What an AI phone call costs, line by line
A per-minute price is only one line of the bill. Here is every part of a call's cost, how to work out cost per outcome, and how to compare with human calling without fooling yourself.
The owner of a coaching institute in Kota sees two quotes for AI calling. One says "₹X per minute". The other says "₹Y per minute plus platform fee plus telephony". The cheaper-looking one turns out to be dearer once she adds up a month. This post gives you the lines to ask about, so you can compare quotes on the same basis.
All rupee rates below are invented example figures chosen for easy arithmetic. They are not Cora Call's prices or any vendor's. For real numbers, use the calculator at /pricing.
The parts of a call's cost
- AI per minute. Speech recognition, the language model and the voice each cost something. A platform may bundle them into one rate, or charge a chain of stages separately. Realtime models are often priced differently from chains of separate vendors.
- Telephony per minute. The carrier or trunk charges for the call itself. Rates differ by destination: mobile, landline, international.
- Pulse rounding. Some carriers bill in blocks (for example 30 or 60 seconds) rather than per second. A 20-second call billed in 60-second pulses costs three times what per-second billing would. Ask which applies; it matters most for short calls.
- Number and channel rentals. A phone number, and each concurrent line (channel), may carry a monthly rental whether or not you call.
- Platform fee. A monthly plan fee for the software, often with some usage or credit included.
Connected minutes versus attempts
Most billing runs on connected minutes: time from when the call is answered. Unanswered attempts usually cost little or nothing in AI terms, but check whether your carrier bills ringing time or voicemail pickup, and whether a call answered by voicemail counts as connected. A campaign that reaches few of its numbers pays for far fewer minutes than one that reaches most of them, and yet needs more attempts to deliver the same outcomes.
How bundled credit works varies. A plan may include credit that is spent first, with usage beyond it billed at a set rate. Ask three things: does unused credit roll over, what is the rate after it runs out, and does credit cover telephony as well as AI.
Reading a quote line by line
When two quotes arrive, rebuild each one on the same sheet. Write the five lines above down the left, put each vendor in a column, and fill every cell. Where a quote says "all-inclusive", ask what is inside: does the rate cover the carrier leg, the number, recording storage and transfers to a person? Where a cell stays empty, that is a question for the vendor, not a zero.
Also note which costs scale with calls and which do not. Rentals and platform fees are fixed, so they weigh heavily in a small pilot and lightly at volume. Per-minute charges are the opposite. A vendor with a low fixed fee and a high minute rate suits a trial; the reverse suits a steady high volume.
Cost per outcome is the number that matters
A cheap minute that books nothing is expensive. Divide total spend by the outcomes you wanted: appointments booked, leads qualified, payments collected.
A worked example (invented rates)
A Pune dental clinic runs a reminder and rebooking campaign for one month. Every figure below is an invented example.
| Line | Example rate | Quantity | Cost |
|---|---|---|---|
| AI per connected minute | ₹4 | 1,500 min | ₹6,000 |
| Telephony per minute | ₹0.60 | 1,500 min | ₹900 |
| Pulse rounding allowance (billed minutes ~10% above talk time) | ₹0.60 | 150 min | ₹90 |
| Number rental | ₹500 a month | 1 | ₹500 |
| Platform fee | ₹3,000 a month | 1 | ₹3,000 |
| Total | ₹10,490 |
Suppose the campaign makes 1,000 connected calls averaging 90 seconds (1,500 minutes) and books 120 appointments. Cost per booked appointment is ₹10,490 ÷ 120, about ₹87. If a better opening line lifts bookings to 150 at the same spend, it falls to about ₹70. The lever was the script, not the rate.
What moves the bill up or down
Four things change the total more than the headline rate does. Call length: a long, rambling script costs more per outcome than a tight one. Answer rate: lists with stale numbers waste attempts and platform time. Retries: calling each number three times triples the attempts for a modest gain in connections. Handoffs: every call transferred to a person adds human time on top of the AI minutes already spent.
Ask also about charges that appear only on some plans: recording storage, extra concurrent lines during a busy hour, and fees for calling from rented numbers rather than your own. None of these is unusual, but each belongs on the comparison sheet. A quote that leaves them out is not wrong, only incomplete.
Finally, check how the vendor reports cost. A call-by-call figure beside the transcript lets you spot the three-minute calls that went nowhere. A single monthly invoice does not.
Comparing with human calling, fairly
Compare on the same footing, and use your own numbers. Here is a structure with assumptions you should replace.
| Assumption (edit these) | Example value |
|---|---|
| Caller's monthly cost to you, including salary, benefits, seat and tools | ₹30,000 |
| Productive calling hours per month | 100 |
| Calls handled per productive hour | 12 |
| Appointments per 100 connected calls | 12 |
On those invented figures, a person costs ₹300 an hour, handles 1,200 calls a month and books about 144 appointments, roughly ₹208 each. That is more than the AI example above, but the comparison is thin. It ignores supervision, training, sick days and attrition on the human side. It also ignores setup time, script tuning, calls the AI cannot finish and hands to a person, and quality differences on the AI side. A human may also convert better on complex or emotional calls.
Treat the result as a way to find your break-even, not a promise of savings. The honest picture usually mixes the two: AI takes first contact, reminders and after-hours calls, and people take the conversations that need judgement. See handing an AI call to a person for how that works.
Mistakes that make AI calling look cheaper than it is
- Counting only talk time. Billed minutes can exceed talk time once pulses are applied.
- Leaving out setup. Writing and testing the script, connecting your calendar or CRM, and reviewing the first few hundred calls take staff hours.
- Ignoring the calls that fail. A call that ends in a confused caller still costs minutes, and may cost you a customer.
- Comparing a full month of human work with a pilot week of AI. Compare like periods, and include the same outcomes on both sides.
Where to start
- Ask every vendor for the five lines above in writing, including pulse size and what credit covers.
- Estimate connected minutes from your list size, expected answer rate and call length.
- Run the sums in the pricing calculator with your own numbers.
- Run a small pilot and measure cost per outcome before you scale.
- On Cora Call, each call shows its cost alongside the transcript and summary, which makes this measurement easier. Pricing is per connected minute plus a plan fee, so use /pricing for figures.