Every vendor conversation about answering machine detection eventually turns into a pricing conversation — and most of those conversations get the comparison wrong. Buyers line up quoted rates side by side ($0.0075 per call vs $0.004 per call vs a flat monthly fee) and pick whichever number looks smallest, without ever connecting the pricing model to what it's actually buying them in accuracy.
That's backwards. The real cost of AMD isn't the metered fee on the invoice — it's what a missed or misrouted live answer costs you, multiplied across however many thousand calls you dial this month. A vendor charging half a cent less per call but running a 20% false positive rate can cost you far more than one charging double the rate at a 2% false positive rate. This piece breaks down the three pricing models AMD vendors actually use, when each one makes sense, and the break-even math you need to run before signing anything — building on the broader AMD false positive cost framework we've covered before.
The three pricing models
Every AMD vendor prices one of three ways, and the model itself changes what you should optimize for.
1. Per-call (metered add-on)
This is the CPaaS default. You pay a small fee every time AMD runs, stacked on top of your regular per-minute voice rate. As of late 2026, published and street rates look roughly like this:
| Vendor | AMD fee | Billing unit | Notes |
|---|---|---|---|
| Twilio | ~$0.0075 | per call | Stacks on top of $0.0140/min outbound voice rate |
| Telnyx (Premium AMD) | ~$0.0065 | per call | Standard tier runs closer to $0.0020 |
| Plivo | ~$0.0040 | per request | Machine detection billed separately from voice |
| Vonage/Nexmo | Custom | per call | Priced in volume contracts, rarely published |
| Purpose-built AI AMD (e.g., amdify.io) | Flat per-seat or per-channel | per license | Unmetered — no per-call surcharge |
At a glance, $0.004–$0.0075 per call looks trivial. It isn't, once you run the math against real campaign volume.
2. Bundled into per-minute voice pricing
Some AI voice platforms (Vapi, Retell, Bland, ElevenLabs-style builders) fold AMD into a blended per-minute rate instead of itemizing it. You don't see an AMD line item on the invoice — it's baked into the $0.08–$0.15/min you're already paying for the call leg. This looks simpler, but it means you can never isolate what AMD itself is costing you, and you have zero leverage to negotiate accuracy improvements separately from call minutes.
3. Flat-rate or per-seat licensing
Purpose-built AMD engines designed for dialer platforms (VICIdial, GoAutoDial, Asterisk-based stacks) typically price by concurrent channel, agent seat, or server license — not by call volume. You pay the same whether you dial 50,000 calls a month or 500,000. This is the model built for centers running high call volume per agent, where metered pricing becomes a tax on productivity rather than a cost of infrastructure.
Why "cost per call" is the wrong number to compare
Vendors want you comparing $0.0075 vs $0.0065 vs $0.004, because at that resolution every option looks cheap. The number that actually matters is cost per correctly identified live answer — and that depends entirely on accuracy, not sticker price.
Here's the math most buyers skip. Take a mid-size operation running 40 seats, each seat averaging 300 dial attempts per shift, 22 working days a month:
- Call volume: 40 seats × 300 calls × 22 days = 264,000 calls/month
- Metered AMD cost at Twilio's $0.0075/call: 264,000 × $0.0075 = $1,980/month in AMD fees alone, before the underlying voice minutes
- Metered AMD cost at Telnyx Standard's $0.0020/call: 264,000 × $0.0020 = $528/month
That gap — roughly $1,450/month — is real money. But it's dwarfed by the accuracy gap. Asterisk's native AMD and most CPaaS detection engines run at a 15–25% false positive rate, meaning somewhere between 15 and 25 live human answers out of every 100 get misrouted to a voicemail drop, hung up on, or never connected to an agent. Purpose-built AI AMD engines bring that down to 1–3%.
Run that same 264,000-call base through a false positive cost framework and the numbers invert fast. If 30% of those calls connect to a live person — roughly 79,200 live answers — a 20% false positive rate silently kills about 15,840 real conversations a month. At even a conservative $8 average value per completed live connection (blended across appointment-set, sale, or payment-resolved outcomes), that's over $126,000 a month in lost opportunity. The $1,450/month difference in per-call AMD fees isn't the number that should be driving the vendor decision.
