Quick answer: Voice broadcasting is the outbound voice pattern where one recorded message — a pre-recorded audio file or a text-to-speech script — goes to every contact on a list, paced so a threshold of concurrent calls respects your list and your carriers. In Devotel Orbit the dedicated broadcast wizard lives in the dashboard under Outbound → Voice Broadcasts (https://orbit.devotel.io), and everything you need — audio, caller ID, audience, schedule, and an optional A/B variant — is one six-step screen from a compliant send.
This guide is the complete playbook for running that wizard well: which metric actually measures success, how the A/B test works, which compliance guardrails are enforced for you versus which stay tenant-owned, and when a broadcast is the wrong tool entirely.
Answer rate is the north-star metric, not dials
The number a broadcast program should be judged on is answer rate — the share of dialed contacts whose call actually connects — for three reasons.
First, a dial is a cost, not an outcome. A broadcast that dials 50,000 numbers and answers 8,000 of them delivered a message to 8,000 people; reporting "50,000 dials" measures effort, and effort is the same regardless of whether anyone heard the script.
Second, answer rate is the one engagement signal a voice-only send can honestly record. Orbit's A/B engine understands opened, clicked, and replied for SMS and email; a phone call has none of those. The connected/disconnected outcome is the whole measurement surface, and answer rate is the platform winner metric that maps to it. That is why the broadcast wizard fixes the success metric to answer rate — the alternative winner metrics are structurally unmeasurable on voice and would leave a test that can never conclude.
Third, answer rate is where the levers actually are. Caller-ID ownership and attestation posture, the time of day you dial relative to each recipient's business hours, how clean the list is before you start, and whether numbers that went nowhere keep getting redialed — all of these move answer rate, and none of them move dials. Optimize the metric the levers attach to.
A/B-variant testing — two audio scripts, one measurable winner
The broadcast wizard ships an optional A/B step: you write (or upload) the primary script and a second "Variant B" script, and a holdout share of your audience hears the two scripts split evenly while the rest waits for a winner. The configuration is concrete:
- Test traffic — the share of the audience put into the holdout, 1–50%, split evenly between the two scripts. The remainder hears only the winning script once one is declared.
- Success metric — fixed to answer rate, for the measurability reasons above.
- Auto-optimize — when enabled, Orbit continuously shifts more calls toward the leading script as answers accumulate, instead of holding an even split until formal winner selection.
Winner selection is auditable, not a vibes call: each variant needs at least 30 answered calls before it counts toward a result, Orbit waits 24 hours from the first call so late answers still enter the comparison, and the leader wins once the result clears the statistical-confidence bar or leads the runner-up by at least 10%. You can review live per-script results and override the winner at any time on the campaign page.
Two working notes that save real campaigns:
- Both scripts dial real contacts, so treat Variant B like the primary: run the same listen-back preview on it before you create the broadcast, not after the first wave of calls.
- Change the message content between variants, not the delivery. Keep the same caller ID, TTS voice, and schedule across A and B or the test measures your setup rather than your script.
Deliverability and the compliance guardrails
A voice broadcast lives or dies on whether carriers and recipients let it through. Some of the guardrails below are platform-enforced; the ones that are tenant-owned are on you to keep there.
Answered-number provenance — exclusion math, before the first dial. The wizard's audience step computes a net hourly projection for your chosen audience and breaks it down before you send: contacts with no valid phone number, contacts opted out of calls, and contacts on your do-not-call suppression list are each counted and subtracted. The eligible count you see is the number the send path will actually dial. That means a "we called someone we shouldn't have" incident becomes traceable to the exclusion breakdown you saw at authoring time — check it before every send, and treat a surprising exclusion count as a list problem, not a platform surprise.
STIR/SHAKEN positioning. US recipients increasingly see caller-verified or "Spam Likely" labels based on the attestation level your calling number carries. Numbers your organization owns through Orbit attest at the highest level because number authorization is provider-verified, while external caller IDs and leased pool numbers cap at a lower level and underperform accordingly. If your broadcast answer rate is depressed, the attestation posture of the caller ID you picked is the first thing to check. The full mechanics are covered in the STIR/SHAKEN attestation guide.
Quiet hours — two layers, one yours to set. TCPA federal quiet-hours rules are always enforced on Orbit voice broadcasts; the wizard cannot schedule around them. On top of that, the "Respect business hours per recipient" option (on by default) gates dialing to 09:00–18:00 in each recipient's own local timezone. Keep it on for marketing-adjacent traffic and reminder traffic alike — the 9-to-6 window is the highest-answer band for most lists, and evening dials are where complaints originate. State-level windows narrower than the federal rule still apply where relevant; the business-hours gate handles common cases, your legal review handles the rest.
