Mostly no, but the exceptions are where outbound campaigns leak money. Twilio’s documentation treats a call answered by voicemail as a completed call, and “Completed calls, regardless of the outcome, are charged against your project balance.” Retell publishes a 10-second minimum charge on some calls, and Twilio lists answering machine detection at $0.0075 per call.
The awkward part is what is not published. Every AI voice pricing page we opened on 8 September 2026 — Retell AI, Bland, Vapi — prices connected minutes. None of them prices the dials that ring out, hit a busy tone, get spam-labelled by an Australian carrier, or die three seconds after pickup. On a normal outbound list that is the large majority of the campaign, and it is the part nobody quotes you a rate for.
Do I get charged for AI calls that nobody answers?
Split the question into five separate meters, because a single dial can bill on all five: the telephony leg, a detection fee, the platform’s per-minute charge, the LLM, and the text-to-speech voice. A call outcome switches each meter on or off independently, and the vendors do not agree on which switch flips when.
The matrix below is the answer. Every cell is either a behaviour the vendor publishes on its own page — linked — or is explicitly marked as something we could not locate. All pages were opened on 8 September 2026. Currency note: Twilio’s and Retell’s pricing pages publish a bare “$” with no currency code stated on the page, so read every rate below as unconverted.
Outcome to billing-event matrix
| Call outcome | Telephony leg (Twilio Programmable Voice) | Detection fee (Twilio AMD) | Platform minute (Retell AI) | LLM + TTS (Retell AI) | Bland / Vapi published position |
|---|---|---|---|---|---|
| No answer — rings out | Terminal status no-answer: “There was no answer or the call was rejected”. Twilio’s Call resource docs also state that duration “is empty for busy, failed, unanswered, or ongoing calls”. The docs state the positive case only — we could not locate an explicit statement on that page that a no-answer call carries no connectivity charge. |
Not located. We could not find a statement on Twilio’s Australian voice pricing page or in the Answering Machine Detection guide as to whether the $0.0075-per-call fee applies when nobody answers. | Not billed. Retell’s pricing FAQ: “You are only charged for connected calls. If a call fails to connect, you won’t be billed.” | Not billed — same statement covers the whole per-minute stack. | We could not locate any statement about unanswered calls on either pricing page (checked 8 September 2026). |
| Busy signal | Terminal status busy: “The caller received a busy signal”. Duration empty. Same documentation gap as no-answer. |
Not located — same gap as the no-answer row. | Not billed — “fails to connect”. | Not billed. | Not located on either pricing page. |
| Voicemail answers and the agent talks to it | Billed. Twilio: “A completed call indicates that a connection was established, and audio data was transferred. This can occur when a call is answered by a person, an IVR phone tree menu, or even a voicemail. Completed calls, regardless of the outcome, are charged against your project balance.” Australian mobile: $0.0750/min; Australian local: $0.0252/min. | $0.0075 per call, published. This is the outcome AMD exists to cut short. | Partly billed. Retell: “For calls that reach voicemail, billing applies only for the duration the AI agent is active on the line.” | Billed for that same duration. Retell also states billing continues through silence: “Billing covers the entire duration of the call because the speech-to-text engine remains active and listening throughout, even during silence.” | We could not locate the word “voicemail” on either pricing page. |
| Spam-labelled or carrier-rejected before answer | Terminal status failed: “The call could not be completed as dialed, most likely because the provided number was invalid” — Twilio’s own gloss points at an invalid number rather than at carrier spam blocking, so read this as the closest published status, not as a documented spam outcome. Not a completed call. Explicit no-charge statement not located on that page. |
Not located — same gap as the no-answer row. | Not billed — “fails to connect”. | Not billed. | Not located. Note the paid mitigations are per-call or per-month, not per-minute: Retell publishes “Branded Call +$0.10/outbound call” and “Verified Phone Number $10.00/Phone number/month”. |
| Answered, then hung up inside 10 seconds | Billed. The StatusCallback parameters in Twilio’s Call resource docs describe Duration as “The duration in minutes of the just-completed call; calls are billed by the minute.” | $0.0075 per call, published. | Billed at a floor. Retell’s billing-exceptions page scopes this to “Calls shorter than 10 seconds that use dynamic opening messages when AI speaks first”, which then carry a “Minimum charge of 10 seconds”. A 6-second call bills as 10 seconds. | Billed against the same 10-second floor. Retell separately scales billed duration for prompts over 4,000 tokens: “Billed Duration = Original Duration × Scaling Factor (rounded up)”. | We could not locate a published minimum billed duration, or a per-second versus per-minute rounding rule, on either pricing page. |
| Connected, then transferred to a human | Billed for the whole call. Twilio notes the call price “only reflects the charge for connectivity. Charges for other call-related features such as Answering Machine Detection, Text-To-Speech, and SIP REFER are not included in this value.” Twilio’s Australian page lists SIP interface refer at $0.1000/refer. | Not applicable after answer. | Stops. Retell: “Once the call is transferred, the AI voice agent fee stops. Only the telephony fee continues for the remainder of the transferred call.” | Stops at transfer, per the same statement. | Bland publishes a separate Transfer Time Rate — $0.05/min (Start), $0.04/min (Build), $0.03/min (Scale) — distinct from its Talk Time Rate of $0.14/$0.12/$0.11 per minute. We could not locate a transfer rate on Vapi’s pricing page. |
Read the matrix and one thing stands out: the two outcomes that dominate an outbound list — no answer and busy — are the two rows with the most “not published” cells. Vendors document what they charge for. They rarely document what they don’t.
