If you run a service business, the question eventually stops being "should I get an AI phone agent" and starts being "which pricing model won't quietly double my bill in month four." Per-minute, per-call, and subscription pricing all look similar on a sales page. They behave very differently once your call volume is real and uneven.
This post is a math-grounded framework for picking between the three models. It assumes you've already decided an AI phone agent is the right move (we've written about that elsewhere — see the links at the bottom). What it covers: how each pricing model actually bills, where each one punishes you, the break-even math between them, and a five-step decision you can run on your own numbers in a spreadsheet.
The three pricing models in plain English
Vendors use different names, but the underlying structures collapse to three shapes. Once you see them, you can map almost any quote to one of these.
Per-minute billing
You pay for the wall-clock time the agent is on a call, usually rounded up to the nearest 6, 10, or 15 seconds. Common ranges we've seen across the market: $0.10–$0.25 per minute for the AI voice itself, with separate per-minute charges for the underlying telephony (often $0.01–$0.04) and any premium features like sentiment detection or live transfer.
The honest description: per-minute is the most flexible model for low or unpredictable volume. You're not committing to anything. You pay for what you use, and a quiet month costs you almost nothing.
The trap: per-minute is the most expensive model the moment your call volume grows. The same call that costs $1.20 in month one costs $1.20 in month twelve, even though you've sent the vendor ten times the call volume. There's no volume discount because the unit of billing is the minute, not the relationship.
Per-call billing
You pay a flat fee per completed call, regardless of whether the call lasted 22 seconds or 22 minutes. Common ranges: $0.50–$2.50 per call, sometimes tiered by call type (inbound vs. transfer, simple vs. multi-step).
The honest description: per-call is the closest analog to "pay per lead" thinking, which a lot of business owners find intuitive. It's also the model that punishes you hardest for short calls — a 30-second "we're closed, leave a message" call costs you the same as a 12-minute intake that books a $4,000 job.
The trap: per-call is the most expensive model for call types where duration actually corresponds to value. If your AI agent is doing real intake, qualifying leads, and booking appointments, you want longer calls, and per-call pricing charges you the same whether the call was valuable or not.
Subscription billing
You pay a flat monthly or annual fee for a bundled allocation — typically a number of minutes (1,000, 2,500, 5,000) or a number of calls (200, 500, 1,000). Overage rates apply above the bundle. Common ranges: $99–$499/month for a small-business tier, with overage at $0.05–$0.15/minute or $0.40–$1.50/call.
The honest description: subscription is the most predictable model and the most generous when you're using what you paid for. The break-even math almost always works in your favor once you're hitting 60–70% of your bundle.
The trap: subscription is the most expensive model when you underuse it. If you're a one-person consultancy that gets 80 calls a month and you bought a 2,000-minute plan, you're paying for 1,920 minutes of capacity you'll never use. And most subscription contracts auto-renew at the same tier unless you actively downgrade.
The break-even math between models
Here's the question to ask any vendor: at what monthly call volume does subscription become cheaper than per-minute, and at what call duration does per-call become cheaper than per-minute? Run the math with real numbers from your phone bill before you sign anything.
Example math for a 3-person HVAC company
Let's say you average 280 calls a month through the agent, with an average call duration of 3.5 minutes. That's 980 minutes of AI voice time per month.
At $0.18/minute per-minute pricing, you're paying $176.40/month, plus telephony. No commitment, no overage risk.
At $1.40/call per-call pricing (assuming that's the average across call types), you're paying $392/month. Per-call is roughly 2.2x more expensive than per-minute at this volume and duration.
At $199/month subscription for 1,000 minutes with $0.12/minute overage, you're paying $199/month as long as you stay under the bundle. 980 minutes fits. Subscription is the cheapest of the three by $73/month.
Now flip the inputs. Same company, but slow season — only 90 calls a month. Per-minute drops to $56.70. Per-call drops to $126. Subscription stays at $199. Per-minute wins by $73/month.
This is the structural reality: subscription wins in your busy season, per-minute wins in your slow season. The question is which season is longer.
Where per-call wins
Per-call only wins on average when either (a) your call durations are very short (under 60 seconds), and most of those calls are unqualified "wrong number" or "we're closed" interactions, or (b) you genuinely value predictability on a per-lead basis and you don't care that you're overpaying for the qualified leads to subsidize the unqualified ones.
If you're a pest-control company that gets 400 calls a month, 60% of which are 30-second "do you service my zip code" calls, per-call can win. Run the numbers — but know what you're buying.
