Research-backed calculator guide
Inbound vs Outbound Voice AI Cost: Compare Answered Minutes, Dials and Connection Rates
Compare inbound and outbound voice AI cost with a denominator-aware model for answered calls, dial attempts, connection rates, telephony, and AI minutes.
Quick answer
Inbound and outbound voice AI cost should not be compared with one headline per-minute number. Inbound cost usually starts with offered calls, then converts to answered calls and connected minutes. Outbound cost starts with dial attempts, then applies a connection rate before calculating the conversation minutes that the AI actually handles. The same agent rate can therefore produce very different cost per answered call, cost per dial, and monthly spend.
Use this basic model:
- Inbound answered calls = offered calls × answer rate.
- Outbound answered calls = dial attempts × connection rate.
- Connected AI minutes = answered calls × average connected minutes.
- Total voice AI cost = telephony charges + AI voice charges + model, number, feature, and operating charges.
For a comparable US example, 1,000 inbound calls answered 90% of the time at four minutes each create 3,600 connected minutes. Two thousand outbound dials connected 25% of the time at three minutes each create 1,500 connected minutes. Using illustrative pay-as-you-go rates observed on 2026-11-02, the inbound flow costs more in total because it creates more connected minutes, even though the outbound campaign makes twice as many call attempts. The outbound flow still needs an attempt-cost check: a carrier or dialer may charge for some no-answer, voicemail, or machine-detection activity.
The practical comparison is not “inbound is cheap” or “outbound is expensive.” It is: what is the billable unit, what fraction of attempts become conversations, and which denominator will you use to judge the result?
Why inbound and outbound have different cost denominators
An inbound line begins with a customer deciding to call. “1,000 inbound calls” can mean 1,000 offered calls, 1,000 answered calls, or 1,000 completed conversations. Those are not interchangeable. A queue abandonment rate, after-hours routing rule, or short abandoned call can change the number of minutes without changing the offered-call count.
An outbound campaign begins with a dial plan. “2,000 outbound calls” usually means 2,000 attempts. Some calls fail immediately, some reach voicemail, some reach a person who declines, and some become a conversation. Connection rate is the bridge between attempts and answered calls:
outbound answered calls = outbound dials × human connection rateA cost report that divides outbound spend by dials answers “what did each attempt cost?” A report that divides by answered calls answers “what did each live connection cost?” Both can be useful. They answer different management questions. For sales or qualification, cost per connected call and cost per qualified outcome may be more useful than cost per dial. For capacity planning, connected minutes and peak simultaneous calls matter more.
Build the cost in layers
Start with a call-flow ledger rather than a vendor headline. Keep these layers separate:
- Attempt layer. Count inbound offers or outbound dials. Record answer, abandonment, voicemail, busy, failed, and rejected outcomes.
- Conversation layer. Count answered calls and connected minutes. Record the average and the distribution of call length, because a long tail can dominate spend.
- Telephony layer. Apply the carrier or contact-center rates to the correct direction, destination, number type, and billed duration. Twilio’s US rate card, for example, lists local calls at $0.0140 per minute to make and $0.0085 per minute to receive, while toll-free receive is $0.0220 per minute. It also lists separate monthly number prices and optional features such as answering-machine detection. These are vendor quotes observed on the page on 2026-11-02, not universal market rates. [1]
- AI layer. Apply the voice-agent, STT, TTS, and LLM meters. Deepgram lists a bundled Standard Voice Agent rate of $0.075 per connection minute on pay-as-you-go, and lower tiers when you bring your own TTS or language model. Its page says the Voice Agent meter is based on WebSocket connection time, so confirm whether your implementation keeps a session open during hold or idle time. [3]
- Fixed and operational layer. Add phone-number rental, platform subscriptions, recordings, summaries, CRM integration, human transfer, compliance, support, and implementation if they apply.
The split between service and telephony is important. Amazon Connect documents a $0.018-per-minute voice service charge for both inbound and outbound usage, then adds communication-service charges. Its US example uses a $0.0022-per-minute inbound DID rate, a $0.0048-per-minute outbound destination rate, a $0.0300 daily DID charge, and per-second billing with stated minimums. That structure demonstrates why an “inbound rate” and an “outbound rate” may each be only one line in a larger bill. [2]
For a broader provider comparison, use the model pricing comparison to keep the model or API layer separate from carrier and orchestration charges. If an estimate will be used for finance, the AI API invoice reconciliation calculator can help map those assumptions to invoice rows after launch.
Worked cost example
The example below is intentionally inspectable rather than a quote for any particular production stack. It uses US local-number assumptions and rates observed on 2026-11-02: Twilio local receive at $0.0085/minute, Twilio local make at $0.0140/minute, and Deepgram Standard Voice Agent at $0.075/connection minute. The example applies the AI rate only to connected minutes and excludes failed-attempt carrier charges, number rental, taxes, LLM calls outside the bundled agent rate, human labor, and fixed platform fees. Replace each rate with the rate card for your actual destination and stack.
