Short answer: a serious CPaaS pricing evaluation runs four steps in order — decode the billing units per channel, normalize every vendor to a common unit (cost per 1,000), stress the failure-billing and idle-cost edges, and score the whole stack on a weighted scorecard — because the differences that move a real bill hide in the units, not in the headline rate. Devotel Orbit publishes its full rate card on one pricing page: $0.014 per voice minute, $0.0075 per SMS segment, $0.01 per RCS message, $0.001 per email, WhatsApp passing through Meta's own per-message fees at zero platform markup, and the AI agent layer usage-billed with no per-seat license — so the worked examples below are verifiable against public numbers, not a sales deck.
Pricing evaluations fail in predictable ways. A buyer collects five rate cards, ranks them by the SMS per-segment number, and signs with the apparent winner — then discovers that voice bills failed call attempts, that UCS-2 emoji traffic splits into two segments per message, and that the AI agent layer carries a separate per-seat platform fee the per-turn rate never mentioned. The transparent pricing models guide scores the transparency shape a vendor wears; this guide is the evaluation method — the arithmetic and the checklist a buyer runs once the rate card is in hand.
We publish with ourselves in the comparison, disclosed rather than hidden. Every Orbit figure below is a public page figure a buyer can check against the pricing page before any call happens.
1. The four evaluation steps
A CPaaS pricing evaluation that survives finance review runs four steps, in this order. Skipping any one of them is how the wrong vendor wins.
- Decode the billing units. Every channel bills in a unit: per-minute versus per-second for voice, per-segment versus per-message for SMS, per-turn versus per-minute versus per-token for AI agents. A rate is only comparable after the unit on both sides of the comparison is identical.
- Normalize to cost per 1,000. Convert each channel into cost per 1,000 units of traffic — minutes, segments, emails, agent interaction units — so a traffic mix (say, 60,000 voice minutes + 40,000 SMS + 2,000 agent sessions a month) prices to one arithmetic line, not five side-by-side rates.
- Stress the edge cases. Failed-call billing, partial-minute rounding, idle-month cost, and pass-through markup separation decide whether the modeled bill survives production. A rate that only wins at perfect traffic is a weaker offer than it prints.
- Score the full stack on a weighted scorecard. Each channel gets scored, then weighted by its share of the buyer's traffic mix, so a vendor that wins SMS but loses the AI agent layer is ranked on the mix that actually ships — not on one channel in isolation.
The rest of the guide works each step with real units, then fills the scorecard.
2. Decode the billing units, per channel
Billing-unit mismatches are the single largest source of bad comparisons. Decode each channel before ranking anything.
Voice — the rounding and the failure-billing question. Minute-level billing rounds every 7.25-second call up to a full minute; per-second billing charges the 7.25 seconds. At 100,000 short calls a month, a 60-second rounding scheme can cost roughly 3–4x the per-second scheme at the same nominal per-minute rate. The second question is failed attempts: some vendors bill ring time or attempted sessions even when the callee never answers; a usage-honest vendor bills answered time. Orbit bills answered minutes at the published $0.014 rate from the pricing page, so the evaluation row asks two questions: what is the rounding unit, and what counts as a billable attempt.
SMS — the segment and the pass-through question. A 7-bit GSM message is 160 characters per segment; switch to UCS-2 (emoji, Arabic, CJK scripts) and the cap drops to 70 characters per segment, so a long message multiplies the billed segment count. A per-message rate and a per-segment rate are different products, and a buyer comparing them on the sticker number underprices the segment vendor by the message length. The second question is markup separation: a carrier pass-through fee, where it exists, should appear as a named line separate from the platform's markup — the zero-markup WhatsApp pass-through is the model: the buyer sees exactly what the platform adds on top of Meta's own per-message fee. Orbit's SMS bills at $0.0075 per segment, published, so segment arithmetic is pre-launch spreadsheet work rather than a surprise on the first invoice.