When metered pricing actually makes sense
Per-call pricing isn't always the wrong answer. It fits:
- Low, variable volume. If you're running under 10,000–15,000 calls a month, or volume swings seasonally, a flat per-seat license can cost more than you'd ever spend metered.
- Multi-channel platforms already on a CPaaS stack. If your dialer is already routing through Twilio or Telnyx for SMS, voice, and AMD in one bill, consolidating vendors has real operational value that can outweigh a few hundred dollars a month in AMD fees.
- Pilot and proof-of-concept phases. Before committing to a flat license across 40+ seats, metered AMD lets you test an engine's accuracy against your actual lead lists without a fixed commitment.
Where it stops making sense is exactly where most outbound operations actually live: sustained volume above roughly 20,000–30,000 calls a month, running day in, day out, where the metered fee becomes a permanent line item instead of a flexible cost.
Flat-rate licensing: the break-even math
Purpose-built AI AMD engines priced per-seat or per-channel usually land somewhere between $40–$150 per concurrent channel per month, depending on deployment size and whether it's self-hosted alongside VICIdial/Asterisk or delivered as a managed API.
Using the 40-seat example above, flat licensing at $100/channel/month = $4,000/month, flat, regardless of call volume. Compare that to Twilio's metered $1,980/month — on pure per-call fees, metered looks cheaper. But metered pricing is charging you the same $0.0075 whether the detection is right or wrong. Flat-rate AI AMD vendors compete on accuracy because that's the only lever they have left to justify the model — there's no volume upsell hiding behind a "detection fee" line item.
The break-even isn't about call count. It's about what a single misrouted live answer is worth to your center. If connect value per call is low (survey work, low-ticket outbound), metered pricing with a cheaper accuracy tier may be the rational choice. If connect value per call is high (insurance, final expense, debt resolution, high-ticket sales), the false positive tax at 15–25% erases any savings from the cheaper per-call rate within the first few thousand calls.
A pricing comparison checklist before you sign
Before locking into any AMD pricing model, get vendors to answer these in writing:
- Is the AMD fee itemized, or bundled into the voice rate? If bundled, ask for the AMD-only rate separately — you need it to compare against alternatives.
- Does the published accuracy number include a false positive and false negative breakdown, or just an aggregate "accuracy" percentage that hides which direction the errors skew?
- What's the fee at your actual monthly call volume, not the vendor's example volume? Metered pricing tiers often break at thresholds (10K, 50K, 250K calls) — know which tier you land in.
- Is there a minimum commitment or concurrent channel cap on the flat-rate option, and what happens if you exceed it mid-month?
- Can you run a side-by-side pilot — same lead list, same time window, two AMD engines — before switching your whole dial plan over? Most legitimate vendors will support this; if a vendor won't, that's a signal.
For a broader checklist covering accuracy claims, integration depth, and support SLAs beyond pricing alone, see the AMD vendor evaluation checklist.
The bottom line
Per-call pricing optimizes for the vendor's billing simplicity, not your connect rate. Bundled per-minute pricing hides the AMD cost entirely, which makes it impossible to negotiate or benchmark. Flat-rate licensing forces the vendor to compete on accuracy instead of volume, because there's no per-call meter to pad.
If you're running meaningful outbound volume — multiple seats, daily dialing, revenue tied to live connects — the pricing model that minimizes your invoice is rarely the one that maximizes your connect rate. Run the false-positive math against your own call volume and average connect value before comparing sticker prices. The vendor charging a quarter-cent less per call can easily cost you ten times that in missed live answers.
amdify.io runs on a flat, unmetered license specifically because of this math — our AI AMD engine holds false positive rates to 1–3% without charging a per-call detection fee, so the savings from better accuracy aren't offset by a metered surcharge. If you want a side-by-side accuracy and pricing comparison against your current AMD setup, reach out at amdify.io and we'll run the numbers against your real call volume.