Dialer

Single line or parallel, decided per contact

Parallel dialing buys connects and costs hang-ups, so it is not applied uniformly. Each contact is scored on value and on how likely the line is to be picked up right now, and the two decide the mode.

• A high-value contact, or one likely to answer, gets a dedicated single-line dial. Nothing is worth hanging up on a hot lead.
• Lower-value, lower-pickup contacts are batched onto parallel lines, more lines the less likely a pickup is, up to a ceiling of five simultaneous lines.

Pickup likelihood is estimated from the recipient's own local hour and day, whether the line is mobile, and that contact's past connect rate where one exists. Mid-morning and late afternoon local time score highest; late night and the lunch hour score lowest.

When a call may be placed

Two windows apply, both in the called party's local time, and both starting at 08:00.

08:00 to 17:00

The business window. The default for a cold contact.

08:00 to 21:00

The TCPA floor. The hard ceiling nothing crosses.

An existing customer, or a contact who asked to be contacted, can be called past 17:00, but the relaxation is clamped by construction: the end hour is the lower of what was asked for and 21:00, so no exception can push a call outside the legal window rather than merely being told not to.

Recording consent is resolved from the number

Whether a call may be recorded depends on where the person you are calling is, not on where you are. The callee's US state is resolved from the area code, and fourteen US states are treated as requiring all-party consent, alongside seven countries. A workspace can carry its own per-jurisdiction overrides on top.

Each entry in that list carries its own provenance, so a classification taken from verified statutory text is not presented at the same confidence as one carried over from a seed list. Two states are flagged as possibly misclassified and are deliberately erred toward all-party, which is the safe direction.

Disclosing an artificial voice is a separate obligation and is handled separately. A voicemail this dialer drops is synthesized, so where state law requires the caller to say so, the disclosure is added. A jurisdiction that cannot be established as a US state gets the disclosure too, because an unnecessary sentence in a voicemail costs nothing and an omitted one is a per-violation penalty.

Abandoned calls

A call is abandoned when a live person answers and no rep is bridged to them within two seconds of their greeting. That is the FCC definition, and three things happen when it occurs.

1. 1A durable abandonment record is stamped on the call session. It is both the safe-harbor record and the only thing counted as an abandonment, so the rate cannot be inflated or deflated by a raw call status.
2. 2The required recorded message, naming the caller and the phone number, is recorded as pending on the session and marked undelivered until the telephony plane confirms it played. The record never claims audio it cannot confirm.
3. 3The number is suppressed for the campaign, so an abandoned party is not dialed again.

The abandonment rate is governed against a 3 percent cap measured over a rolling 30-day window. A workspace can be configured to a stricter target; it cannot be configured to a looser one, because the configured value is clamped down to the cap.

Before the first call

Voice has its own channel switch on the outbound gate, which is closed unless it has been deliberately opened for the environment, and the platform-wide kill switch and the per-workspace pause apply to calling exactly as they apply to mail. Suppression is checked on the phone number as well as the email address, so an opt-out on one channel stops the other.