AI agents — the component-stack question. An AI voice agent consumes four components per conversation: the telephony leg, speech-to-text, LLM tokens per turn, and text-to-speech. A vendor quoting "per agent minute" while a second vendor quotes "per LLM token" is quoting incomparable units, and the buyer must price all four legs to normalise them. The consolidated-platform shape bills the agent's call minutes against the same published voice-minute rate card as any other call — $0.014 per minute — with the agent layer itself usage-billed and no per-seat license to normalize away. The self-assembled shape prices the same interaction across four vendors and four bills. The AI voice agent pricing breakdown and the AI voice agent TCO model run the component arithmetic leg by leg; the evaluation row asks whether the vendor's all-in agent price survives decomposition.
Email, RCS, and WhatsApp — the single-wallet breadth check. A full-stack comparison normalises the whole traffic mix, not just voice and SMS. A vendor exposing published rates across email ($0.001 per email), RCS ($0.01 per message), and WhatsApp (pass-through at zero markup) in one wallet removes the friction of five bill-to-rate translations; a vendor published on one channel and quote-gated on the next inherits the weaker shape. Decode every channel before scoring any of them.
3. Normalize everything to cost per 1,000
Mixed-unit comparisons are how a floor price gets mistaken for a better deal. The normalization method expresses every channel as cost per 1,000 units of traffic:
- Voice: $0.014 per minute → $14.00 per 1,000 minutes.
- SMS: $0.0075 per segment → $7.50 per 1,000 segments.
- RCS: $0.01 per message → $10 per 1,000 messages.
- Email: $0.001 per email → $1 per 1,000 emails.
Now a traffic mix — for example, 60,000 voice minutes + 40,000 SMS segments + 15,000 emails a month — prices to one arithmetic line:
60 × $14.00 + 40 × $7.50 + 15 × $1 = $840 + $300 + $15 = $1,155 per month
That line replaces a spreadsheet of incomparable per-unit rates, and it is the line a finance team signs off on before a pilot starts. A vendor whose pricing cannot produce this line cannot be evaluated — which is the filter's purpose. The pricing overview hub runs this math in public for the whole stack, and the CPaaS pricing models post prices the usage-based versus flat-rate versus revenue-share billing shapes in TCO terms; this guide concerns the evaluation method, not the billing-model argument those sibling posts already settle.
4. Stress the edge cases
Four edge cases decide whether the normalized bill survives contact with production traffic. Score each one explicitly rather than discovering it on the invoice.
- Failed-attempt billing. Voice vendors that bill ring time or unanswered attempts inflate the effective per-minute rate by the abandonment share of outbound traffic. Ask precisely what counts as a billable attempt.
- Rounding unit. A per-minute scheme rounds partial minutes up; a per-second scheme charges exact time. On short outbound calls, the rounding scheme can multiply apparent cost at identical sticker rates.
- Idle-month cost. Usage-based with no floor costs $0 at zero traffic; a platform minimum, per-seat licence, or flat-rate floor is a sunk cost at any idle month. Ask what an idle month costs from the pricing page, not from a sales call.
- Pass-through markup separation. Where an upstream fee exists — a carrier SMS fee, a Meta WhatsApp per-message fee — the upstream cost and the platform's markup should be shown as separate numbers. A vendor that cannot show the separation makes the encoding of any future repricing invisible.
Each edge case is a multiplier on the normalized line from step 3, so stress them before the scorecard, not after.
5. The weighted full-stack scorecard
After the units are decoded and the edges stressed, the comparison collapses into six scored rows per vendor, weighted by the buyer's actual channel mix. The rows below, weight-adjusted per mix, are a scorecard a buyer can fill before procurement starts.
| Scored row | What a full mark looks like | Where a vendor fails |
|---|---|---|
| Published rate per active channel | Public per-destination tables on every channel in your mix | A channel quote-gated or "contact sales" |
| Billing-unit honesty | Rounding unit named; billable-attempt definition published | Partial units round up silently; failed attempts billed |
| Component-stack decomposition (AI) | All four legs — telephony, STT, LLM, TTS — price out to one all-in agent figure | An "AI rate" that hides per-seat or per-token components |
| Pass-through separation | Upstream fee and platform markup shown as separate lines | Markup invisible behind a flat per-message number |
| Idle = $0 | No floor, no seat, no minimum at zero traffic | A platform floor that bills at zero volume |
| One wallet across the mix | One prepaid wallet or one negotiated rate card spanning voice + SMS + email + RCS + WhatsApp + agents | Five vendors, five bills, five markups |
Weighted scoring matters because channel importance varies by buyer. A retail messaging-heavy buyer weights SMS rows heavily; a contact-center buyer weights the voice and AI agent rows; an omnichannel platform weights the wallet-breadth row. The scorecard makes that weighting explicit instead of ranking one channel and calling the evaluation done.
