Quick answer: the save happens before the cancel page, not on it
By the time a customer opens your cancellation page, you are negotiating, not saving. A discount offered at that moment usually buys a delay, not a renewal. The saveable window opens weeks earlier and is visible in behaviour: usage decay, seat reductions, a quiet champion, an unresolved ticket, a failed payment nobody chased. From 1 July 2027 Australian law puts a floor under the other end of this: new section 48F(1) of the Australian Consumer Law requires a way to end a subscription that is “easy to find” and “straightforward”, and it reaches small business subscribers as well as consumers.
The line that governs the whole design: phoning a customer whose usage is fading is ordinary customer contact. Phoning a customer who has already asked to cancel, in order to talk them out of it, is not — that is a retention obstacle, however politely it is scripted. A churn-save agent works the first window and stops dead at the cancel button. Everything below assumes that rule is enforced in code, not in a prompt.
Why the cancellation page is the wrong place to fight
Someone at the cancel button has usually justified the decision to a manager, checked the contract, and often chosen a replacement. A discount reverses none of that. It buys one more billing period from a person whose decision is made, and teaches your base that asking to leave is how you get a better rate. It does nothing about the cause: if value stopped in March, an offer in June does not restore it.
The window where a conversation changes the outcome is earlier, while someone is still deciding whether the product is worth persisting with. In that window they will tell you what is wrong, because they have not yet committed to a story about why they left.
The signal set: what fading actually looks like
Churn risk is not one number. The shape of the decay matters more than its size, because different shapes need different conversations.
| Signal | What it usually means | What the agent must find out |
|---|---|---|
| Usage cliff (days) | A person left, or a project ended. Not dissatisfaction | Who owns this now, is the use case live |
| Slow fade (weeks) | Value erosion. It works, it stopped being worth it | What changed, and what replaced you |
| Seat reduction | Budget scrutiny has started above your champion | Who is asking, and what number |
| Champion goes quiet | They have left, or lost the internal argument | Still in role? Who else decides? |
| Ticket closed unfixed | An unpaid grievance. These wait for renewal | What broke, is it still broken |
| Failed payment, unchased | Not loyalty. Operations | Nothing. Fix the payment |
| New admin login | Ownership changed. Equity reset to zero | Who they are, what they are measured on |
| Usage narrows to one job | You are a point tool now, easy to cut | Did the other jobs move, or stop |
The first two look identical in a dashboard and mean opposite things: both read as “usage down 80 per cent”. Calling both with one script is how retention outreach earns its reputation for tone-deafness.
Why a call beats an email for this job
Retention email works when the action is mechanical: update a card, reactivate a seat, book training. It fails at diagnosis, because the reason a customer leaves is usually not the reason they give. Exit surveys collect the socially efficient answer, because that answer ends the interaction. Budget and not the right fit right now are polite exits. Underneath sit things people say aloud but will not type: an implementation that never finished, a feature promised in a sales cycle that never arrived, a bad support experience nobody escalated.
A conversation gets past that because it can ask one more question. When you say budget, was that across the board or was ours the one that got picked? You can ask a person that; you cannot put it in a form. Voice also closes loops live: if the fix is a solutions engineer, the agent books it there. See handing context from an AI agent to a human for what travels with that handoff.
Conversation design: diagnosis first, offers late or never
The biggest design error is leading with the offer. It becomes a negotiation before you know what you are negotiating about, and hands the customer a reason to churn they did not previously have. A good save agent opens with an observation and a question, then listens.
Authorised
- Diagnose and record the reason in the customer’s own words, tagged to a cause, in the CRM.
- Book a human for anything technical, contractual or emotionally warm.
- Escalate on the call: unresolved tickets, security concerns, complaints.
- Log a product gap when the cause is a missing capability, not a usage problem.
- Fix mechanics: payment-update link, seat reactivation, onboarding resend.
Never
- Call anyone who has asked to cancel. A bright line, not a judgement call. The cancellation proceeds.
- Invent commercial terms. No discounts, credits or contract variations. The agent has no authority to offer a concession of any size, and we do not publish suggested numbers for anyone else’s business either. A generative system will happily produce a plausible concession nobody authorised.
