Deflection is the wrong goal, outcomes are

By Faraha Jebi · June 4, 2026 · 2 min read

Key takeaways

• Deflection counts tickets the AI handled, not whether the answer was right or the customer stayed.
• The four metrics the business actually buys are handle time, first response, CSAT, and recoverability.
• Put outcomes in the contract and treat deflection as a diagnostic, not a goal.

Deflection rate is the metric every AI-for-support vendor leads with. It is also the metric most likely to fool a buyer into spending six figures on a system that does not change anything that matters. Deflection counts the tickets the AI handled. It does not count whether the answer was right, whether the customer came back, whether the queue got faster, or whether the cost line on next quarter’s P&L moved. It is an activity metric in a job that pays for outcomes. Vendors lead with it because it scales fastest on a slide. Buyers chase it because a bonus is tied to a target. The result is a market that celebrates 60 percent deflection and never asks whether that number included the customer who churned silently after the bot misrouted them.

What deflection hides

Deflection is the share of incoming tickets that closed without a human. It is easy to count, easy to dashboard, easy to celebrate, and that ease is the problem. A 60 percent deflection rate hides four things the business cares about more. Whether the answer was right: a confident wrong answer the customer accepted still counts as deflection, even if it triggers a chargeback weeks later. Whether the customer recovered: if the AI got it wrong and the customer left instead of escalating, that is a deflection on the report and a churn in the cohort. Whether handle time and first response moved: deflection is silent on the benchmarks that govern SLA attainment. And whether CSAT held: Zendesk’s own 2026 CX Trends data puts AI-handled CSAT at 4.10 out of 5 against 4.30 for human agents, a gap that compounds across millions of interactions.

The outcomes that move the business

If deflection is the activity metric, four outcome metrics are what the business actually buys. Average handle time drives cost-per-contact, the line finance reads; the Auralis cohort sees roughly 30 percent lower AHT in steady state. First response time is how customers judge urgency; Auralis customers see roughly 35 percent faster FRT. CSAT is the only metric customers fill out themselves, and the hybrid path is engineered to recover the customer rather than just close the ticket. Recoverability is the one most vendors never measure: when the AI is wrong, does the system catch it before the customer leaves? Auralis Audit instruments it as a first-class signal rather than leaving it off the dashboard.

Why the market still leads with deflection

The honest answer is that the category grew up around chatbots between 2018 and 2022, when “the bot did it without a human” was a real technical achievement, and the metric outlived its moment because the dashboards did. Forrester’s April 2026 analysis documents the gap directly: vendors sell autonomy, but “there is zero appetite in this market for fully autonomous agentic applications.” Buyers know the metric is hollow; they buy it because vendors quote it, and vendors quote it because buyers ask. The way out is to change what you ask for. Put handle time, first response, CSAT, and recoverability in the contract, and treat deflection as a diagnostic, not a goal.