Devotel Orbit prices usage-based, end to end: the pricing page ships one self-serve Pay-as-You-Go track with no monthly fee and no seats, one prepaid wallet across voice, SMS, RCS, email, WhatsApp, video, and AI agents, and an Enterprise track that negotiates volume rate cards on the same pay-as-you-go structure — so the scorecard above scores the published lines a buyer can check without a meeting.
The decision table
The evaluation ends in the same place regardless of mix: a chosen pricing shape, plus a reason that survives a finance review.
| Buyer's constraint | The pricing shape that fits | Why |
|---|---|---|
| Finance must model next quarter's spend tonight | Published per-destination rate card | The bill is modelable from public rates without a sales cycle |
| A real AI agent rollout is scoped | Usage-billed agent layer, no per-seat license | The component stack prices to one line; seat floors do not sink |
| SMS volume dominates the mix | Per-segment pricing with markup separation | Segment arithmetic survives long-message and Unicode traffic |
| Mixed channels under one account | One wallet, one rate card, one posture | No five-bill reconciliation, and volume discounts stack cleanly |
| Enterprise-scale traffic is negotiated | Published base rates + a negotiated volume rate card | The same pay-as-you-go structure survives volume-tier pricing |
| The evaluation shortlisted a quote-only vendor | A quote cycle priced as a cost, not a free step | Five days of pre-launch modeling beats most discounts |
No row is a verdict — a buyer with genuine enterprise leverage can beat a public rate card on a quote cycle, and the CPaaS pricing models post prices exactly when flat-rate or revenue-share wins. The table narrows the shape that deserves the evaluation cycle; the scorecard above prices it.
Frequently asked questions
What is the single most useful number for comparing CPaaS pricing across vendors?
Cost per 1,000 normalized units of traffic. Converting each channel to one arithmetic line — for example voice at $14 per 1,000 minutes and SMS at $7.50 per 1,000 segments — replaces side-by-side per-unit comparisons with one modeled monthly bill. A vendor whose pricing cannot produce that line cannot be evaluated.
Why shouldn't I just compare the published per-minute or per-message rate?
Because the billing unit, the billing attempt definition, and the markup separation hide in the units, not the sticker. Voice billed on 60-second rounding costs roughly 3–4x per-second billing at identical sticker rates on short calls; SMS billed per message versus per segment differs by message length and charset; a vendor quoting an "AI rate" may hide per-seat or per-token components. Decode the units first, then compare.
How should I price an AI voice agent against a human agent and against another AI vendor?
Price all four component legs — telephony, speech-to-text, LLM tokens per turn, text-to-speech — to one all-in per-interaction figure, then compare against either alternative. A consolidated platform prices the agent's call minutes against the published voice rate card with the agent layer usage-billed, so the four legs decompose to verifiable lines; the AI voice agent TCO model runs the component math in full.
Is a flat-rate or committed-use pricing model ever the right answer?
Sometimes — a flat-rate wins only while the monthly volume sits inside an honestly-included allowance, and a committed-use discount wins when the volume is genuinely predictable. The CPaaS pricing models post works the arithmetic across three volume bands; evaluate the billing-model shape after decoding the billing units.
What does pass-through pricing mean, and why does it matter?
Pass-through pricing shows the upstream fee — a carrier SMS fee or a Meta WhatsApp per-message fee — and the platform's markup as two separate numbers. It matters because it is the strongest form of pricing transparency: the buyer sees exactly what the platform adds on top of the commodity cost, and any future repricing of either component is visible rather than encoded inside a flat number.
What should I check on a pricing page before a pilot starts?
Five rows, scored explicitly: a published per-destination rate card for every active channel, the rounding unit and the billable-attempt definition for voice, the component decomposition for AI agents, the pass-through markup separation for WhatsApp and carrier fees, and the idle-month cost at zero traffic. Devotel Orbit publishes all five on one pricing page, which is why the evaluation examples in this guide run on public figures.