| Flow | Volume assumption | Conversion to answered calls | Billable AI minutes | Telephony input | AI layer input | Modeled total | Useful denominator |
|---|---|---|---|---|---|---|---|
| Inbound | 1,000 offered calls; 4.0 min average | 1,000 × 90% = 900 answered | 900 × 4.0 = 3,600 | 3,600 × $0.0085 = $30.60 | 3,600 × $0.075 = $270.00 | $300.60 | $0.334 per answered call; $0.0835 per connected min |
| Outbound | 2,000 dials; 3.0 min average | 2,000 × 25% = 500 answered | 500 × 3.0 = 1,500 | 1,500 × $0.0140 = $21.00 | 1,500 × $0.075 = $112.50 | $133.50 | $0.267 per answered call; $0.0668 per dial; $0.089 per connected min |
The inbound total is higher here because the scenario produces 3,600 connected minutes, not because inbound calls are inherently more expensive. The outbound campaign makes twice as many attempts but produces fewer live minutes. If the carrier bills an outbound attempt before a person answers, add that cost separately:
outbound total = (dials × attempt charge)
+ (answered calls × connected minutes × telephony rate)
+ (answered calls × connected minutes × AI rate)
+ fixed and feature chargesThe effective outbound cost per dial is $133.50 ÷ 2,000 = $0.06675 under the simplified assumptions. The effective outbound cost per answered call is $133.50 ÷ 500 = $0.267. If a manager looked only at the $0.089 connected-minute figure, the campaign’s reach efficiency would disappear. If a manager looked only at cost per dial, a campaign with a poor connection rate could look artificially efficient because it generated fewer conversations.
Connection-rate sensitivity
Hold 2,000 outbound dials, a three-minute connected conversation, and the same combined connected-minute rate of $0.089 constant. The connection rate changes both spend and the denominator used for a campaign decision:
| Human connection rate | Answered calls | Connected minutes | Modeled connected-minute spend | Spend per answered call | Spend per dial |
|---|---|---|---|---|---|
| 15% | 300 | 900 | $80.10 | $0.267 | $0.0401 |
| 25% | 500 | 1,500 | $133.50 | $0.267 | $0.0668 |
| 40% | 800 | 2,400 | $213.60 | $0.267 | $0.1068 |
This table is not an ROI forecast. It shows the mechanical consequence of the assumptions. At a fixed three-minute conversation length, the cost per answered call stays constant because every answered call consumes the same connected minutes. Total spend rises with connection rate because more people actually talk to the agent. If the business values conversations, a higher connection rate is not automatically a cost problem; it is a volume and outcome decision. If outbound attempts incur a separate charge, the actual spend curve becomes higher and should be modeled from the carrier’s attempt rules.
What to collect before comparing vendors
Ask every vendor to define “minute.” It may mean carrier duration, active conversation duration, audio processed, WebSocket connection time, or a rounded billing block. Deepgram states that its ordinary speech APIs bill by audio second and its Voice Agent rate uses connection time, while AWS describes minimum billing for separate service and telephony layers. Those meters cannot be compared without converting them to the same call-flow assumptions. [2] [3]
Record whether inbound and outbound use the same number type. A local inbound number, toll-free inbound number, outbound caller ID, international destination, and SIP trunk can each have a different rate. Record number rental and daily charges separately from minutes. Also ask whether call recording, transcription, summaries, sentiment, AMD, transfer legs, and storage are optional add-ons. A rate card that bundles these features is not directly comparable with one that itemizes them.
For outbound, request examples for no answer, busy, voicemail, invalid number, and early hang-up. An AI dialer can spend money before a conversation begins through setup time, carrier duration, AMD, or a short prompt. For inbound, request queue, abandoned-call, transfer, and after-hours examples. A missed inbound call may still create a telephony event, while a connected transfer can create a second billable leg.
Finally, choose the outcome denominator before launch. A support team may track cost per resolved call. A lead-qualification team may track cost per human connection and cost per qualified lead. A collections team may need cost per right-party contact. The observability ROI calculator is useful when the voice system’s value depends on measured outcomes rather than minutes alone.
FAQ
Is outbound voice AI always more expensive than inbound voice AI?
No. Outbound often has more attempts per live conversation, but total cost depends on connection rate, call length, attempt billing, destination, caller-ID type, and the AI meter. Inbound may create more connected minutes because callers reach the agent directly. Compare the same volume window using both cost per answered call and cost per connected minute.
How do I calculate AI outbound call cost per connected call?
Add all outbound campaign charges for the period, then divide by human-answered calls. If you are modeling before launch, estimate dials × connection rate for answered calls. Include attempt charges, connected telephony, AI minutes, transfers, and fixed fees in the numerator. Do not divide by dials unless you specifically want cost per attempt.
What is the right denominator for voice bot inbound call pricing?
Use at least two denominators. Cost per answered call shows what it costs to handle a caller who reaches the bot. Cost per offered call shows the cost across the whole inbound funnel, including calls that abandon or never connect. For staffing and provider capacity, also retain connected minutes and peak concurrency.
Do no-answer outbound calls cost money?
They can. The answer depends on the carrier, platform, destination, minimum billing increment, voicemail or answering-machine detection, and how long the call remains open. A conservative forecast keeps an explicit attempt-cost line instead of assuming every unanswered call is free. Ask for rate-card examples for busy, no-answer, voicemail, and invalid-number outcomes.
Should I use one blended per-minute rate for both directions?
Use a blended rate only after you have verified that telephony, AI, TTS, STT, LLM, and feature meters are genuinely bundled and direction-neutral. Otherwise retain separate inbound and outbound telephony rows. A blended rate is convenient for a stable workload, but it can hide a change in connection rate or a different outbound destination mix.
What costs are excluded from this worked example?
The example excludes taxes, number rental, subscription minimums, implementation, CRM work, human escalation, compliance, recording storage, post-call analysis, failed-attempt carrier charges, and any LLM or TTS layer outside the assumed bundled agent rate. It also uses illustrative US local rates observed on 2026-11-02, not a promise of future pricing. Recalculate with your destination, plan, billing increment, and measured call logs.
Run your scenario
Turn this framework into an editable estimate.
Use dated, source-linked assumptions and see cost per request, successful task and month.
Primary sources checked
Pricing and product behavior can change. These sources support the time-sensitive claims in this guide; verify current provider terms before committing spend.