How much does an AI support agent cost per user?
A conversational agent answering customer questions against your help centre and ticket history.
The short answer
On Claude Sonnet 5, at 6,000 input and 350 output tokens per call. Six thousand input tokens assumes a system prompt, four or five retrieved help articles, and a few turns of conversation history carried forward. The output is short because a good support answer is short.
Your numbers
Now put your own in
The fields open with the profile above. Change them to match your feature and the read updates as you type.
Your numbers
Rough numbers are fine. You are checking whether you are near the line, not being exact.
The usual right answer for production features. Near-frontier quality at a fraction of the cost.
Everything you send it
What it sends back
How often one user uses it
What it costs you
What drives the cost
Input tokens, by a wide margin. Every turn resends the whole conversation plus the retrieved articles, so a six-turn conversation pays for the context six times. Most teams model the first turn and are surprised by the bill for the sixth.
Where the estimate goes wrong
The count of calls per user is the number that moves. Support volume is not evenly distributed: a small fraction of users open most of the conversations, and those users are also the ones who go ten turns deep. Model your power users rather than your median, or the estimate will be low by a multiple rather than a percentage.
What this does not price
This prices the conversation and not the escalation. An agent that answers eight questions and hands the ninth to a person has not removed that ninth ticket, and the handover itself is work. Where the business case rests on deflection rate, the deflection rate is the number to measure, not the token cost.
Other features people cost
Next step
If customer support agent is on your roadmap and the number here made you pause, that is exactly the conversation worth having before an engineer starts.