What does each pricing model actually charge for?
Usage pricing charges for work done: an answered message, a started voice minute. WireDesk bills $0.02 per agent reply on chat and email and $0.20 per started voice minute, with no seats and no platform fee. If the queue is quiet, the bill is small; if a product incident triples your volume for a day, the bill moves with it — which is why every workspace has a hard monthly spend cap.
Per-resolution pricing charges a fixed fee, commonly around a dollar, each time the agent closes a conversation without a human. It is easy to reason about and it rewards a vendor for closing conversations, which is not always the same as answering them well — the definition of 'resolved' is the vendor's, and it is worth reading closely.
Per-seat pricing charges for the humans, monthly, whether or not the AI answers anything. Deflection makes the seats quieter but not cheaper, so savings only arrive when you eventually run fewer seats.
A mixed month, priced three ways
Take a mid-market support org: 25,000 agent replies on chat, 5,000 answered emails, and 8,000 started voice minutes in a month. At published rates that is $500 + $100 + $1,600 = $2,200.
Turn that into conversations. At roughly four replies per chat that is about 6,250 chat conversations; emails answered in one reply, 5,000; voice at about four minutes a call, 2,000 calls. Call it 13,250 conversations for $2,200 — near 17 cents each, including the voice.
The same month on a per-resolution plan: if the agent resolves 60% of those conversations without a person, that is roughly 7,950 resolutions. At a dollar each, about $7,950 — more than three times the usage bill for identical work. On a per-seat suite the arithmetic doesn't even reference the 13,250 conversations; it references how many people you employ, and the AI is an add-on line on top.
Why is voice priced per started minute?
Because that is what it costs to run. A live voice session holds a realtime model, speech in both directions and a telephony leg open for as long as the caller is talking, so the honest unit is time, not outcomes. Per started minute means a 2.2-minute call bills as three — the same rounding your telephony provider uses, and the same rounding our own calculators apply so the estimate never comes in under the invoice.
It also makes the cheapest optimisation obvious: answer on chat and email where the unit is two cents, and keep voice for the conversations that need a voice. Most teams that start on chat find the phone volume falls on its own.
What actually moves the number?
Three things, in order. Replies per conversation: an agent that answers in one well-sourced message costs a quarter of one that needs four. Grounding does most of this work — retrieval that returns the right passage the first time ends the conversation the first time.
Call length: at $0.20 a minute, an agent that answers a question in ninety seconds costs half of one that spends three minutes getting there. And deflection: every conversation your team handles instead is priced at their fully loaded hourly rate, which is where this stops being a line item and starts being a staffing question. The support ROI calculator puts both sides of that on one screen.
How do we stop the bill from surprising us?
Set the cap. Every workspace has a hard monthly spend cap — $50 by default for a pilot, raised to whatever finance signed off on — and when it is reached the agent stops answering billable conversations rather than running past it. You are notified at 80% and again at the cap, so it is a decision and not an accident.
Then watch a real month. Volume rates apply once you are at scale, and the number that matters for negotiating them is your own — conversations, replies per conversation, and voice minutes — which the dashboard shows from the first week.