The five-step decision
Step 1: Pull your actual call volume from your phone bill
Not the number your old answering service quoted you. Not the number your old receptionist thinks she handled. The number on the carrier's bill, broken down by hour of day and day of week. If you don't have this, your carrier almost certainly has a self-serve report. Pull the last 90 days, separate business hours from after-hours, separate weekdays from weekends.
Step 2: Estimate average call duration and call type distribution
Best estimate is fine. Listen to ten calls and time them. Mark which ones were qualified leads, which were unqualified, which were after-hours, which were transfers to a human. This is your call mix.
Step 3: Build a three-row spreadsheet with your real numbers
Row 1: per-minute (rate × minutes). Row 2: per-call (rate × calls). Row 3: subscription (flat fee + overage). Use your actual busy-month volume, not your average. Plan for the worst normal month, not the best.
Step 4: Stress-test against a 2x volume scenario
What does each model cost if your call volume doubles (because the AI is actually working and people are calling back, or because you added a service line)? Per-minute scales linearly. Per-call scales linearly. Subscription stays flat until it doesn't, then overage kicks in. This is where subscription's risk profile shows up.
Step 5: Negotiate the overage rate before you sign
Overage is where subscription contracts hide their cost. A 1,000-minute bundle at $199/month looks great until a single viral week puts you at 1,400 minutes and the overage is $0.20/minute — that's $80 of unbudgeted spend. Get the overage rate in writing. Get the rollover policy in writing. Get the downgrades-without-penalty policy in writing.
What we recommend (and why)
For most small service businesses (1–10 people, 100–500 calls per month, mix of qualified and unqualified): subscription wins once you've been running for two months and have real volume data. The predictability is worth more than the theoretical savings of per-minute.
For businesses with truly seasonal volume (landscaping, pool service, certain types of HVAC): per-minute can be cheaper across a full year, because you only pay for the months you're busy. Just make sure your vendor doesn't have a monthly minimum that erases the savings.
For businesses with very high call volumes (more than 2,000 calls/month): custom enterprise pricing usually beats all three published models. Ask for it. The worst they can say is no.
Common pricing claims to push back on
"Unlimited minutes" — there is no such thing. There are always fair-use caps, throttling, or feature restrictions. Read the fine print.
"No setup fees" — sometimes true. Often there are one-time charges labeled differently: onboarding, training, integration, custom-prompt authoring. Ask for the all-in first-month cost.
"Month-to-month, cancel anytime" — usually true for per-minute and per-call. Often NOT true for subscription, where 12-month commitments unlock the lowest rate. Get clarity before you assume flexibility.
"Pay only for answered calls" — verify whether "answered" means connected to a human, completed by the AI, or initiated. There's a meaningful difference, especially for per-call pricing.
The math behind the math
If you remember nothing else: pricing models aren't about which is cheapest in isolation. They're about which matches your actual call shape — volume, duration distribution, seasonality. The vendor's sales page is optimized for the average customer; you're not average. Run the numbers on your shape, not theirs.
And if a vendor won't give you transparent per-minute, per-call, and subscription quotes up front so you can compare them honestly, that's a signal. Move on.
Related reading
Which call types should an AI agent handle vs. a human — covers the call-mix question from a routing angle.
The true cost of an AI receptionist for a small business in 2026 — covers the TCO question across all-in pricing dimensions.
How to evaluate an AI phone agent: the 4-part buyer framework — covers the broader RFP and trial process.
FAQ
How do I know my actual call volume before I sign up for a plan?
Pull the last 90 days of call detail records from your phone carrier. Most carriers (RingCentral, Dialpad, Ooma, Grasshopper, even standard Verizon/AT&T business lines) have a self-serve report. Count total inbound calls, sum total minutes, and note the hour-of-day distribution. That gives you the inputs you need to compare plans honestly.
What about per-resolution or per-booking pricing?
It's emerging, especially in sales-focused AI agent products. The idea is appealing: you only pay when the AI actually books a meeting or qualifies a lead. The structural problem is that the vendor then has an incentive to game the metric — declaring any conversation a "resolution" to bill for it. Read the definition of "resolution" carefully, ask for examples, and weigh the alignment risk against the cost savings.
Should I sign a 12-month subscription contract for the lower rate?
Only if you've already run a real trial with real call volume and you trust the vendor's reliability. A 12-month commit at $199/month is $2,388 — that's a meaningful amount to lose if the product doesn't perform or the vendor's service degrades. Most subscription contracts have a 30-day out clause; use it as a built-in checkpoint.
What if my call volume is highly seasonal?
Per-minute pricing is structurally better for seasonal businesses, because you don't pay for capacity you don't use. The downside is no volume discount during peak season. Run a full-year scenario, not just one month, before deciding.