Tenant-owned obligations, not platform settings. Recording-consent disclosure, list provenance (that every contact consented to receive calls), and honoring opt-outs promptly remain your organization's responsibilities. The wizard requires acknowledging recording-consent disclosure before any voice campaign can save — the acknowledgment is a reminder, not a transfer of the obligation. Frequency caps and suppression lists are enforced at execution time against your configured policy; tune those per your program, and scrub known-litigator lists where you operate in the US.
When to use a broadcast vs agent-driven outbound vs an IVR tree
| Pattern | Use it when | The message |
|---|---|---|
| Voice broadcast | One message, thousands of contacts, no interaction needed | Outage notices, appointment reminders, policy updates, event announcements |
| Agent-driven outbound | The call needs a human conversation | Renewal negotiation, qualification, anything with branching live discussion |
| IVR / AI voice agent | Recipients respond, choose, or transact | Surveys with menus, payment collection, self-service intake |
The broadcast is the cheapest per contact and the least flexible. Broadcast first, and route contacts who respond to the message toward an agent or an IVR via the callback number in the script. If half your "broadcast" calls keep turning into conversations, that traffic belongs on a dialer queue, and the honest move is to split the list.
Worked example — a 50,000-contact campaign
A facility-services company sends a quarterly maintenance reminder to a 50,000-contact export.
Step 1 — check the exclusion breakdown. The net projection shows 50,000 matching contacts, minus 900 with no valid number, minus 2,100 opted out of calls, minus 300 on the suppression list. That leaves 46,700 eligible recipients — the figure everything below should be computed against, not the raw export.
Step 2 — baseline. At a conservative 35% answer rate for an A-attested own-number broadcast in business hours, that is roughly 16,345 answered calls. At the 15% answer rate typical of unattributed external caller IDs, the same list answers about 7,005 — ownership of the caller ID more than doubles the result before any script work happens.
Step 3 — run the A/B holdout. With test traffic at 20% (the wizard default), about 4,670 recipients per variant hold the even split while the remaining 70% (roughly 32,690) wait for the winner. Both variants cross the 30-answered minimum on the first dialing window, the 24-hour evaluation window closes, and the leading script — say it answers at 38% versus the base at 35% — clears the 10% lift bar.
Step 4 — the winner ships. The remaining audience hears the winner. Expected answered calls: the holdout answered about 1,730, and the remainder at 38% adds about 12,420 — roughly 14,150 connected contacts, versus the roughly 16,345 that the plain-send projection quoted at the pooled baseline. Modeling the shortfall explicitly — that the test reserve a fraction of the audience to learn on — is the difference between a report and an excuse.
Step 5 — keep the recording. Both scripts, the holdout size, the per-script answers, and the declared winner are reviewable on the campaign page after the fact, so the next quarterly campaign starts from a tested script rather than a blank one.
Frequently asked questions
Do I need consent to send a voice broadcast?
Yes. Voice broadcasts to US numbers fall under TCPA rules for prerecorded and artificial-voice calls, and marketing-adjacent broadcasts generally require the recipient's prior express consent — and for telemarketing, written consent. Orbit enforces the mechanical guardrails (quiet hours, opt-out suppression, frequency caps) and blocks a voice campaign from saving without recording-consent acknowledgment, but the consent itself is a tenant-owned legal obligation. Run your list provenance past counsel before you dial.
What answer rate should I expect from a voice broadcast?
It swings with caller-ID attestation and list quality more than with script quality. Owned, A-attested numbers with a clean list routinely answer in the mid-30% range in business hours; external unattributed caller IDs on the same list often land below 15%. Attestation posture is the single biggest lever, which is why the wizard's caller-ID picker defaults to your active owned numbers.
Can I use my own recorded audio instead of text-to-speech?
Yes — the wizard's Recorded Audio tab accepts MP3, WAV, or M4A up to 25 MB, uploads the clip as a server-side asset the voice gateway can play, and gives you a listen-back before the send. TTS is the better default when you need per-contact personalization ({{first_name}} and other variables render per recipient); recorded audio wins when brand voice consistency matters more than personalization.
How is a voice broadcast different from a voice campaign in Orbit?
Same execution backbone, different authoring surface. A broadcast is the specialized wizard for "one message to a list" — audio or TTS, list or segment, schedule, optional A/B — with quiet-hours and frequency-cap enforcement at execution time. The general campaign wizard is the multi-step surface for channel mixes, journeys, and cadenced sends. If the message genuinely is "record one thing and call the list," the broadcast wizard is fewer clicks and fewer ways to misconfigure.
Can I override the A/B winner after it is declared?
Yes. The campaign page shows live per-script results and lets you override the winner selection at any time — useful when a script wins on answer rate but you know the reply quality differs. Orbit picks winners on the metric you can measure, not on the metric you might prefer; the override is the audit trail that says a human reviewed the engine's choice.
Orbit's broadcast wizard, attestation posture tooling, and tenant-owned compliance controls all live in one account — the pricing page covers the voice plan that fits your list size, and the dialer and AI-agent surfaces cover the traffic a broadcast shouldn't carry.