Ask most teams how they chose a channel for their agent and you will hear a price. "SMS is cheap." "Voice is expensive." That is like choosing a vehicle by comparing the price of fuel per mile and ignoring that one of them is a bicycle and the other is a truck.
The unit that matters is not the message. It is the completed outcome: the appointment confirmed, the invoice paid, the question answered. A channel that is cheap per message and fails half the time is not cheap. A channel that is expensive per attempt and works the first time might be the bargain.
This post is a framework for thinking about that. It deliberately uses relative terms, not prices, because real rates vary by provider, country, and volume, and they change. The structure of the trade-off is what lasts.
The five drivers of cost per outcome
1. Direct delivery cost. What it costs to send or place one attempt. Voice is billed by duration, so a long call costs more than a short one. Text channels tend to cost by message or segment. Email is usually the cheapest to send, but "cheap to send" is not the same as "cheap to deliver."
2. Attention value. What the recipient gives you. A phone call demands real-time attention, which is why the whitepaper calls voice the highest-trust, highest-attention channel. A text gets glanced at. An email waits in a pile. Attention is a feature you pay for: higher attention means higher completion on urgent or high-stakes tasks.
3. Failure cost. What it costs when the attempt does not land. A missed voicemail is a lost attempt. A text flagged as spam may fail silently, with the agent logging "sent" while the recipient sees nothing. An email in the spam folder looks identical to an email nobody read. Failure is expensive not just in wasted attempts but in decisions made on the false belief that the message arrived.
4. Retries and follow-ups. Every channel needs a plan for non-response. The cost of the plan is multiplied by how often you need it. A channel with low response rates needs more nudges per outcome, and each nudge can damage reputation if you overdo it.
5. Compliance and setup overhead. The fixed and ongoing cost of being allowed to send. US SMS needs brand and campaign registration, which the whitepaper describes as taking weeks. WhatsApp has template approval and session-window rules. Email needs domain authentication and a warm-up period. Voice needs consent handling and, for AI callers, disclosure. These are mostly up-front costs, but they shape which channels are even available on day one.
A comparison table
Everything below is relative: how each channel tends to compare to the others, not a quote.
| Driver | Voice | SMS | ||
|---|---|---|---|---|
| Direct cost per attempt | High | Low to medium | Low to medium | Very low |
| Recipient attention | Very high | High | High | Low to medium |
| Speed to a response | Immediate | Minutes | Minutes | Hours to days |
| Silent failure risk | Medium | Medium to high | Low to medium | High |
| Best content type | Urgent, nuanced, emotional | Short, time-bound | Rich, conversational | Detailed, structured, formal |
| Retry tolerance | Low (calls annoy) | Medium | Medium | Medium to high |
| Compliance and setup overhead | Medium | High | Medium to high | Medium |
| Reputation fragility | High | High | Medium | High |
Read the table by column, not by row. No channel wins every row, which is exactly why the interesting question is the match between channel and task.
Matching channel to task
Think of the cost of the outcome as three numbers multiplied together: the cost per attempt, the number of attempts you need, and the cost of being wrong.
Notice the pattern: the answer is often a sequence, not a single channel. Cheap and low attention first, expensive and high attention only when the stakes or silence justify it. This is the logic behind the whitepaper's section on intelligent outbound routing: at 23:45 in the recipient's timezone, do not call, send a text; if the message is long, switch to email; if urgency is high, prefer text over email. The agent's intent stays the same and the channel decision becomes data-driven.
The hidden multiplier: fragmentation
There is one cost that appears in none of the table rows: the cost of running several channels at once. In the typical setup, each channel has its own vendor, its own identity, its own billing, its own consent tracking, and no shared memory. A conversation that starts as a call and continues as a text restarts from zero.
That is a tax on every outcome that crosses channels, and cross-channel sequences are precisely the cheap-first, escalate-later pattern the framework recommends. If switching channels is expensive to build and loses context, teams default to one channel and pay the failure cost of using the wrong one.
A shared session layer, one identity and one thread across channels, removes that tax. That is the argument the whitepaper makes for a channel abstraction layer, and it is the kind of problem Dial handles at the infrastructure layer so your agent does not have to.
How to actually run the numbers
You do not need exact prices to use this. Do this exercise for your top three agent workflows:
You will often find that the expensive channel is cheaper per outcome for high-stakes workflows, and the cheap channel is better for routine ones. You will also find that your biggest cost is not any channel, it is the cases where the agent believed a message landed and it did not.
The takeaway
Stop comparing channels by sticker price. Compare them by cost per completed outcome, which depends on attention, failure, retries, and compliance as much as on delivery. Design sequences that start cheap and escalate with the stakes, and make sure your infrastructure lets the conversation survive the switch.