AI Support Cost per Resolved Ticket

An AI support cost calculator works out what a ticket genuinely costs to resolve once an AI agent sits in front of your human support team — and whether that arrangement actually saves money. Vendors sell AI support on a deflection or containment rate: the share of conversations the bot handles without a human. That number is seductive and almost always overstates the saving, because it quietly ignores three costs. This calculator puts all three back in.

The first hidden cost is that you pay the AI for every conversation, not just the ones it wins. If the bot handles 10,000 conversations and contains 45% of them, you still pay inference costs on all 10,000 — the 5,500 that escalate cost you the AI attempt *and* the full human handling afterwards. The calculator therefore charges AI spend across every conversation and human spend only on escalations, which is how the money actually flows. The second hidden cost is reopens. A ticket the bot marks resolved but the customer returns to a day later was never resolved; it just deferred a human touch and added a frustrated customer. The reopen rate applies human cost to that share of supposedly-contained tickets. The third is the baseline itself: the calculator computes what the same volume would have cost with humans only, so the saving is a real comparison rather than a vendor's framing.

Dividing total cost by tickets gives the cost per resolved ticket, which is the number worth managing. It exposes the break-even honestly: at low containment or high reopen rates, the AI layer can cost more than the human team it was meant to relieve, because you are paying for inference on every conversation and still paying humans for most of them. As containment climbs and reopens fall, the economics flip sharply — the model is highly sensitive to both. The source research is explicit that containment quality, not raw containment, is what matters, and that unsupported deflection claims should be avoided. Use this to pressure-test a vendor's business case with your own volumes: ask what containment rate they will commit to, measure your reopen rate after launch, and re-run the numbers. It models direct cost only and does not price CSAT damage, brand risk from bad answers, or the engineering time to build and maintain the integration — all of which belong in the decision.