Every remote team has The Dog. Somewhere on your roster is an excellent agent whose household includes a dog with strong opinions about delivery vans, and at least once a week those opinions reach a customer. The bark is the single most recognizable WFH noise event — instantly audible, mildly comic, impossible to schedule — and how a team handles it says everything about whether its audio policy was written by adults.
Why the bark became the mascot
The bark earned its status honestly: it's loud (a burst that rides over speech), transient (gone before anyone can react), and universal (dog ownership is common across every geography WFH agents work from). It is the perfect specimen of our A2 defect class — the transient noise event — precisely because no reasonable policy can prevent it. A doorbell can be disconnected during shift; a TV can leave the room; the dog is a family member with a job of its own, and its job is announcing vans.
That's what makes it such a useful policy test. Any audio standard that would, read literally, require an agent to not have a dog is a standard that selects against normal human households — and your best agents will read it exactly that literally, then quietly start interviewing elsewhere.
What policy should actually say
Three lines, no more:
- Transient pet noise is expected and never disciplined. Mute, handle it, apologize briefly if it broke through, move on. A two-second "sorry — dog versus courier" costs nothing; most callers are on the dog's side anyway.
- Sustained pet noise is a setup conversation. If the dog barks through entire calls daily, the fix is spatial (a different room, a closed door, shift timing around the dog-walker) — worked through together, not decreed. The escalation ladder in our WFH standard applies unchanged: schedule first, eligibility only if scheduling fails.
- The software layer is mandatory and company-paid. Modern noise cancellation eats the mid-bark burst impressively well, and what it can't fully swallow, it shortens from an event into a blip. This is exactly the job the agent-side layer exists for — see the ops guide for what that layer costs and how it deploys.
What software catches, honestly stated
Docs-based honesty, since we don't publish fabricated tests: vendors in this category train specifically on dog barks — Krisp's call-center materials list background noises, voices and echoes as removal targets, and barks are the canonical demo (krisp.ai, read ). Two real limits to keep in your expectations. A bark that overlaps the agent's own speech is genuinely hard: suppression must separate two overlapping sounds, and the residue is sometimes audible. And no outbound layer helps when the bark is on the caller's side — that's the inbound half of the problem, and it's your agent who needs the rescue.
The one metric worth tracking
If pets keep appearing in QA notes, stop counting barks and start counting recoveries: did the agent mute promptly, keep the thread of the call, and re-engage cleanly? A2 events with clean recovery are noise in the statistical sense — they carry no customer-experience signal. A2 events with fumbled recovery are a coaching topic about composure, not about dogs. Tag them accordingly (the scheme is in the metrics guide) and The Dog stops being a QA argument at all.
- Mute promptlythe two-second reflex
- Keep the threadthe call carries on where it was
- Re-engage cleanlya clean recovery carries no customer-experience signal
The wrap
"Get rid of the dog" is a resignation letter with your signature on it. The grown-up stack is: a three-line policy that assumes good faith, a paid-for software layer that makes most barks academic, and a QA lens that scores the recovery, not the pet. Do that, and The Dog becomes what it always deserved to be — the team's unofficial mascot, occasionally audible, professionally irrelevant.