12 min read · Compliance pillar for Australian outbound · Last updated 22 May 2026
The Australia do not call register AI voice agent telemarketing law gap is now your director\'s personal exposure. If your agent does not hear the caller say "stop calling" and act on it inside the same call, the next call goes out anyway. Then the complaint goes to ACMA. Then the infringement notice arrives at your office, not the dialler vendor\'s.
Most outbound stacks treat the register as a one off scrub at the start of a campaign. Upload list, wash against the federal register, dial. Anything the caller says mid conversation drops on the floor.
The number stays in rotation. Next week it gets called again. ACMA does not care that you washed the list in March if you ring the same person in May after they asked you to stop.
Waboom AI handles this differently. Every conversation is listened to in real time for 24 opt out trigger phrases. Hit one and the number flags on the client\'s own DNC list automatically.
No human in the loop. No "we will action it in the next sync". The next dialler queue checks that list before it picks up the phone.
This piece is the operator walkthrough for brand safe Australian outbound in 2026. The law that drives it. The detection that catches it. The questions to ask any provider before you sign a contract that puts your director\'s name on the dock.
In this article
- 1. Why agentic detection beats a list scrub
- 2. The 24 phrases the AI voice agent listens for on every call
- 3. What happens the second a phrase is heard
- 4. CSV upload wash: catching the breach before the dial
- 5. Per client DNC isolation and why it matters for brand risk
- 6. Webhooks: keeping HubSpot, Pipedrive and Salesforce in step
- 7. What this replaces in a normal outbound stack
- 8. Five questions to ask any provider before signing
- 9. Frequently Asked Questions
The Why
Why does agentic detection matter more than a list scrub?
Because a list scrub looks backwards. It checks if the number was on the AU DNC Register the morning the campaign launched. It does not catch what the caller said two minutes into yesterday's call.
ACMA enforces the Do Not Call Register Act 2006 and the Telecommunications (Telemarketing and Research Calls) Industry Standard 2017. Two duties matter to outbound operators.
First, do not call a number on the federal register. Second, honour any opt out the recipient makes during a call, before the next dial.
The second duty is where most stacks fail. A telemarketer in Manila reads the script. The caller in Footscray says "take me off your list". The telemarketer types a note in the CRM. The note never syncs back to the dialler.
Three days later the same caller picks up to the same voice. That is a breach. ACMA can issue an infringement notice for it without going near a courtroom.
The penalty surface is brutal. AUD 222,000 per day for infringement notices. AUD 2.22 million per day in the Federal Court for body corporate breaches [ACMA case summary]. The director personally hook is now a precedent thanks to V Marketing Australia.
AUD 1.5M
V Marketing Australia, fined March 2025 by the Federal Court. Plus AUD 60,000 against the sole director personally. 1.1 million calls to numbers on the Do Not Call Register across two campaigns.
The agentic answer is to make the second duty automatic. Every conversation gets listened to. Every trigger phrase flags the number. The list updates before the next dial.
That is what closes the gap between "we wash the AU DNC Register" and "we honour the opt out you gave us 90 seconds ago".
If you are still building this with overseas telemarketers and a CRM note field, you are taking the V Marketing risk every day you keep dialling. DIY voice stacks miss the same gap for the same reason: the detection is not engineered, it is hoped for.
The Detection
Which phrases trigger the Australia do not call register AI voice agent telemarketing law flag?
24 phrases, in 5 intent clusters, scored in real time against the live transcript of every Australian conversation. Hit any one with intent and the number lands on the client's DNC list before the call ends.
The 24 phrases were tuned from thousands of hours of real AU outbound recordings. Then checked against ACMA\'s published guidance on what counts as a clear opt out signal. Here is the working set the engine evaluates on every call:
| Intent cluster | Phrases the agent flags on |
|---|---|
| Direct stop | "stop calling", "do not call", "don't call", "never call", "no more calls" |
| Removal request | "remove me", "take me off", "remove my number", "delete my number" |
| Opt out language | "unsubscribe", "opt out", "leave me alone", "don\'t contact" |
| Harassment signal | "harassment", "harassing me", "stop contacting", "block my number" |
| Reporting threat | "reporting you", "report you", plus the variants of the above |
The detection runs against the live transcript, not on a single keyword match. Context matters.
"Don't call me back today, I'm in a meeting" is not a DNC signal. "Don't call me, full stop" is. The engine reads the surrounding turns of conversation before it commits to flagging.
False positives matter. If the agent flags a curious sceptic who used the word "harassment" jokingly, you lose a warm lead. So the rule is simple: detection errs on the operator protective side.
Borderline cases get logged with the transcript snippet visible inside the portal. You get the audit trail either way.
This is the layer that separates a brand safe agent from a high throughput dialler. The Sydney boutique agent campaign that produced 141 warm transferred vendor leads in 90 days at AUD 32.74 per lead ran with this detection on from day one.
1,997 real conversations across the run. The DNC flags accumulated quietly in the background while the warm leads stacked up.
The Action
What happens the second a phrase is heard?
