Short answer
Ticket deflection counts contacts that never reached a person. It says nothing about whether the problem got solved. Measure it by intent and tower against a signed baseline, pair it with confirmed resolution and repeat contact, and check satisfaction. Report cost separately, because a ticket that doesn't arrive isn't yet money saved.
Last reviewed Nov 2, 2026
Two numbers that get confused
Zendesk draws the line clearly, and it's a useful working definition even for teams on another platform. "Ticket deflection routes customers away from human support queues," while "ticket resolution means the issue is fully completed during the interaction" (Zendesk, Jul 8, 2026). Its example is an IT password flow: of 1,500 avoided tickets, 1,200 employees reset their passwords, 200 contacted IT within a day, and 100 abandoned the flow. A dashboard reports 1,500. The honest number is 1,200.
The gap is familiar. A Gartner survey of 5,728 customers found 14% of customer service issues were fully resolved in self-service, and the rate was only 36% for issues customers called "very simple" (Gartner, reported by CX Today, Aug 19, 2024). That study covers customer service rather than internal IT, and it dates from before the current generation of agents, so don't apply the number to your desk. Apply the lesson: self-reported success and system-reported success can differ a lot.
Employee preference matters too. In a Gartner survey of more than 5,000 digital workers, chatbot support ranked sixth, with fewer than 10% choosing it as their most preferred option (CNBC, Dec 1, 2024). People who don't trust the channel route around it, and that demand doesn't disappear.
Read vendor numbers with their denominators
Published figures use different units. ServiceNow says Now Assist "is already resolving 98% of initial touchpoints" at the City of Raleigh, and that its Autonomous Workforce handles more than 90% of ServiceNow's own employee IT requests (ServiceNow, Feb 26, 2026). Both are vendor-reported, and each has its own denominator. A touchpoint isn't a ticket, and one company's request mix isn't yours. Ask any vendor, including us, for the unit, the period and the exclusions.
A method that survives review
Define the scope first. Write down what L1, L2, L3 and L4 mean in your organization, by tower. Teams use these labels differently, so an undefined "L1–L4" number can't be compared with anything.
Baseline by intent. Pull at least two full quarters of tickets and classify by intent: status, how-to, access, policy, fault. Include email and chat inboxes, or you'll count a channel shift as a reduction. Have finance sign the baseline and the cost per ticket before anything goes live.
Count three outcomes per intent. Report deflected contacts, confirmed resolutions and repeat contacts. Zendesk's formula works: confirmed resolved interactions divided by total AI or self-service interactions, times 100. Confirmation signals include completed workflow steps and customer feedback, plus no repeat contact within a window such as 24 to 48 hours.
Hunt for shadow demand. Ask managers and floor leads where people go when the channel fails. Direct messages to known experts and walk-ups won't show on your dashboard, and they're often the first sign that deflection has become avoidance.
Hold satisfaction flat or better. Compare agent-handled and human-handled requests with the same survey. If satisfaction falls, stop widening scope until you know why.
Normalize for volume. Headcount, acquisitions, seasonal peaks and releases all move ticket counts. Report per employee served, and note any events in the period.
From tickets to money
This is where most business cases go wrong. The Hackett Group (Mar 24, 2026) found 72% of GBS leaders cite misalignment between expected and actual AI benefits as a significant concern, and loose unit conversion is part of how that gap opens.
Keep two lines. Capacity freed is hours returned to the tower. Hard savings are vendor fees that fell, roles not hired, or contracts that changed. Only the second reaches the budget. Never add them together, and never use a ticket percentage as evidence of a cost percentage unless you've done the calculation for that client and that contract.
What to report each month
Keep it to one page per tower: the ten largest intents, deflected, confirmed resolved, repeat contact, satisfaction, and agent cost against plan. Add a short list of intents you've retired from automation. A list of things you stopped does more for credibility than another rising chart.
How GCC Now helps
GCC Now™ is the ServiceNow Accelerator by ANSR × inMorphis. It helps global capability centers (GCCs) and global business services (GBS) organizations run their ServiceNow services more efficiently and at lower vendor cost. A Deflection Sprint applies this method to your top intents, on your data. McDonald's and Tricon Energy: ~40% reduction in L1–L4 tickets through deflection. That is a ticket figure from those case studies, and a new client's number comes from its own baseline.