and auto-expire after QA logs 3 clean calls… We also cap at one alert per call phase (greeting/discovery/wrap) to avoid supervisor alert fatigue. If your 6-week pilot used multi-step behaviors, split them into sequential micro-prompts tied to the rubric to restore QA alignment.