- Argue. Rebutting the stated reason turns diagnosis into a debate the customer wins by hanging up.
- Guilt. No sunk-cost framing.
- Delay or obstruct. Nothing in the flow may make cancelling slower, harder or conditional on a conversation.
- Repeat. One attempt, one follow-up channel. Persistence reads as pressure.
Those belong in enforced policy, not the prompt. See writing guardrails for autonomous AI agents.
Involuntary churn: an operations problem in a loyalty costume
A meaningful slice of what shows up as cancellation is not a decision at all. Cards expire, issuers decline, fraud rules produce false positives, and a customer who wanted to stay is cut off.
Paddle, which processes subscription payments and publishes its own churn analysis, wrote in a post dated 1 September 2023 that typically 20-40% of churn is involuntary (particularly if you take payments through cards). Paddle illustrates that range with a single sample seller’s twelve-month breakdown, so treat it as one processor’s observation rather than an industry constant. The point holds at any share: part of your churn number is a dunning problem, and the cheapest kind to fix because there is no objection to overcome.
Keep them in a separate queue. Running a retention script at someone whose card expired is insulting, and the call only needs thirty seconds: your payment did not go through, here is the link.
One hard rule sits on top of the dunning logic. Retries follow the customer’s instruction. If someone has cancelled, disputed the charge or told you to stop charging them, recovery stops with them — no further retry attempts, no card updater, no “one last try” before the period ends. Recovering a payment from a customer who has asked to leave is not dunning, it is charging someone who withdrew their authority. Involuntary-churn work is only defensible while the customer’s intention is to keep paying.
The line: when a save call becomes a retention trap
Australia: legislated, with a known start date, and it reaches small business
The Competition and Consumer Amendment (Unfair Trading Practices) Bill 2026 finally passed both Houses on 2 July 2026 and received assent on 6 July 2026 as Act No. 64 of 2026. Per the Parliament of Australia bill page, it amends the Competition and Consumer Act 2010 to “require a person offering goods or services under a subscription contract to provide subscribers with certain pre-contract and ongoing information and an easy and straightforward way to end a subscription, including an online cancellation option in certain circumstances.”
The authorised text on the Federal Register of Legislation carries the notation “[Assented to 6 July 2026]”, and the commencement table in section 2 sets one date for the whole Act: 1 July 2027. The start date is in the Act. You do not need a media release for it.
Two different coverage tests sit inside this Act, and the difference decides whether it applies to you. The general prohibition, new section 28B of the Australian Consumer Law, is headed “Unfair trading practices towards consumers” and is consumer-facing in a drafted, narrow sense. Subsection 28B(3): “Subsection (1) does not apply if the consumer is a body corporate.” Subsection 28B(4): “Subsection (1) does not apply if the supply is in the course of the consumer carrying on a business.” A supplier who wants to rely on either carries an evidential burden under subsection 28B(5).
The subscription rules are drawn wider, and this is the part most commentary gets wrong. They sit in a new Division 4A of Part 3-1 of the ACL. Section 48E(1)(b) applies where the contract “meets the consumer requirement or meets the small business requirement”. Section 48F(1), the exit-method duty, applies to a person who supplies “under a subscription contract that meets the consumer requirement or meets the small business requirement”. A business-to-business SaaS subscription can therefore sit outside section 28B and squarely inside the cancellation obligations. If you sell subscriptions to Australian small businesses, Division 4A is your problem, not somebody else’s.
Section 48G defines both tests, so nobody has to guess a threshold. A contract “meets the consumer requirement” if it is one “under which an individual acquires the goods or services wholly or predominantly for personal, domestic or household use or consumption” (s 48G(1)). It “meets the small business requirement” if two things hold (s 48G(2)): the contract “is a standard form contract for supply of goods or services”, and the subscriber satisfies either or both of “the subscriber makes the contract in the course of carrying on a business and at a time when the subscriber employs fewer than 100 persons” and a turnover test, that the subscriber’s turnover for its last income year “is less than $10,000,000” — A$10 million, worked out under s 48G(4) as the sum of the values of all supplies the subscriber made in the period, less input-taxed supplies and several other listed categories. Under s 48G(3) a casual employee “is not to be counted unless employed on a regular and systematic basis” and a part-time employee counts as “an appropriate fraction of a full-time equivalent”. And the standard-form limb is not a shield: under s 48H(1), where a party alleges a contract is a standard form contract it “is presumed to be a standard form contract unless another party to the proceeding proves otherwise”.
