The cost is not in the conversation count
Agentforce runs on consumption pricing, and the unit is the action, not the conversation. A standard or custom action is twenty Flex Credits flat, a voice action thirty, and every prompt call is billed on top at two, four or sixteen credits per two-thousand-token chunk depending on the model tier.
That last part is where estimates go wrong. Topic classification is a model call. Each reasoning step is a model call. Every Prompt Builder template is a model call. One customer-visible reply is routinely five or six billable events, not one.
Which means the bill is set by the architecture
Two agents that answer the same question can differ several-fold in cost, and the difference is entirely in how the actions were designed. An agent that chains six narrow actions where two would do pays for six. An agent that re-reasons on every turn pays for every turn.
None of that shows up in a demo, because a demo is one conversation. It shows up at fifty thousand, on an invoice, a month after go-live.
What the simulator returns
Build one average conversation — standard actions, voice actions, prompt calls, model tier, average prompt size — then set your monthly volume, retry rate and contracted credit price. It returns your monthly burn in dollars and credits, the cost per conversation and per thousand conversations, and a breakdown bar showing exactly which of the four drivers is spending your money.
The same architecture is then costed at ten, fifty, a hundred and five hundred thousand conversations a month, so you can see where the design stops being affordable.
Retries, and how fast the free tier goes
Retries are the silent multiplier. The tool models what happens if your retry rate drifts from three per cent to twelve, and prices the difference — the same output, for more money, every month until someone notices.
It also sizes the Foundations allocation honestly. The two hundred thousand included Flex Credits are a one-time per-term pool, not a monthly refill, and at a realistic action design that is often a few days of traffic rather than a runway.
What’s inside
- Per-conversation credit cost from your own action design, not a generic benchmark
- The four cost drivers split out: standard actions, voice actions, prompt calls, retries
- Prompt billing modelled properly — per 2,000-token chunk, at Basic, Standard or Advanced tier
- Monthly burn in dollars at your contracted Flex Credit rate
- The same architecture costed at 10k, 50k, 100k and 500k conversations a month
- What a retry rate drifting from 3% to 12% adds to the bill
- How long the 200,000 Foundations credits actually last at your volume
- The rate card it uses, stated in full, so you can check it against your contract