Why “we could not locate it” is the honest answer, and why that matters
We are deliberately not writing that Bland or Vapi “publish nothing” about rounding. What we can say precisely is this: on Bland’s pricing page and Vapi’s pricing page, opened 8 September 2026, we could not locate the strings “per second”, “rounding”, “unanswered” or “voicemail” anywhere on the page. Bland publishes a Talk Time Rate of $0.14/min on Start and Vapi publishes hosting at “$0.05 / min” with “Model Provider Cost (STT, LLM, TTS)” charged “At cost”, and neither page states the granularity that turns those rates into an invoice. A rate without a rounding rule is not a price.
Retell is the counter-example, and it is worth crediting: its pricing FAQ states “Each call is tracked to the nearest second. At the end of each billing cycle, we charge based on your total accumulated minutes. No rounding up per call, no inflated bills.” That is a specific, checkable commitment. Twilio’s is equally specific in the other direction — “calls are billed by the minute” — which on a campaign of short calls is a materially different invoice for the same audio.
The clearest evidence that this information is hard to get is that people resort to instrumenting the API to find it. A February 2026 write-up published by Classmethod’s DevelopersIO took before-and-after snapshots of Twilio’s Usage Records API around test calls purely to establish whether unanswered ring time is billed. We are not citing that test as authority — every billing claim on this page comes from the vendor’s own documentation. We are citing the fact that the test had to be run at all. That is the state of pricing transparency in this category, and it is why we published the matrix above rather than a rate card.
What 1,000 dials actually cost per conversation, not per minute
Cost per minute is the wrong unit for outbound. You do not buy minutes; you buy conversations. Here is the arithmetic end to end, using published rates only, so you can substitute your own figures.
The connect rate below is an assumption you must replace with your own measured number. It is not a Zian measurement, not an industry benchmark, and not a claim about what any campaign achieves. Pull it from your own dialler logs: connected calls divided by dials attempted, over at least a full week.
Inputs
| Input | Value used | Where it comes from |
|---|---|---|
| Dials attempted | 1,000 | Assumption — substitute your own |
| Connect rate | 12% | Assumption — substitute your own measured rate |
| Average connected talk time | 1.8 minutes | Assumption — substitute your own |
| Telephony, outbound to Australian mobile | $0.0750/min | Published on Twilio’s Australian voice pricing page |
| Platform voice infrastructure | $0.055/min | Published: Retell Voice Infra |
| Text-to-speech | $0.015/min | Published: Retell Platform Voices |
| LLM | $0.040/min | Published: Retell GPT 5, standard tier |
| Answering machine detection | $0.0075/call | Published on Twilio’s Australian voice pricing page |
| Do Not Call Register washing | A$0.00585/number | Derived: A$585 ÷ 100,000 wash credits, subscription Type C |
Two caveats before the sum. First, adding Retell’s three published component rates gives $0.110/min; that is our arithmetic, not a bundle Retell publishes, though it sits inside the “$0.07-$0.31 / min” range Retell states for voice agents. Second, the register fee is in Australian dollars while the vendor rates are a bare “$” with no currency code on the page, so the totals below mix units. We have kept the register line itemised so you can restate it in your own currency.
The sum
- Connected calls: 1,000 × 12% = 120 conversations. 880 dials produced nothing.
- Connected minutes: 120 × 1.8 = 216 minutes.
- Telephony: 216 × $0.0750 = $16.20
- Platform + TTS + LLM: 216 × $0.110 = $23.76
- Detection, on answered calls: 120 × $0.0075 = $0.90
- Register washing, on every number: 1,000 × A$0.00585 = A$5.85
- Total: $46.71
Now convert that into the two units that matter. Cost per connected minute: $46.71 ÷ 216 = $0.216. Cost per conversation: $46.71 ÷ 120 = $0.389. If one conversation in twelve books a meeting — again, your number, not ours — that is 10 meetings and $4.67 per booked meeting.