What section 48F(1) actually requires is worth reading against your own cancel flow. The supplier must “provide a way for the subscriber to end the contract” and ensure each way it provides “is easy to find”, “is straightforward”, and “requires the subscriber to take only steps that are reasonably necessary to end the contract and protect the subscriber’s interests”. Where the subscriber entered online, or the supplier offers an online way of entering that kind of subscription, one of the exit routes must be online (s 48F(1)(c) and (2)). A save call that a subscriber has to get through before the cancellation completes is the shape of thing this drafting is aimed at.
Timing matters as much as scope. Section 312 of the ACL, inserted by the same Act, provides that sections 48B, 48E and 48F “apply in relation to a contract entered into on or after the commencement of Part 3 of Schedule 1 to that Act” — 1 July 2027 — and do not apply to a contract entered into before then. But if such a contract “is renewed, extended or otherwise continued” on or after that date, or is varied on or after that date, those sections apply to it from the day the renewal, extension, continuation or variation takes effect. Your existing base rolls into scope at its next renewal, not never.
Penalties are not nominal. Division 4A sits inside Part 3-1, and sections 48D(2), 48E(1) and 48F(1) each carry the note “A pecuniary penalty may be imposed for a contravention of this subsection.” Item 2 of the table in ACL section 224(3) caps a Part 3-1 contravention at “the greater of the amounts mentioned in subsection (3A)” for a body corporate and A$2,500,000 for a person who is not one. Since the Treasury Laws Amendment (Doubling Penalties for ACCC Enforcement) Act 2026 (Act No. 19 of 2026) commenced on 28 March 2026, those subsection (3A) amounts are A$100,000,000, three times the value of the benefit reasonably attributable to the act or omission, or, where the court cannot determine that value, 30 per cent of the body corporate’s adjusted turnover during the breach turnover period. Item 6 of that Act’s Schedule 1 replaced “$50,000,000” with “$100,000,000” in ACL paragraph 224(3A)(a), and ACL section 310 applies the new figure to “offences committed, or acts or omissions that occurred, on or after” commencement. Advice still quoting an A$50 million ACL maximum is describing the law as it was before 28 March 2026. The ACCC’s fines and penalties page states the same three-limb maximum and the same A$2,500,000 figure for individuals.
Assistant Minister Andrew Leigh’s media release of 2 July 2026 states that “unfair trading practices and subscription traps will be banned from 1 July 2027” and that “Consultation is already underway on extending protections to small businesses and franchisees, with submissions open until 10 July”. Read that alongside the Act, not instead of it. The provision confined to consumers is section 28B, with the 28B(3) and 28B(4) exclusions. Division 4A’s subscription and cancellation duties already reach small business subscribers as section 48G(2) defines them. We could not verify the outcome of that consultation, and it does not narrow what Division 4A already says.
Announcing the ACCC’s 2026-27 priorities on 19 February 2026, Chair Gina Cass-Gottlieb said: “This priority recognises the emergence of practices including subscription traps and other dark patterns that manipulate consumer behaviour and unfairly impact consumer choice.”
Sell to Australian small businesses on standard terms and the unfair contract terms regime already bites. The ACCC’s guidance on contracts records that changes to the law took effect on 9 November 2023 and that from that date “proposing, using or relying on unfair contract terms in standard form contracts will be banned and penalties for breaches of the law will apply”. Read that page for what it is: general unfair-contract-terms guidance. It does not mention subscriptions, auto-renewal or cancellation flows, and we are not going to pretend it does. Treating auto-renewal and termination clauses as review candidates under that regime is our inference, not an ACCC instruction, and it is a question for your own lawyers.