Three things, in roughly this order, with no human asked to intervene. The agent closes the conversation respectfully. The number lands on the client's DNC list. A webhook fires to any CRM the client has wired up. All inside the same call.
Step one. The conversation closes politely. The agent acknowledges the request, confirms the number will not be called again, and ends the call inside 8 seconds.
No hard sell. No "are you sure". No transfer attempt. The recipient gets the experience the law expects them to get.
Step two. The phone number is written to the client\'s DNC table inside the Waboom AI portal. Timestamped. Tagged with the phrase that triggered it. Linked back to the call ID, the transcript and the recording reference.
If ACMA ever asks "when did you action this opt-out", the answer is in the portal in two clicks.
Step three. A dnc_detected webhook fires to any subscribed endpoint the client has configured. HubSpot, Pipedrive, Salesforce, an internal data warehouse, a Slack channel. The CRM updates the contact's communication preferences before your sales team even hears about it.
The whole sequence runs in roughly two seconds of agent runtime. That is the difference between an opt-out the law accepts and one ACMA can build a case around.
What you do not get: a queue of opt-out requests piled up in a CSV waiting for a Monday morning sync. A telemarketer in Manila typing a CRM note that never syncs back to the dialler. A 72-hour window where the same number gets called again because the system did not know.
You get one rule. The agent heard it. The agent acted on it. The audit trail is live before the call hangs up.
The Pre-flight
How does the CSV wash catch the breach before the dial?
Every contact list uploaded to the Waboom AI portal is washed against the client\'s existing DNC list before it can be queued. Flagged matches surface in the upload UI before any call goes out, showing the date and trigger phrase that put each number on the list.
This catches the most common breach pattern in Australian outbound. Same number, new campaign, new list, called again because no one checked.
The dormant database reactivation lists are the worst offenders. A list of 50,000 contacts from a five-year-old CRM export will always contain people who opted out at some point. If the dialler does not know that, you ring them, they complain, the file lands at ACMA.
The wash runs in seconds. A 50,000-row upload returns the match set before the operator can finish a coffee. Three outputs:
The audit trail for every wash is retained. If ACMA ever asks how a given number was excluded from a March campaign, the wash log shows the timestamp, the source list, and the rule that flagged it.
2 gates
The upload wash catches breaches before the dial. The agentic in call detection catches them mid conversation. Both layers run on every Australian campaign before any number sees a queue.
For a HubSpot or Salesforce export, the upload wash is the first gate. The agentic detection during calls is the second. Together they cover the two main entry points where a non-compliant number can reach a live dial.
Want this live on your next campaign?
Live same day for a focused build. Days for a standard build with one CRM integration. Two to three weeks max for multi-path orchestration.
The Isolation
Why is per-client DNC isolation the bit that protects your brand?
Because your DNC list is not a community list. If a homeowner in Marrickville tells your real estate agent's voice campaign to never call again, that is your record.
It should not bleed into a separate Brisbane agency's dial queue. And a different brand's opt-out should not silently exclude your prospects.
Every Waboom AI tenant gets a scoped DNC list. The detection writes to that tenant's table. The wash reads from that tenant's table. Cross-tenant queries do not run.
A solar firm in Footscray and a mortgage broker in Melbourne CBD share the same engine, but their DNC lists never see each other.
This matters for two reasons. First, brand specific compliance: if a recipient opts out from one brand they have not opted out from every advertiser on the platform. Second, brand specific risk: ACMA enforcement names the brand on the call, not the platform underneath.
If your tenant\'s data is clean and your campaigns honour your tenant\'s list, you control your own enforcement exposure.
The federal register is still washed at upload as the floor. The per client list sits on top, capturing every recipient who opted out specifically from a campaign run under your brand. Both layers are checked before a number is queued.
The Integration
How do webhooks keep HubSpot, Pipedrive and Salesforce in step?
Every DNC detection fires a dnc_detected webhook event to any endpoint the client has configured. The payload includes the phone number, the trigger phrase, the timestamp, the call ID and the transcript snippet.
CRMs can subscribe, update the contact's preferences, and lock the record from future outbound sequences.
The common wiring patterns:
If you do not want to write webhook code, the Waboom AI portal exposes the same data through a daily CSV export by tenant. Your operations team can drop it into any system that takes a flat file.
The point of the webhook layer: compliance does not stop at the call. The opt out the agent heard at 2:47pm on a Tuesday is visible in your CRM at 2:47pm on the same Tuesday.
The next email send, the next SMS broadcast, the next reactivation campaign respects it. Without anyone in your team needing to remember.
The Replacement
What does this replace in a normal outbound stack?
Four manual processes that most outbound teams either do badly or skip entirely. Each one is a real-world cause of an ACMA infringement notice.
The quarterly DNC scrub. A nominated ops person runs every contact list through the federal register file once a quarter. Anything new opts in between scrubs gets called anyway. Waboom AI washes every upload in seconds and detects in real time on every call. The scrub window collapses to zero.
The CRM opt out note. A telemarketer types "DNC" into a free text field. The dialler does not parse free text fields. The contact stays in rotation. Waboom AI writes the flag into the structured DNC table that the dialler queries on every queue check.