Here is the result the arithmetic above produced, which we did not expect. On this modelled stack, the 880 dials that never became a conversation cost A$5.15 of the $46.71 total — about 11%. Everything else was spent on calls that connected. Wasted dials are cheap, because almost every meter in the stack is a per-minute meter and unanswered calls have no minutes.
The per-dial fees that flip the arithmetic
That conclusion survives exactly as long as your stack has no meaningful per-dial charge. Add one and the picture inverts. Retell publishes two add-ons priced per dial rather than per minute: “Batch Call +$0.005/dial” and “Branded Call +$0.10/outbound call”. Branded calling is the standard mitigation when carriers start spam-labelling your number, so it is the add-on a struggling outbound campaign is most likely to switch on. Whether that per-outbound-call fee applies to a dial that is never answered is not stated on Retell’s pricing page (checked 8 September 2026); the scenario below models it as the page describes it, per outbound call.
Same 1,000 dials, same talk time, same rates, plus batch calling and branded calling. Per-dial cost becomes A$0.00585 + $0.005 + $0.10 = $0.11085 per dial, or $110.85 across the campaign, before a single conversation happens.
Sensitivity: cost per conversation as connect rate moves
Variable cost per connected call is fixed at 1.8 × ($0.0750 + $0.110) + $0.0075 = $0.3405. So cost per conversation = $0.3405 + (total per-dial cost ÷ number of conversations). Substitute your own connect rate into the left column.
| Connect rate (your assumption) | Conversations from 1,000 dials | Scenario A: register wash only, no vendor per-dial fees | Scenario B: batch + branded calling |
|---|---|---|---|
| 4% | 40 | $0.487 | $3.112 |
| 8% | 80 | $0.414 | $1.726 |
| 12% | 120 | $0.389 | $1.264 |
| 20% | 200 | $0.370 | $0.895 |
| 30% | 300 | $0.360 | $0.710 |
| Spread, 4% to 30% | 1.35× | 4.4× | |
That spread is the actual insight on this page. With no vendor per-dial fee — the register wash at A$0.00585 a number is the only per-dial line in Scenario A — connect rate barely moves your cost per conversation; with a $0.10 per-outbound-call fee attached, connect rate becomes the single largest driver of it. In Scenario B at a 12% connect rate, the 880 dials that produced no conversation account for $97.55 of the $151.71 campaign total — 64% of the spend, on calls nobody had. In Scenario A the same dials account for 11%.
So the first question to ask a vendor is not “what is your per-minute rate”. It is “which of your charges are per dial, and do they apply when the dial does not connect”. Two vendors quoting the same per-minute figure can invoice you amounts that differ by a multiple, and the difference is decided entirely by list quality. The same logic separates AI phone agents from predictive diallers, where the dial-to-connect ratio has always been the number that decides the economics.
Apply for early access to Zian’s autonomous sales agents. Zian AI runs live phone, SMS, email and WhatsApp outreach through SmartReach AI™ and PrecisionPitch AI™. We are in a partnership-application beta and reviewing applications now. Apply For Partnership
The costs that land before a single number is dialled
The per-minute rate is the only line most buyers model. These are the lines that arrive whether or not anyone picks up.
- List washing against the Do Not Call Register. An Australian obligation with a published, per-number price. The ACMA’s subscription schedule runs from Type A (500 wash credits, free) to Type H (100,000,000 credits, A$142,271); Type C is 100,000 credits for A$585. Crucially, “In general you can rely on the washed list for 30 days” — so this is a recurring per-number cost on any list you re-dial across months, not a one-off. Our guide to Do Not Call Register obligations and ACMA enforcement for AI voice agents covers the compliance side.
- Number rental and concurrency. Twilio’s Australian page lists local numbers at $3.00/month. Retell publishes concurrency beyond the first 20 at “$8.00/Concurrency/month” and verified phone numbers at “$10.00/Phone number/month”. Bland’s plans carry platform fees of $299/month (Build) and $499/month (Scale) on top of per-minute rates. Vapi publishes concurrency at “10 included + $10 / line / mo”.
- Detection you pay for on every answered call. AMD at $0.0075/call is trivial per call and stops being trivial at volume: on 100,000 answered calls it is $750, and it buys you the ability to hang up on voicemail before the per-minute meters run.
- The rounding rule. Twilio bills voice by the minute. Retell states it tracks to the nearest second. On a campaign whose median connected call is 40 seconds, that single difference changes the telephony line by roughly a third.