United States: the click-to-cancel rule is gone, and its replacement is unsettled
Much published advice still describes a rule that has not existed for over a year. The FTC published an amended Negative Option Rule at 16 CFR part 425 — formally the Rule Concerning Recurring Subscriptions and Other Negative Option Programs, known everywhere as click-to-cancel — in the Federal Register on 15 November 2024. On 8 July 2025, in Custom Communications, Inc. v. Federal Trade Commission, the Eighth Circuit set it aside, holding that “the procedural deficiencies of the Commission’s rulemaking process are fatal here” and concluding: “Accordingly, we grant the petitions for review and vacate the Rule.” The court decided on procedure. It expressly did not endorse the conduct the rule targeted.
The FTC then conformed its books to the judgment. A final rule effective 12 February 2026 revised part 425 “to recodify the text of the Negative Option Rule as it existed before the effective date of the Commission’s 2024 final rule amending it”, restoring its older name, Use of Prenotification Negative Option Plans. So there is a Negative Option Rule in force in the United States — the pre-2024 one, aimed at prenotification plans. What is not in force, and has not been since July 2025, is anything resembling click-to-cancel.
What replaces it is open. On 11 March 2026 the FTC announced an Advance Notice of Proposed Rulemaking seeking comment on ways to address unfair or deceptive negative option practices, “including retaining the current Rule, adopting provisions of the vacated 2024 Rule or some other provisions, or implementing alternatives to regulation”. Christopher Mufarrige, Director of the FTC’s Bureau of Consumer Protection, said in the announcement: “The Commission’s enforcement track record suggests, however, that negative option subscriptions continue to be plagued by difficult cancellation processes, unlawful retention tactics, and a suite of other impediments that prevent consumers from easily switching or ending subscription services.” As at 4 September 2026 that advance notice is the most recent rulemaking action listed on the FTC’s Negative Option Rule page, and no successor rule has been proposed, let alone finalised. The federal position is genuinely unsettled, and anyone telling you which way it lands is guessing.
The statute underneath it is not unsettled, and it is the thing to design against. The Restore Online Shoppers’ Confidence Act, at 15 U.S.C. § 8403, makes it unlawful to charge a consumer for goods or services “sold in a transaction effected on the Internet through a negative option feature” unless the seller, among other requirements, “provides simple mechanisms for a consumer to stop recurring charges from being placed on the consumer’s credit card, debit card, bank account, or other financial account”. Nothing in the rulemaking touched that duty, and state automatic-renewal laws are likewise unaffected by it.
The bright line for your agent
- Fading usage, no cancellation request: a save call is legitimate customer contact. Normal consent and outbound-calling rules apply, but the act itself is unremarkable — it is the same category as a check-in from an account manager.
- Cancellation requested: the cancellation proceeds, at the same speed it would have without the agent. An agent that phones someone who has asked to leave, to talk them out of it, is a retention obstacle however politely scripted. That is the conduct both regulators have named: “subscription traps and other dark patterns” in the ACCC’s words, “difficult cancellation processes, unlawful retention tactics” in the FTC’s.
Post-cancellation contact is defensible only in documented cases: off-boarding, a billing dispute, or a win-back long after the fact under whatever consent basis you actually hold. Put the flag in agent eligibility, not campaign config, so no campaign can switch it off.
Measurement: save rate is the vanity number
| Metric | What it tells you | How it gets gamed |
|---|---|---|
| Save rate | Little. Counts anyone who did not cancel today | Concede everything, count the deferral |
| Saved and still active at 90 days | Whether the save was real. The honest one | Hard to game. That is the point |
| Same, excluding saves that involved a concession | Whether you fixed it or bought time | Tag concessions loosely |
| Product causes surfaced and actioned | Whether the agent diagnoses or absorbs pressure | Log every cause, action none |
| Involuntary recovery rate | Dunning effectiveness, kept out of retention | Blend it into save rate |
The most valuable output is usually the reason data, not the saves. Log it as you would any other agent output; see AI agent observability.
