Every published margin figure in this category is asserted by a company selling the platform that sits on the cost side of the calculation. We went looking for one that is not, read eleven of them back to their source, and did not find a single disinterested number.
That is worth saying plainly before any arithmetic, because the arithmetic turns out to be less interesting than where the money actually comes from.
The margin claims, and who is making them
| Claim | Asserted by | Working shown? |
|---|---|---|
| 65–85% gross margin, 97% at 280 clients | Ringlyn, a white-label platform | Yes |
| 50–70% profit at $297–997 per client | Trillet, a white-label platform | Yes |
| 50–75%, top performers past 80% | Trillet, second article | Yes, and it contradicts its own headline |
| 70–90% gross margin | Leadlock, citing Famulor | No |
| 50–75% and 65–80%, ROI up to 760% | Blck Alpaca, citing Famulor | No |
| Platform $0.04–0.10 vs agency $0.20–0.35 | Famulor, a voice platform | Partially |
| 65% gross margin | Cekura, a voice QA vendor | Yes |
| Setup $2,500–$15,000, retainers $500–$5,000 | A paid agency directory | No |
Two of those trace to the same origin. Leadlock cites Famulor. Blck Alpaca cites Famulor. Famulor is itself a voice platform whose own tier is the favourable comparator in its own analysis. The 70 to 90% figure has one source and at least two apparent independent confirmations.
The most useful one is Trillet's, because it argues against itself. The headline says 50 to 75%. The table inside the same article computes 28 to 64% for a single client, and only reaches 85% by dividing a $299 platform fee across twenty clients while holding usage flat. That is fixed-cost dilution presented as margin improvement. The per-minute economics are identical in both rows.
The one vendor that itemises, and where it hides the rate card
GoHighLevel is the only participant publishing a complete component-level rate card with effective dates. It is in a support article rather than on the pricing page, which is why almost nobody quoting GHL economics has read it.
| Component | Rate | Effective |
|---|---|---|
| Voice engine (GHL's own charge) | $0.045/min | 20 May 2026 |
| TTS — OpenAI or Cartesia | $0.015/min | 20 Apr 2026 |
| TTS — ElevenLabs V2.5 | $0.035/min | 20 May 2026 |
| TTS — ElevenLabs V3 | $0.170/min | 20 Apr 2026 |
| Telephony | Not published, stated as “the same amount as Twilio” | — |
GHL's own orchestration charge of $0.045 a minute is lower than Retell's $0.055 infrastructure fee and lower than Vapi's $0.05 hosting fee. GoHighLevel is not the expensive option in this category. It is the cheapest orchestration layer of the ones that publish, and it is the only one that dates its components.
One dropdown moves the whole business
Combine GHL's published components with our verified $0.0140 Twilio rate and a low-latency model at $0.0108, and the voice choice does this:
| Voice | All-in per minute | × our component floor |
|---|---|---|
| OpenAI or Cartesia | $0.0848 | 1.82× |
| ElevenLabs V2.5 | $0.1048 | 2.25× |
| ElevenLabs V3 | $0.2398 | 5.16× |
A single menu selection moves the unit cost 2.8 times, inside one vendor's own rate card. On a $297 monthly retainer, the break-even point moves from about 2,916 minutes to about 1,031 minutes. Choosing the better voice costs a reseller roughly 1,885 minutes of headroom per client per month, and it is a dropdown in a settings panel.
The margin is the retainer, not the markup
Here is the arithmetic the literature does not run. Ten clients, one number each, $297 a month, platform fees divided across the book.
| Path | At 500 min/client | At 2,000 min/client |
|---|---|---|
| Wrapper over Retell | 75% | 27% |
| Trillet Agency | 71% | 10% |
| GHL, cheap voice | 69% | 26% |
| GHL, ElevenLabs V3 | 43% | −78% |
| Ringlyn white-label bundle | 16% | 16% |
At 500 minutes the reseller's variable cost is around $42 to $48 against a $297 retainer, so roughly 85% of the gross profit is the retainer itself rather than any markup on minutes. The margin exists because the client bought a subscription and used a seventh of it.
At 2,000 minutes the same business runs at 10 to 27%, and the premium-voice configuration loses money on every client. Nothing about the reseller changed. The client talked more.
Setup fees do not rescue this. A published $499 setup is 1.68 months of a $297 retainer, so across a twelve-month life it adds about 14% to gross revenue and nothing at all to steady-state margin. It recovers acquisition cost. It is not a margin source.
What the buyer pays, against what it costs
A client paying $297 for 500 used minutes is paying $0.594 a minute against a verified component floor of $0.0465. That is 12.8 times. At 2,000 minutes it falls to $0.1485, or 3.2 times.
Of that 12.8 times, the platform takes roughly 2 times and the reseller takes the remaining 6. The reseller's share evaporates the moment the phone actually rings, which is the part no margin table shows.
The clearest retail rate card we found runs even higher. One productised receptionist service publishes overage at $0.70 to $0.85 a minute, which is 15 to 18 times the floor, and prices its included minutes at $0.60 to $0.73 a minute. Its bundled rate is dearer than most competitors' overage rate. That is a rate card built on the assumption the minutes will not be used.
GoHighLevel says the quiet part in its own documentation
The rebilling documentation is unusually candid, and it contradicts most of what is written about it: “Your agency is always charged by HighLevel based on actual token consumption, regardless of the pricing model you configure for sub-accounts… If you set rates too low, you will absorb the difference.”
The $0.10 per voice minute and $0.50 per message figures that circulate as “GHL's rebill rates” are, on that page, explicitly examples of what an agency might charge a client. They are not rates GHL charges anyone. Every third-party article we read repeats them without the qualification.
The number that would settle this is the one nobody publishes
Every margin figure above, ours included, rests on an assumed monthly minute volume per client. Nobody publishes what a real small-business client actually consumes. Not the platforms, not the resellers, not the directories.
Since the entire margin structure is a bet on under-utilisation, that missing input makes every published claim unfalsifiable. The reseller margin claims are not so much wrong as untestable, and they are being sold to people who will discover the answer one client at a time.