Editorial
Guide
Post-Acquisition Guide: How to Bring Two Digital Experiences Together
The Gist
- Why do agents escalate cases they already know how to resolve? Because escalation is really about authority and backing, not missing information — AI just removes the excuse that used to disguise it.
- What makes AI expose the ownership gap instead of closing it? Once AI supplies a confident, policy-grounded answer, "I wasn't sure" stops working as cover, leaving the real question — who owns this decision — impossible to dodge.
- What should CX leaders change first? Retire deflection rate as the headline KPI, replace it with Time to Effective Escalation, Outcome Certainty and Escalation Resolution Rate, and publish explicit decision rights that back good-faith judgment calls.
For a decade, enterprise technology companies have treated the frontline's hesitation as an information problem. The fix was always more: more training, more knowledge bases, better search, and now AI that hands an agent a confident answer in seconds. The logic seemed airtight — if people escalate because they don't know the answer, give them the answer.
But what if the answer was never the problem?
That is the uncomfortable question worth sitting with as AI floods the contact center. Because the tool everyone expects to reduce escalations may turn out to do something more interesting: it strips away the last excuse, and exposes a gap that information was quietly hiding all along.
Why Agents Escalate Cases They Already Know How to Resolve
Here is the pattern anyone who has worked a service floor will recognize. A case lands that falls slightly outside the usual shape. The agent already knows the policy. They know the likely answer. They probably know what they'd do if the call were entirely theirs.
And they escalate anyway.
We read that as a skill gap. Often it isn't. The agent isn't asking what is the answer — they're asking three quieter questions: Can I make this call? If I do, will it be supported later? If the outcome is challenged, who carries it? Those questions never appear on a process map or a dashboard, yet they shape an enormous amount of day-to-day behavior.
What looks like uncertainty about the answer is often uncertainty about ownership. And no amount of information resolves that, because information was never the thing in short supply.
The obvious reading is that customers want their issue solved. But some problems can't be solved on the spot — not by AI, and not always by the human they're escalated to. So customers watch something subtler: how you pursue it, whether you ask the right questions, the customer experience itself, and whether you capture it properly. When the fix isn't available, the process is the resolution.
What Matters Here: Why Do Agents Escalate Cases They Already Know How to Resolve?
Agents escalate not because they lack the answer but because they're unsure whether they have authority to act on it and whether they'll be backed if the outcome is challenged. This makes the handoff a question of decision rights, not knowledge.
How AI Removes the Cover for Avoiding Ownership Decisions
For years, the ownership gap had cover. “I wasn't sure” was a socially acceptable reason to pass a case up, and it was impossible to disprove. The handoff looked like a knowledge problem, so we kept investing against it.
AI removes the cover story. Once the system has already surfaced a correct, policy-grounded answer with a confidence score attached, “I wasn't sure” stops being available. What's left standing in the open is the real question: am I allowed to own this, and will anyone back me if it goes wrong? How can I get hold of the solution makers? Marketing, Product, R&D for example.
This is the counterintuitive part. AI doesn't create the ownership gap — it removes its camouflage. The handoffs that used to look like hesitation now plainly look like what they always were: rational risk avoidance. And that reframes AI from a tool that should cut escalations into a forcing function that makes the ownership question impossible to dodge.
The structural shift is what makes this urgent, and the major analyst houses now describe it in strikingly similar terms:
- BCG warns that businesses often fail to assign responsibility for autonomous AI agents, so no one holds clear accountability when something goes wrong.
- Deloitte frames the new model as one where AI handles routine, data-intensive work while humans focus on judgment, empathy and complex problem-solving — but only for organizations that redesign roles, metrics and governance for that collaboration.
- And Accenture argues the entire shift rests on trust: systems are only ever as autonomous as they are trustworthy, and someone must remain answerable for what they do.
Notice what every one of these has in common. AI isn't removing human judgment from the loop — it's concentrating it. The easy, structured volume gets automated, and what reaches a human is, by definition, the ambiguous, high-stakes, judgment-heavy residue — exactly the cases where ownership matters most and is least clear.
The economics seal it: Gartner projects that by 2030 the cost per resolution for generative AI in service will exceed that of many offshore human agents. If AI makes judgment faster but not cheaper, the human decision becomes more valuable, not less — and who is empowered to make it becomes the binding constraint.
What Matters Here: How Does AI Remove the Cover for Escalation Avoidance?
Once AI surfaces a confident, policy-grounded answer, "I wasn't sure" no longer excuses a handoff, exposing that the real hesitation is about accountability, not information.
Escalation Failure: How Handoffs Mask Missing Decision Rights
This is the heart of what I have elsewhere called escalation failure. Customers aren't rejecting AI; they're rejecting AI that obscures ownership and stalls human judgment when the stakes are high. And as I noted in The Economics of Trust in AI-Driven CX, the winning model is orchestration, not replacement — one where clear accountability isn't optional but foundational.
Deploying an automated tool does not transfer accountability to the tool. Someone in the organization always owns the outcome. The only real question is whether that ownership is designed — assigned clearly, in advance — or left to be discovered after something goes wrong.