Where Zian fits
Zian AI builds autonomous agents for live phone calls, SMS, email and WhatsApp across 30+ languages, with CRM integration. SmartReach AI orchestrates message, channel and timing, which is where the difference between a check-in and a nuisance lives. PrecisionPitch AI split-tests scripts against real outcomes, which is how you learn whether diagnosis-first openings beat offer-first ones. The multi-channel design behind reactivating dormant CRM records applies here, with stricter eligibility rules — and eligibility is the part that matters, because the cancellation flag has to sit where no campaign can override it.
Zian is in partnership-application beta. No public pricing, no free trial, no self-serve signup. If you want to build a save program on it, you apply and we talk.
How we sourced this
Every legal and statistical claim above was checked by opening the primary source on the date shown against it. The Australian statutory text was read in the authorised version on the Federal Register of Legislation on 4 September 2026.
| Claim or figure | Owner (organisation) | Source | Date checked |
|---|---|---|---|
| Subscription obligations; passage 2 July 2026, assent 6 July 2026 as Act No. 64 of 2026 | Parliament of Australia | Bill page, Competition and Consumer Amendment (Unfair Trading Practices) Bill 2026 | 4 Sep 2026 |
| Commencement 1 July 2027 (s 2 table); s 28B(1)–(6), incl. the 28B(3) body-corporate and 28B(4) business-purpose exclusions; Division 4A ss 48B–48H; the s 48E(1)(b) and s 48F(1) “meets the consumer requirement or meets the small business requirement” test; the s 48G(1) and 48G(2) definitions; s 48F(1)(b)–(c) exit-method wording; s 312 application to contracts renewed, continued or varied after commencement | Commonwealth of Australia (Federal Register of Legislation) | Competition and Consumer Amendment (Unfair Trading Practices) Act 2026, authorised text | 4 Sep 2026 |
| ACL s 224(3) table item 2 (Part 3-1) and s 224(3A): body corporate maximum is the greater of A$100,000,000, 3× the attributable benefit, or 30% of adjusted turnover in the breach turnover period; A$2,500,000 for a person who is not a body corporate | Commonwealth of Australia (Federal Register of Legislation) | Competition and Consumer Act 2010, Schedule 2, compilation in force 1 July 2026 | 4 Sep 2026 |
| A$50,000,000 replaced by A$100,000,000 in ACL para 224(3A)(a) (Sch 1 item 6); commenced 28 March 2026; ACL s 310 applies it to acts or omissions on or after that date | Commonwealth of Australia (Federal Register of Legislation) | Treasury Laws Amendment (Doubling Penalties for ACCC Enforcement) Act 2026 (No. 19, 2026) | 4 Sep 2026 |
| Same three-limb body-corporate maximum and the A$2,500,000 individual maximum, stated by the regulator | ACCC | Fines and penalties | 4 Sep 2026 |
| “unfair trading practices and subscription traps will be banned from 1 July 2027”; “Consultation is already underway on extending protections to small businesses and franchisees, with submissions open until 10 July” | Assistant Minister Andrew Leigh (Treasury portfolio) | Media release, 2 July 2026 | 4 Sep 2026 |
| “subscription traps and other dark patterns” as an enforcement priority | ACCC (Chair Gina Cass-Gottlieb) | Media release, 19 February 2026 | 28 Aug 2026 |
| “Changes to the law on unfair contract terms came into effect on 9 November 2023”; no mention of subscriptions, auto-renewal or cancellation on that page | ACCC | Contracts guidance | 4 Sep 2026 |
| 2024 click-to-cancel rule vacated on procedural grounds, 8 July 2025 | US Court of Appeals for the Eighth Circuit | Opinion, Custom Communications, Inc. v. FTC, No. 24-3137 | 28 Aug 2026 |
| Pre-2024 rule text recodified, effective 12 February 2026 | US Federal Trade Commission | Final rule notice | 28 Aug 2026 |
| Advance notice of proposed rulemaking; Mufarrige statement; options under consideration | US Federal Trade Commission | Announcement, 11 March 2026 and the Negative Option Rule page | Announcement 28 Aug 2026; rule page re-checked 4 Sep 2026 |
| Duty to provide simple cancellation mechanisms (ROSCA) | United States Code | 15 U.S.C. § 8403 | 28 Aug 2026 |
| “typically 20-40% of churn is involuntary” | Paddle (post dated 1 September 2023) | Reduce churn | 28 Aug 2026 |
Not claimed, because we could not verify it: any industry-wide save-call success rate, any benchmark for a good 90-day retained-save rate, and the outcome of the Treasury consultation on extending protections to small businesses and franchisees. That consultation goes to the general prohibition in section 28B, which is the provision confined to consumers; Division 4A’s subscription and cancellation duties already reach small business subscribers as section 48G(2) defines them. We also do not quote a penalty figure specific to Division 4A, because the Act does not create one — those contraventions fall under the general Part 3-1 maximum in ACL s 224(3). Zian AI publishes no churn-save figures, and nothing above is legal advice.
