Share with your CMO
AI in the contact center is exposing a structural flaw that better knowledge management never could: agents who escalate cases they already know how to resolve, not because they lack the answer but because they lack the authority and the protection to act on it. Ricardo Saltz Gulko’s ownership gap argument draws on Klarna’s public reversal after its AI-only support model degraded quality on complex cases, plus Gartner’s projection that AI cost-per-resolution will exceed offshore human agents by 2030, to argue that deflection rate is the wrong headline metric and Escalation Resolution Rate should replace it.
What this means for your business
The CMO who owns CX metrics sits at the center of this argument, and the exposure depends on one thing: whether your current dashboard rewards avoided contact or resolved ownership. If deflection rate is your headline KPI, you have likely been measuring a system that looks efficient while quietly accumulating customer effort on the other side. Organizations running high deflection numbers with flat or declining NPS scores are already living this contradiction, they just haven’t named it.
The core claim holds up. The incentive math Gulko describes is real and underappreciated: a frontline agent who owns a borderline decision and gets it wrong faces a coaching conversation or a quality mark, while an agent who escalates faces essentially nothing. Given that asymmetry, escalation isn’t timidity, it’s rational self-protection. No AI implementation changes that calculus unless decision rights are published and a good-faith guarantee is written into policy. The Klarna case is the clearest public evidence. The company’s CEO didn’t say the AI failed on volume; he said it failed on judgment-heavy, emotionally charged cases, which is precisely where unresolved ownership does the most damage to retention.
The analyst consensus Gulko cites, BCG on governance gaps, Deloitte on role redesign, Accenture on trust as an enabling condition, all converges on the same uncomfortable point: AI concentrates judgment rather than distributing it. The easy, structured volume gets automated, and what reaches a human is, by definition, the ambiguous residue where accountability matters most. That’s not an argument for slowing AI deployment. It’s an argument that deploying AI without redesigning decision rights produces a system where the highest-stakes cases are handled by people who have every incentive to pass them up.
The metric swap Gulko recommends, retiring deflection rate in favor of Escalation Resolution Rate and Time to Effective Escalation, is the decision this reframes for the CMO already defending a CX budget. If your board review shows deflection climbing and you can’t pair that with escalation resolution data, you’re presenting a number that measures cost avoidance and calling it customer experience. The CFO will eventually notice the gap between deflection performance and renewal rates. Better to reframe the dashboard on your terms before that conversation happens.
Concept deep-dive: Escalation Resolution Rate
Escalation Resolution Rate measures what fraction of cases that get handed to a higher-authority human are actually resolved at that level, without bouncing further or returning as repeat contacts. It exists because deflection rate, the older standard, only tracks whether a contact reached a human at all, not whether the human who received it had the authority and context to close it. Think of it as the difference between counting how many patients reached a doctor and counting how many left with a diagnosis. The business connection is direct: low Escalation Resolution Rate is a leading indicator of churn on complex accounts.
Based on reporting from The Next CX Metric Isn’t Deflection Rate. It’s Escalation Resolution Rate., originally published 2026-07-28 11:56:00.