What Matters Here: What Is Escalation Failure and Why Does It Persist?
Escalation failure happens when handoffs obscure who owns a decision. Deploying an automated tool doesn't transfer accountability to it — ownership still has to be assigned, or it gets discovered only after something goes wrong.
Why Escalation Is the Rational Choice for Frontline Agents
It's tempting to call hesitant agents timid. They aren't. They're doing accurate math. Owning a borderline decision that goes wrong can cost an agent — a coaching conversation, a quality ding, a mark against their numbers. Escalating costs them almost nothing. Faced with that asymmetry, escalation is the rational choice, every time. People aren't failing to understand the policy; they're understanding the incentive structure perfectly.
And the evidence says fixing this pays. McKinsey's research on employee engagement and empowerment found that organizations that actively listen to and act on recommendations from frontline employees are far more likely than their peers to implement new and better ways of working. Its work on the psychological needs of employees documents that giving frontline workers genuine discretion over appropriate decisions — like accepting a return or issuing a voucher — is one of the levers that raises both wellbeing and performance. Discretion isn't a soft perk. It's an operational input.
This is also where the AI conversation quietly turns into an ownership conversation. In its analysis of managing AI agents, BCG argues that because agentic systems sit between tool and colleague, organizations must “upgrade governance and redesign decision rights” — and reports that most leading organizations already expect their decision-making rights to change. Redesigning who can decide is not a side effect of deploying AI. It is the work.
What Matters Here: Why Do Agents Choose Escalation Over Ownership?
Owning a borderline call that goes wrong can cost an agent a coaching conversation or a quality ding, while escalating costs almost nothing — given that asymmetry, escalation is the rational choice, not a sign of hesitancy.
Why Deflection Rate Hides the Ownership Gap
There's one more reason the ownership gap survives every tooling upgrade: we measure the wrong thing. A dashboard optimized for deflection rate — how many contacts we kept away from a human — will look healthy even as customers are bounced through handoffs and effort quietly climbs. Deflection rewards the very behavior that escalation failure depends on.
The corrective is to stop asking “how many contacts did we deflect?” and start asking “how reliably did we resolve outcomes at the right level of accountability?” That shift points to metrics that actually surface the ownership gap — Time to Effective Escalation, Outcome Certainty and Escalation Resolution Rate — treating them not as CX vanity numbers but as financial levers. You cannot manage an ownership problem you have deliberately designed your dashboard not to see.
From Automation to Intelligence: How Leading Teams Are Rethinking Operations
Register
Content Leaders Collective: When Your Documentation Tools Can't Keep Up
Register
The Next Era of Enterprise Experience: AI, Automation and What It Means for CX Leaders
Register
How to Win the War for Agentic Citations: The AEO Playbook You Need Now
Register
What Matters Here: Why Does Tracking Deflection Rate Hide the Ownership Gap?
A dashboard built around deflection rate can look healthy even as handoffs and effort quietly climb, because it rewards keeping contacts away from a human rather than measuring whether ownership was resolved at the right level.
Related Article: The Measurement Crisis Holding Customer Service Back
How to Design Decision Rights Into AI-Assisted Escalations
The fix isn't more knowledge, and it isn't more confident AI. It's making ownership legible: clear decision rights, paired with a credible promise that good-faith judgment will be backed even when the outcome isn't perfect. The specifics differ for B2C and B2B, because the stakes and the failure modes differ.
Five B2C Steps to Build Ownership Into Escalation Design
-
Publish decision rights at the point of action. Define explicit thresholds an agent can own outright — refund up to X, waive a fee under Y conditions, issue a credit within Z — and surface them in the workspace alongside the AI's recommended answer. The AI proposes; the agent is pre-authorized to dispose.
-
Create a good-faith decision guarantee. Make it policy, in writing, that a defensible judgment call made within published limits will not be penalized if the outcome disappoints. Remove the personal downside and you remove the reason to escalate reflexively.
-
Re-engineer escalation triggers around stakes, not friction. Escalate on emotion, explicit human requests and genuinely high-stakes or out-of-policy cases — not at the first sign of mild confusion the agent could resolve.
-
Swap deflection for ownership metrics. Retire deflection rate as a headline KPI. Track Time to Effective Escalation, Escalation Resolution Rate and repeat-contact rate so the dashboard rewards resolved ownership, not avoided contact.
-
Treat the knowledge base as accountable infrastructure. Assign a named owner to every policy the AI can cite, and audit it on a cycle so the source of truth is current and someone is responsible for it.
Five B2B Steps to Make Escalation Ownership Contractual
-
Map decision rights onto account tiers and contracts. An escalation can disrupt billing, integration, SLAs or compliance. Define who owns which decision by account value and risk — and ensure a named relationship manager or SME, not a queue, owns the ambiguous exception.
-
Build the handover object. When a case escalates, package conversation history, customer context and transaction state so the receiving human inherits full context. Escalation failure in B2B is often a context failure; the human re-asks, the customer re-explains, and trust erodes.