None of these appear in a per-minute quote, and none of them are visible until the first invoice. If you are comparing published figures across vendors, our walkthrough of how to read an Australian AI phone agent pricing page sets out the five pricing models these fees hide behind.
How to work out your own cost per conversation this week
You do not need a vendor to do this, and you should not take anyone’s benchmark for it. The method takes about an hour with your existing call logs.
- Export your dial log for a full week, not a day. Include every attempt, with terminal disposition. If your platform does not expose dispositions, that is your first finding.
- Count dispositions separately — answered by human, answered by machine, no answer, busy, failed. Voicemail-answered is the row that quietly bills, so never fold it into “no answer”.
- Compute your real connect rate: human-answered dials divided by total dials. Substitute it into the sensitivity table above.
- Separate your invoice into per-dial and per-minute lines. Anything per-dial multiplies by your failure volume, not your success volume.
- Divide total spend by conversations, then by booked meetings. Cost per booked meeting is the only figure that tells you whether the campaign works.
Then the honest part. Doing this properly is not hard, it is relentless: you re-run it every time you change a script, a voice, a calling window or a list source, because every one of those moves the connect rate, and the connect rate is what the sensitivity table says you are actually buying. That is the operational cost the per-minute quote never shows — the analyst hours, the disposition hygiene, the re-washing schedule. It is also why volume changes the calculus rather than the rate: Zian’s learning engine tracks around 420,000 data points across more than 10,000 leads a day, and 28x more contact attempts only pays for itself if you know which attempts became conversations. Whether you keep that in-house or hand it over, the measurement discipline is the same, and the gap between a booked meeting and an attended one is a separate problem again — we covered it in the gap between AI-booked meetings and meetings that show.
Frequently asked questions
Do I get charged if an AI call rings out and nobody answers?
On the platform layer, Retell states plainly: “You are only charged for connected calls. If a call fails to connect, you won’t be billed.” On the telephony layer, Twilio’s Call resource documentation assigns an unanswered call the terminal status no-answer and states that duration “is empty for busy, failed, unanswered, or ongoing calls”, while charging applies to completed calls. Twilio documents the positive case; we could not locate an explicit statement on that page that a no-answer call carries no connectivity charge (checked 8 September 2026).
Am I charged when the call goes to voicemail?
Yes, on both layers. Twilio’s documentation states that a completed call “can occur when a call is answered by a person, an IVR phone tree menu, or even a voicemail”, and that “Completed calls, regardless of the outcome, are charged against your project balance.” Retell’s pricing FAQ states that for calls reaching voicemail, “billing applies only for the duration the AI agent is active on the line.” Twilio’s answering machine detection, published at $0.0075 per call, exists to end that call early.
Is there a minimum billed call length on AI voice platforms?
Retell publishes one: on its billing exceptions page, “Calls shorter than 10 seconds that use dynamic opening messages when AI speaks first” carry a “Minimum charge of 10 seconds”, so a 6-second call bills as 10. We could not locate a published minimum billed duration on Bland’s or Vapi’s pricing pages (checked 8 September 2026). Twilio’s status callback reference states that “calls are billed by the minute”, which functions as a one-minute floor on the telephony leg.
What does it cost in Australia before any call is made?
Washing your list against the Do Not Call Register is a published, per-number cost. The ACMA’s subscription schedule lists eight types, from Type A (500 wash credits, no fee) to Type H (100,000,000 credits, A$142,271); Type C provides 100,000 credits for A$585, which works out to A$0.00585 per number. Subscriptions run 12 months, and the register states that “In general you can rely on the washed list for 30 days”, so re-dialled lists incur the cost again.
Does billing stop when the AI transfers the call to a person?
It depends which meter. Retell states: “Once the call is transferred, the AI voice agent fee stops. Only the telephony fee continues for the remainder of the transferred call.” Bland instead publishes a separate Transfer Time Rate — $0.05/min, $0.04/min and $0.03/min across its three self-serve plans — alongside its Talk Time Rate. Twilio notes that its call price “only reflects the charge for connectivity” and that SIP REFER charges sit outside it; the Australian page lists SIP interface refer at $0.1000/refer.
Why should I measure cost per conversation instead of cost per minute?
Because per-minute rates are quoted against connected time, and connected time is the minority of an outbound campaign. Two stacks quoting an identical per-minute rate can differ by more than four times on cost per conversation once a per-dial fee is present, as the sensitivity table on this page shows. Cost per conversation, and then cost per booked meeting, are the units that decide whether the campaign pays.
Zian AI is in a partnership-application beta. Autonomous agents for outbound appointment setting, support and call closing across phone, SMS, email and WhatsApp, in 30+ languages, with private model deployment available on customer infrastructure. Apply For Partnership