Cost per conversation and cost per resolution
Cost per conversation divides by everything handled.
Cost per resolution divides only by what ends up resolved, and is more demanding. Some AI vendors charge per resolution, and when comparing you need both figures on the same basis.
Why it matters
It is the figure that translates support into the language of finance.
It shows what growth costs, lets channels be compared and helps decide what is worth automating. It also shows something uncomfortable. A conversation that should not exist, because the product is confusing or an article is missing, costs the same as a necessary one.
The formula, with an example
Cost per conversation is the total cost of support in the period divided by the conversations handled in that period.
The figures in the example are illustrative.
- Example. €18,000 a month across salaries, tools and overheads, and 3,000 conversations, give €6 per conversation.
- With automation. If an AI agent resolves 1,000 of those conversations at a cost of €800, the average falls, and it is worth calculating separately the cost of those handled by people.
What goes into the cost
The usual mistake is counting only salaries.
Four items go in.
- People. Salaries with employer costs, for those who answer and those who coordinate.
- Tools. Licences and usage-based charges.
- Training and quality. Time not spent answering.
- Overheads. The proportional share of space, equipment and services.
By channel and by request type
The average hides large differences.
A call takes one whole person for its duration, a chat lets one person handle several and a request resolved by an article takes nobody. Calculating it by channel and by topic shows where the spend is.
How to lower it without serving worse
Cost per conversation falls in three ways, and only the first also cuts the number of conversations.
- Avoid the request. Fix what causes it in the product or the documentation.
- Resolve it without a person. With self-service or an AI agent that looks up the data.
- Shorten it. With the customer’s context in view and prepared replies.