-
Put escalation paths in the SLA. Specify the maximum Time to Effective Escalation and the named role accountable for high-stakes outcomes. Make ownership a contractual commitment the customer can see, not an internal hope.
-
Stand up an AI governance boundary. Decide, in advance and in writing, what the AI may resolve autonomously, what requires human sign-off, and where the audit trail lives. Clear boundaries are what let an agent act without seeking reassurance.
-
Tie escalation quality to retention economics. Report escalation resolution against renewal and expansion data so ownership investment is visible as protected revenue, not overhead.
Key Lessons on AI, Escalations and Decision Ownership
The following table highlights the most important lessons, actions and strategic considerations emerging from how AI is exposing the ownership gap behind contact center escalations.
| Key Area | What Happened | Why It Matters | Recommended Action |
|---|---|---|---|
| Escalation drivers | Agents escalate cases they already know how to resolve because they're unsure if they're authorized to act or will be backed if challenged. | Treating this as a knowledge gap misdiagnoses the problem and wastes investment on more training and better answers. | Reframe escalation analysis around decision rights, not information gaps. |
| AI's effect on cover stories | Once AI supplies a confident, policy-grounded answer, "I wasn't sure" no longer works as an excuse to hand off a case. | AI doesn't create the ownership gap, it removes the camouflage that let it go unaddressed for years. | Use rising escalations post-AI deployment as a signal to audit decision rights, not a signal the AI failed. |
| Analyst consensus | BCG, Deloitte and Accenture independently describe the same structural shift: AI absorbs routine volume while judgment-heavy, ambiguous cases concentrate with humans. | The human decision left standing becomes higher-stakes and more valuable, not less relevant. | Redesign roles, metrics and governance around the judgment calls AI can't absorb. |
| Frontline incentives | Owning a borderline call that goes wrong can cost an agent a coaching conversation or quality ding; escalating costs almost nothing. | Given that asymmetry, escalation is the rational choice for agents, not a sign of hesitancy or poor skill. | Publish explicit decision-rights thresholds and back good-faith judgment calls in writing. |
| Measurement | Deflection-rate dashboards can look healthy even as handoffs and customer effort quietly rise. | Deflection rewards avoiding contact, not resolving ownership at the right level. | Replace deflection rate with Time to Effective Escalation, Outcome Certainty and Escalation Resolution Rate. |
| Real-world evidence | Klarna scaled back an AI-only support model after its CEO conceded it produced lower quality on complex, emotional cases. | Value migrates to exactly the decisions where a human must be empowered to own the outcome. | Design escalation and payout steps into AI workflows from the start, as BCG's insurance example and Oracle's guardrail-based governance illustrate. |
What Matters Here: What Concrete Steps Build Ownership Into Escalation Design?
Effective fixes pair published decision-rights thresholds with a good-faith decision guarantee for B2C, and contractual escalation SLAs with a defined AI governance boundary for B2B.
How Klarna, BCG and Oracle Handle AI Escalation Ownership in Practice
The pattern is already visible in named deployments. Klarna is the most public example: its OpenAI-built assistant handled two-thirds of chats and the work of hundreds of agents, yet by 2025 the company was rehiring humans, with its CEO conceding that a cost-first, AI-only model produced “lower quality” on the complex, emotionally charged cases. The AI absorbed the volume; what it couldn't absorb was the judgment — and the ownership of that judgment had to be handed back to people. The lesson isn't “AI failed.” It's that the value migrated to exactly the decisions where someone has to be empowered to own the outcome.
Contrast that with deployments that design ownership in from the start. BCG describes insurance operations where AI agents handle claims end to end — document validation, triage — but with an explicit escalation-and-payout step routed to a human, an approach associated with materially higher net promoter scores. The escalation isn't treated as a failure of the AI; it is a designed, owned step in the workflow.
And on governance, Oracle has taken the approach of embedding AI actions within defined guardrails, requiring human confirmation for high-impact outputs and maintaining full audit trails — treating accountability as a built-in product feature rather than a compliance afterthought. That is what “ownership by design” looks like in a real system: the boundaries of what the AI may decide alone, and where a named human takes over, are settled in advance.
What Matters Here: How Do Klarna, BCG and Oracle Illustrate Escalation Ownership?
Klarna's reversal after an AI-only model showed judgment couldn't be automated away, while BCG's insurance workflow and Oracle's guardrail-based governance show ownership designed in from the start rather than discovered after a failure.
Who Owns the Judgment When AI Handles the Answer
The tool we expected to quietly reduce escalations may be the thing that finally forces the conversation no knowledge base ever could: who owns the judgment? AI didn't break ownership. It just turned the lights on.
Until decision rights are clear and good-faith judgment is genuinely backed, every new tool will do exactly what the last one did — help people find the right answer faster, and hand it off anyway.
This article was inspired by my colleague Masako Takei, whose reflection “Escalations Sometimes Signal Missing Ownership, Not Poor Skill” sparked the central idea here — the observation that when a case falls outside the usual pattern, the real uncertainty is often not about the answer, but about who owns the judgment.
Learn how you can join our contributor community.