Vendors selling AI to support organisations tend to lead with deflection: the share of contacts the system handled without a person. It is an appealing number, and on its own it is close to meaningless. A contact that was deflected and came back three days later, angrier and with a new ticket, has not been saved. It has been moved.
The measure that holds up in a P&L review is cost per resolved case. It adds the cost of the AI system, the cost of the human time that still goes into escalations and reviews, and the cost of rework and repeat contacts, then divides by cases that were actually resolved. Computed the same way before and after a pilot, it is hard to argue with, and it will show honestly when a system looks cheap per contact but expensive per resolution.
Global technical support adds a further consideration. A follow-the-sun operation hands cases between regions across the day, and much waste hides in those handoffs, where the next shift restarts a case the previous one had half understood. A well-scoped AI use is to keep a clean running summary of each case, so the engineer picking it up in a new time zone starts informed. This is a hypothesis to test against the baseline, not a claim to assume.
Scope the pilot narrowly. Choose one queue and one product line with enough volume to measure, record the baseline over a sensible period, and keep a comparison group handled the old way. A pilot defined that way can be stopped or scaled on evidence, and its result is easy to explain to a finance partner.
Technical customers add a quality constraint. An engineer asking about a failing integration cares more about a correct answer than a fast one, and a confident wrong answer costs trust that takes a long time to rebuild. The escalation path to a human has to be quick and obvious, and wrong-answer rate should sit next to cost on the same dashboard.
Finally, plan for the team. When routine contacts are handled by software, the work left for people is harder and more varied, which calls for different skills and a different career path. Organisations that plan that shift get better outcomes than ones that discover it. A good question for any pilot sponsor: what is the cost per resolved case today, and who signs off the figure?
