Two dialers, one queue. Power dials one line at a time for full control; predictive runs up to three lines per rep to keep the conversations coming. Both sit behind answering-machine detection we built ourselves — so your reps hear "hello," never a beep.
Most platforms pass your call to a carrier's answering-machine detection and live with the result. We run our own classifier on a live audio stream, because the carrier's answer wasn't good enough.
The moment a call is answered, the audio is forked to a live speech stream and classified as the greeting arrives — usually settled well before a carrier would even report back.
If the classifier isn't sure in time, the call is treated as a machine. The tradeoff is deliberate: a rep should never be dropped onto a voicemail greeting, so ambiguity resolves the safe way.
Carrier premium AMD runs alongside ours purely to catch the beep, so a dropped voicemail starts at the tone instead of talking over the greeting.
| Setting | SellifyGPT | Typical carrier AMD |
|---|---|---|
| Who decides human vs. machine | Our own classifier, on live audio | The carrier's black box |
| What it listens to | The live greeting, transcribed as it plays | Tone and cadence heuristics |
| On an unclear call | Treated as a machine — rep protected | Frequently bridged to the rep |
| Thresholds | Tuned by us against real call data | Rarely exposed at all |
Same contact queue, same scripts, same dispositions. The difference is how many lines run at once.
One line at a time, dialed the instant you finish the last call. Answering-machine detection still stands between your rep and every voicemail.
Up to three lines per rep, running in parallel, with the pace pulled back automatically whenever your abandon rate climbs.
The FCC's safe harbor caps abandoned calls at 3%. Most dialers make that your problem. Ours makes it the pacer's problem.
Measured as a rolling 30-day rate per campaign — because the regulation treats the campaign, not the individual rep, as the thing being measured.
Cross 2.5% and the pace eases back about 25% per minute until the rate recovers. Older dialers halve your throughput the instant you cross the line; this one steers.
Throttling needs a meaningful sample before it acts, so three unlucky calls on a Monday morning don't strangle the queue. Voicemail drops are excluded from the math entirely.
The mistake that gets teams sued isn't malice — it's arithmetic. You dial a list at 6pm from California and a chunk of it is in Florida, where it's 9pm.
Every campaign call is checked against the local time where that contact actually is, resolved from their state and falling back to their area code. Daylight saving is handled properly, because it's the thing everyone gets wrong.
A contact outside their window isn't lost. On the predictive dialer they're rescheduled to the exact moment their window opens and picked back up automatically — no requeue to manage. On the power dialer the call is simply blocked and the rep moves on.
The default is 9am–9pm local. Set a tighter window per campaign if your industry or your own policy calls for it.
Timezone is inferred from the contact's state, or their area code if state is blank. When neither resolves, the contact is treated as callable and the call goes out. A mobile number that moved across the country still reads as its original area code. This meaningfully reduces your risk; it is not a legal guarantee, and it doesn't replace knowing the rules for what you sell and where.
Calling a 480 number with no 480 in your pool? It picks the closest area code by actual physical distance, so a Phoenix lead sees a Phoenix-area number. Your numbers are yours — it will never reach into another tenant's pool.
Leave a pre-recorded message and move on without waiting. Configure which attempt in a cadence starts dropping — hang up on the first two, leave a message on the third.
"Sale!" opens a deal. "Call Back Later" schedules the callback. "Take Off List" marks the record Do Not Call — stopping the dialer and the texts, and pulling them out of drips and flows. Eleven come built in; rename or replace any of them.
Recording is gated globally, per tenant, and per campaign — and any error means no recording. Reps are responsible for verbal disclosure where their state requires it.
Hand a live call to another rep, or forward calls to your mobile with your own timeout and business-hours rules.
The moment a call bridges, the script opens with that contact's details already filled in — no hunting for the right tab.
That's the specific thing the system is built to prevent. Answering-machine detection runs on the live audio of every power and predictive call, and if it can't reach a confident verdict in time it treats the call as a machine rather than risk bridging a rep onto a greeting. The tradeoff is that a person who answers and then stays completely silent for several seconds can be treated as voicemail.
The power dialer runs one line per rep, so it can't abandon a call — it's the right choice for warmer or higher-value lists, and it's included on every plan from Starter up. The predictive dialer runs up to three lines per rep in parallel to raise conversation volume, with the pace automatically governed by your abandon rate.
The pacer measures your abandoned calls as a rolling 30-day rate per campaign against the FCC's 3% safe harbor. At 2.5% it starts easing the pace back about 25% per minute rather than cutting throughput off a cliff, and it waits for a meaningful sample size before throttling at all. Voicemail drops aren't counted as abandons.
Yes. Campaign calls are checked against the contact's own local time — worked out from their state, or their area code if state is blank — with a 9am–9pm default you can tighten per campaign. On the predictive dialer they're rescheduled to the moment their window opens; on the power dialer the call is blocked and the rep moves on. Two honest limits: a contact whose timezone can't be determined at all is treated as callable, and a rep manually keying a number on the softphone isn't blocked, since that's usually a call the person asked for.
Yes. Add numbers to your pool and the dialer picks the best match for each contact — an exact area-code match first, and if there isn't one, the geographically nearest area code you own, measured by real distance rather than a lookup table.
No. It's off until an admin turns it on, and it can be disabled per campaign. Where your state requires verbal disclosure that a call is recorded, that remains the rep's responsibility on the call.
Start free today — bring your contacts, fire up the dialer, and let the AI coach do the rest.