Frequently asked questions
Should a churn-save agent call before or after someone clicks cancel?
Before, and only before. Once cancellation is requested, a call to reverse it is a retention obstacle, not customer service, and the cancellation must proceed at the speed it otherwise would. Australia’s Competition and Consumer Amendment (Unfair Trading Practices) Act 2026 inserts section 48F(1) of the Australian Consumer Law, which requires a supplier to provide a way for the subscriber to end the contract and to ensure each way it provides “is easy to find”, “is straightforward”, and “requires the subscriber to take only steps that are reasonably necessary to end the contract and protect the subscriber’s interests”. Section 2 of that Act commences the whole Act on 1 July 2027.
Does the Australian subscription law apply to business customers?
Partly, and the split is the whole point. The general unfair trading prohibition, ACL section 28B, does not apply where “the consumer is a body corporate” (s 28B(3)) or where “the supply is in the course of the consumer carrying on a business” (s 28B(4)). The subscription rules in Division 4A are drafted wider: sections 48E(1)(b) and 48F(1) each apply where the contract “meets the consumer requirement or meets the small business requirement”. Under section 48G(2) a contract meets the small business requirement if it is a standard form contract and the subscriber either “employs fewer than 100 persons” when the contract is made or had turnover “less than $10,000,000” (A$10 million) in its last income year. A business-to-business subscription can sit outside section 28B and inside the cancellation duty.
Is the US click-to-cancel rule in force?
No. The Eighth Circuit vacated it on 8 July 2025 on procedural grounds in Custom Communications, Inc. v. FTC, and an FTC final rule effective 12 February 2026 recodified the pre-2024 rule text, so the Negative Option Rule now in force is the older prenotification rule, not click-to-cancel. As at 4 September 2026 the most recent rulemaking action on the FTC’s Negative Option Rule page is an advance notice of proposed rulemaking announced in March 2026; no successor rule has been proposed. The position is unsettled. The separate duty under 15 U.S.C. § 8403 to provide simple mechanisms to stop recurring charges is unaffected, as are state automatic-renewal laws.
What share of churn is actually failed payments?
More than most teams assume. Paddle wrote in a post dated 1 September 2023 that typically 20-40% of churn is involuntary (particularly if you take payments through cards), illustrating it with one sample seller’s twelve-month breakdown. Treat it as one processor’s observation rather than a benchmark, but it is enough to justify separating dunning recovery from real retention in your reporting.
How do we know whether the save program is working?
Report saved-and-still-active at 90 days, split by whether a concession was given, plus the rate at which the agent surfaced a product cause somebody actually actioned. Raw save rate rewards concessions and hides deferred churn. Keep involuntary payment recovery in its own line.
Can we buy this from Zian today?
Not off the shelf. Zian AI is in partnership-application beta: no public pricing, no free trial, no self-serve signup. Deployments are scoped with our team, including private deployment on your own infrastructure.
Build the save conversation that happens before the cancel button
If your retention motion starts at the cancellation page, you are measuring negotiation and calling it loyalty. The agents worth building call earlier, take no for an answer, and hand product a list of causes somebody can act on.