The Monday morning sync. An ops person exports last week\'s call notes, filters for opt out keywords, and bulk updates the CRM. Anything called between Friday lunchtime and Monday morning still slipped through. Waboom AI updates the CRM via webhook inside the same call.
The retrospective audit. ACMA opens a complaint. The brand asks the BPO for proof the opt out was honoured. The BPO sends a spreadsheet with mismatched dates and missing call IDs. Waboom AI keeps the full audit chain in the portal for the life of the tenant. Call ID, transcript snippet, timestamp, webhook delivery confirmation.
Shorter version. The law expects an opt out to be honoured before the next call. The only way to do that at outbound volume is to engineer it into the agent, not bolt it onto a workflow.
Federal calling window rules sit alongside the DNC engine. 9am to 8pm weekdays. 9am to 5pm Saturdays. No Sundays for sales. Seven national public holidays blacked out.
The brand protection buyer\'s guide walks through every window and the cultural blackouts most providers ignore.
The Buyer's Checklist
Which five questions should you ask any provider before signing?
The compliance answers most providers give are vague on purpose. These five questions cut through.
If the answer is "we scrub the register at the start of every campaign" and nothing else, you are signing up for the V Marketing risk pattern.
For Waboom AI the answers: 24 phrases per tenant. Two seconds. Both federal and per tenant lists. Full audit chain inside the portal. Sydney for the portal, transcripts and audit logs.
2 seconds
From the moment a caller says "stop calling" to the next dial being blocked. Portal write, CRM webhook, audit log, all inside the same call window. Anything slower is the V Marketing risk pattern with a fresh coat of paint.
The voice runtime sits offshore (US) because that is where the model and the real time speech infrastructure runs at production latency. The metadata, the audit logs and the structured call data all stay in Sydney.
For more on the law underneath this, the Australian telemarketing law guide walks through every Act and Industry Standard rule. For the operational mechanics on the inbound side, the AI voice agents page covers the full platform.
Run a compliance walkthrough on your next campaign
If you are about to sign with a provider, or you have one running and you are not sure how it would hold up to an ACMA complaint, we will show you the audit chain end-to-end on a live test campaign. 20 minutes.
Frequently Asked Questions
Does the agent actually identify as AI on every call?
Yes. The opening line on every Waboom AI outbound conversation in Australia identifies the call as automated and names the brand running the campaign. Identification is the floor under any operator-friendly reading of the Industry Standard. We do not test the boundaries on this one. The cost of getting it wrong is the entire compliance posture.
What if a recipient uses an unusual phrase to opt out that is not in the 24?
The 24 phrases are the deterministic flag set. The agent also evaluates intent across the conversation using the language model itself. A recipient who says "look mate, I've had enough of this, I want off your list" will be flagged even if the exact phrase is not in the table.
The 24 give us the deterministic floor. The intent evaluation gives us the ceiling. Both write to the same DNC table.
How does this interact with the federal AU DNC Register?
The federal register is washed against every uploaded list at upload time, before any call queues. The per-client DNC list captures everything the recipient said specifically to a campaign run under your brand.
Both are checked on every dial. The federal register is the floor. The per-client list is the brand-specific layer on top.
What happens if a number ends up on the DNC list by mistake?
The flag is reversible inside the portal. An operator can review the trigger transcript, decide it was a false positive, and remove the number from the DNC table.
Every reversal is logged with the operator's user ID and the reason. The audit chain shows the flag, the reversal, the rationale and the user. ACMA's bar is that you act in good faith and keep a record.
Does this work on inbound calls too, or just outbound?
Both. Inbound conversations are listened to with the same detection layer, and any opt-out language from an inbound caller flags the number for outbound exclusion. If a recipient calls in to ask a question and tells the agent never to ring them, that lands in the same DNC table.
What about the calling window rules and public holiday bans?
The calling window engine evaluates each dial at the recipient\'s local time, not the dialler\'s. 9am to 8pm weekdays. 9am to 5pm Saturdays. No Sundays for sales.
The seven national public holidays plus any substituted weekday are blocked at the dialler level. State-level holidays do not auto-block but are flagged in the campaign config for review. The brand-protection guide covers the full list.
Is the Sydney 90-day campaign data really comparable to a fresh build?
Yes. The campaign produced 1,997 real conversations and 141 warm-transferred vendor leads at $32.74 per lead. The detection layer was on from day one.
The numbers in the Sydney case study are the actual portal numbers from a Waboom AI client running on the production stack. If your campaign has comparable list quality and target market, the funnel shape transfers.
Are you covered if something goes wrong?
Waboom AI Ltd holds NZD 1,000,000 of business insurance covering professional indemnity, cyber liability and public liability. Certificate of currency on request.
The agentic detection layer, the CSV wash and the per-client DNC isolation are engineered to keep you well inside ACMA's expectations. The insurance backs the operational stack underneath.
Leonardo Garcia-Curtis
Founder & CEO at Waboom AI. Building voice AI agents that convert.
Ready to Build Your AI Voice Agent?
Let's discuss how Waboom AI can help automate your customer conversations.
Book a Free Demo


