Measure AI support-chat cost per resolved conversation

How should you count a resolved support conversation?

1. Define a conversation boundary and a follow-up window appropriate to the service. Decide in advance whether a return visit concerns the same case; never silently merge it with a newly resolved conversation.
2. Record model calls across all turns, retries and policy retrieval. Include the repeated conversation context in billed input tokens, and use total billed calls per original conversation as average attempts.
3. Label independently resolved cases, agent-assisted resolutions, escalations, abandonments and unresolved cases separately. Human takeover may resolve the case but is not an automated resolution.
4. Evaluate answers against the applicable support policy and verify that the customer’s problem was resolved. Compare cost using the same case mix and completed follow-up window; report return-contact and escalation rates alongside the estimate.

Write the acceptance rule first

A sample resolution rule can require a policy-correct answer, a completed customer action and no reopening within the chosen follow-up window. The example does not assert a real resolution rate or a universal window length.

Worked illustrative example

These numbers are assumptions chosen to demonstrate the calculation. They are not provider quotes, measured success rates or model benchmarks. Both options use $1 per million input tokens, $4 per million output tokens and an assumed 90% final success rate.

The example has 10000 original tasks, 1500 input tokens and 300 output tokens per model call. Option A uses 1.05 average calls per task; Option B uses 1.55. Everything else is held equal to isolate the cost of additional calls.

Monthly token spend = tasks × calls per task × (input tokens × input rate + output tokens × output rate) ÷ 1,000,000. Divide by accepted original tasks for cost per accepted result.

Load this exact example in the calculator

What the estimate leaves out

Search and retrieval services, CRM tools, telephony, human escalation, refunds and taxes are outside the token-only calculation. Do not include customer text or personal information in a share link; the calculator accepts numeric assumptions only.

Replace the assumptions with measured usage and the current rate for the exact provider route and conditions. Review source-linked API prices · Check retry accounting