Night Shift Audio: Keeping Calls Quiet at 3 A.M.

A support call runs at conversation volume, hour after hour, in a flat where everyone else is asleep. Night agents know the consequences intimately: the partner who has to be up at six, the toddler whose room shares a wall, the neighbour who has started timing the noise for the building group chat. So they adapt — by whispering. And whispering is an audio defect: low signal, breathy texture, compressed dynamic range. The caller strains, asks the agent to repeat, and the agent — trying not to wake anyone — repeats quietly again. That loop is our A4 defect code ("agent voice low, distant or muffled") wearing pyjamas; the full code list is in the metrics guide.

The honest fix is not "speak up." It's a setup where normal-adjacent speech is possible: the room farthest from sleepers, a door that closes, mic close enough that a low-but-voiced register still reads full — close-mic'd quiet speech sounds dramatically better than distant-mic'd quiet speech, because the signal-to-room ratio is what callers actually hear.

Night shifts do hand you one gift: the unpredictable noise events that plague day WFH — deliveries, school runs, the neighbour's mower — mostly sleep too. Ambient scores at night are typically the cleanest of any shift. Which produces the night-audio paradox: outbound cleanliness is nearly free at 3 a.m., while the hard problems move to the agent's own voice level and to the inbound side. Night callers skew toward exactly the environments software exists for: drivers on empty highways, warehouse and hospital staff on breaks, travelers in airports where it is very much not 3 a.m. Cleaning what your agent hears is the two-way noise cancellation case, and it's arguably stronger on nights than days.

• Voice-level amnesty, setup accountability. Never coach a night agent to "project more" without first funding the room change that makes projection possible. Volume advice into a sleeping household is policy malpractice.
• Schedule the loud work. Outbound blocks, escalation callbacks and anything script-heavy belong at the shift's edges, when households are more awake. WFM can move noise in time on nights exactly as it does on the day floor.
• Let the software carry the margin. A noise-cancellation layer won't make a whisper loud, but it strips the 3 a.m. artifacts — fridge hum, fan noise, the creak-and-rustle of a quiet room — so a low-voiced agent spends every decibel on speech. The deployment math is the same as any seat; see the ops guide.

Night agents are running the hardest audio brief in the company: be clearly audible and nearly silent, simultaneously, for eight hours. Treat their setup requests as engineering requirements, not comfort items. And if you run QA calibration, score a batch of night calls as their own cohort at least once — the defect mix is so different from days that a single blended threshold quietly punishes the people working while you sleep.

Related reading: the full WFH agent audio standard — the household-noise policy template there includes the scheduling clause night teams rely on.