Synthetic testing for phone trees · macOS and Windows
Test your phone tree the way customers reach it. By calling it.
Earshot dials your number over a SIP trunk, presses the keys a caller would press, records the far end, transcribes it locally, and asserts every prompt, spoken value and timing gap — with the audio slice attached to each result.
Synthetic test calls, automated results. Runs on your Mac or PC. Uses your SIP trunk. Recordings never leave the machine. 14-day free trial.
- ~1,000 scenarioswritten for one customer program.
- 412 s · 11 · 0412 s of call, 11 assertions, 0 people on the phone. A queue-treatment run at 1 AM.
- 90 s + 120 s90 s of hold music and a 120 s delay cadence, proven. Not estimated with a stopwatch.
- 20-bug UAT roundverified overnight, each bug with its own audio slice.
- NICE CXoneSupported
- Avaya Aura · AAEP · Communication Manager
- Genesys Cloud
- Amazon Connect
- Cisco
- Five9
- Twilio
- Asterisk / FreePBX
- Anything that answers a call
Everything else about your contact center is tested. The phone tree is tested by a person with a stopwatch.
A contact-center engineer changes a routing rule, re-records a prompt, or edits an hours profile. To find out whether it worked, somebody picks up a phone, dials in, listens, waits out the hold music, times the delay messages by hand, and types what they heard into a ticket. Then the fix lands and they do it again.
That costs about seven minutes per queue, per retest, per person. It produces no artifact. Nobody can prove the timing afterwards, because nobody recorded it. So UAT bugs read like this:
"did not get 2nd delay message after 120 sec MOH"
"heard ONE after the survey message"
Both are real. Both are unfalsifiable without a recording. And the regressions nobody files are worse: a re-record that drops one wav, an hours edit that sends daytime calls to the closed treatment, a routing change that quietly reassigns a skill. None of these fail loudly. They fail to a customer, weeks later.
No evidence
The ticket says what the tester remembers. The audio is gone.
No repeatability
Every retest is a person, a phone and seven minutes. Nobody retests all forty flows.
No regression net
Prompt re-records, hours changes and routing edits break silently.
A real call, a recording that grades itself, and a test that fixes itself.
You define a scenario — in the builder, in YAML, from a sheet or with your own AI key — and Earshot places the synthetic call, records it, transcribes it on your machine, scores every step and files the verdict in the test log. Then it does the parts a person used to do by ear.
Discover mode, golden baselines, ANI/DNIS matrices, the desktop app and the privacy model → the full product tour
An Avaya to NICE CXone migration, roughly 1,800 numbers.
A regional telecom moved its contact center from Avaya to NICE CXone. The cutover surface was about 1,800 numbers and the UAT window was measured in days.
A 20-bug UAT round, verified overnight.
The testers filed twenty bugs at the end of a day. The fixes went in, a scenario per bug ran unattended overnight, and the tester list came back the next morning with a verdict and an audio slice against every line. No coordination call, no shared phone, no "works for me".
Queue treatment proven in 30 minutes, unattended.
Four skills, each with the same six-checkpoint sequence: QA disclaimer → 90 s of hold music → callback offer → first delay message → 120 s → second delay message. Twenty-four checkpoints, all timed from the recording. By hand that is four people-hours of sitting on hold, and the timings are still somebody's stopwatch.
A billing self-service path, end to end.
Stored-card payment verified all the way through, including the spoken confirmation values — the amount and the confirmation number read back by the IVR, asserted as numbers rather than as strings.
A platform-wide config bug, found by silence.
A subroutine was blanking screenpop and callback flags on every single call. It surfaced because the recordings showed an offer that never played. Nothing in the platform reported an error. The evidence that it was broken was the absence of eight seconds of audio.
"Nobody can prove timing, and nobody has audio evidence"— that is the state of IVR testing that Earshot was written against.
Put your own numbers in.
Seven minutes per manual retest is the measured figure from the case study. Everything else below is yours to set.
Retest cost per month
inputs, not measurementsThe migration preset is the shape of a cutover like the one in the case study — roughly 1,800 numbers collapsing to about 120 distinct flows.
Runs in your browser. Nothing is sent anywhere.
Solo at $49/month pays for itself at 0.7 hours saved.
Set at least one flow to see the number.
Assumes one minute per run to read the result instead of seven minutes on the phone. Calls run unattended, so the wall-clock time is not the engineer's time.
Migrating a contact center, or maintaining an IVR nobody can safely change?
Tell us what you are testing and what it costs you today. We will walk you through a run against your own number, or answer by email from hello@earshot.channel.