To compare per-minute and flat-fee AI voice pricing, convert both into a fully loaded cost per connected minute at the usage your clients will actually generate, then add the line items the headline rate leaves out: telephony, model and voice costs, concurrency limits, overage rules and non-call usage like SMS. A flat fee wins when usage is high and steady. Per-minute wins when usage is small, spiky or unproven. For a reseller, the real question is which model keeps your margin predictable once you have thirty clients with very different call volumes.
Why do sticker rates make the two models hard to compare?
Because "per minute" rarely means one price. A voice call stacks several cost layers: carrier minutes, speech recognition, the language model, text-to-speech, and the platform itself. Some vendors bundle these into one rate. Others charge a platform fee per minute and bill the rest to you separately.
Vapi's public pricing page is a good example of why you have to read carefully: the platform fee is one component, and model, voice and telephony providers are separate. Check current terms there rather than trusting any summary, including this one. Carrier costs alone vary by route and provider, as Twilio's voice pricing page shows by listing rates per country and number type.
So a "cheap" per-minute headline and an "expensive" flat fee can land in the opposite order once everything is loaded. The only fair comparison is the all-in number.
How do you calculate cost per connected minute?
Use one formula for every vendor, flat or metered:
- Total monthly spend = platform fee + usage charges + carrier charges + model and voice charges + any packs or overage you would realistically buy.
- Total connected minutes = minutes the agent was actually on a call across all clients, including calls that end in a transfer.
- Cost per connected minute = spend divided by minutes.
Run it at three usage points per client: a quiet month, a typical month, and a bad month (a storm, a promotion, a campaign). Flat-fee plans look best in the bad month and worst in the quiet one. Metered plans do the opposite. If you only model the typical month, you will pick the wrong one for roughly half your clients.
What costs hide outside the minute rate?
A few that regularly surprise agencies:
- Concurrency. A flat fee that caps simultaneous calls can force an upgrade the day a client's phones get busy. We cover this in what a voice AI concurrency limit means for agency pricing.
- Billing increments. Whether a vendor rounds up per call or per second changes the cost of short calls like wrong numbers and hang-ups. Ask; don't assume.
- Unconnected minutes. Ask whether ringing, voicemail detection on outbound campaigns, hold and transfer time are metered. Answering machine detection means a meaningful share of outbound dials may never reach a person, and you want to know how those are billed.
- Non-voice usage. Texts during and after a call, SMS conversations, missed call text back and appointment reminders are all replies, not minutes. A plan that looks cheap on voice can be pricey once messaging is metered separately.
- Testing. Calling the agent from the dashboard before launch, and rerunning automated agent tests after every change, consume usage. Budget for it.
When does a flat fee actually win?
When a client's volume is high and predictable enough that they would blow past a metered bill. A multi-location clinic with steady inbound traffic is the classic case. You know the number, they know the number, and nobody gets a surprise invoice.
The catch is utilization. A flat fee only beats metered pricing if the client really uses what they pay for. Across a portfolio, a handful of heavy users often subsidize a long tail of small accounts that barely touch the allowance, and that tail is where agencies quietly lose margin or lose the client's trust at renewal. I can't give you a benchmark for how common that is; it depends on your vertical, so measure your own accounts.
When does per-minute win?
When you're early, when clients are small, or when volume is seasonal. Metered cost scales down to zero if the phone doesn't ring, which protects your margin on trial clients and slow months. Roofing and HVAC clients with weather-driven spikes are a natural fit, and our roofing storm-season piece shows how lumpy that demand can be.
The downside is client psychology. Small businesses dislike variable bills. If you resell metered usage as-is, the client feels every busy month. Most agencies solve this by packaging, not by passing the meter through.
How does this change when you resell under your own brand?
Your cost model and your client's price don't have to match. This is the part most comparisons skip. If you resell NovaReps as a white-label platform, you set client plans with your own prices, so you can sell a flat monthly fee to the client while your underlying cost is metered, or the reverse. The risk is yours to size, which is why you need the loaded cost per minute first.
Two mechanisms matter for that math:
- Bring your own keys with no per-minute markup, or managed minutes if you'd rather not hold provider accounts. With BYOK you pay model and voice providers directly, so the spread you're comparing is much easier to see. Our BYOK savings calculator lets you plug in your own assumptions, and what BYOK means for AI voice platforms explains the tradeoffs, including that you take on provider billing.
- Agency analytics for usage, revenue and margin per client, so you can see which flat-priced clients are costing you money rather than guessing.
Minute and reply packs with optional auto top up also give you a middle path: a bounded, prepaid chunk that behaves like a flat fee for the client without an open-ended bill for you.
What should you ask any vendor before deciding?
Put these in writing before you commit:
- What exactly is included in the quoted rate, and what is billed by a third party?
- How are partial minutes, transfers, voicemail and failed dials metered?
- What happens at the allowance limit: hard stop, overage, or forced upgrade?
- Is there a concurrency cap, and what does raising it cost?
- Are SMS, chat and reminders metered separately from voice?
- Can you set your own client prices, and can you see per-client margin?
What's the practical recommendation?
Don't pick a model first. Build a small spreadsheet with your five most representative clients, three usage scenarios each, and the full cost stack for every vendor you're considering. Choose the pricing structure whose worst month you can survive, then set client prices above it. A cheaper average that breaks in a spike month is the more expensive plan.