Nearly one in five consumers who have used AI for customer service saw no benefit from the experience whatsoever, according to the Qualtrics 2026 Customer Experience Trends Report. That failure rate is almost four times higher than for AI use in general, and it is arriving at precisely the moment boards are demanding proof of return on investment. Welcome to enterprise AI’s reckoning.
The 95% Problem Is Not a Technology Problem
MIT’s Project NANDA, published in early 2026, puts the headline figure starkly: 95% of enterprise AI pilots fail to reach production or deliver measurable profit-and-loss impact. RAND Corporation data breaks this down further: 33.8% of projects are abandoned before production, 28.4% are completed but deliver no measurable value, and 18.1% simply cannot justify their costs. S&P Global survey data of over 1,000 enterprise respondents found 42% had abandoned AI projects outright.
The common diagnosis is instructive. Analysts from Gartner, Forrester, and MIT converge on the same root causes: 73% of failed projects lack clear executive alignment on success metrics; 68% underinvest in data governance; 61% treat AI as an IT initiative rather than a business transformation; and 56% lose active C-suite sponsorship within six months of launch.
These are not technical failures. They are organisational failures, governance failures, and integration failures. The technology rarely lets organisations down. The scaffolding around it almost always does.
Klarna Showed Us What Happens When You Optimise for Efficiency, Not Outcomes
No case study illustrates the current AI correction more vividly than Klarna. Between 2022 and 2024, the fintech giant replaced approximately 700 customer service workers with an AI assistant developed in partnership with OpenAI, at one point claiming the system was doing the work of 700 people. By mid-2025, the experiment had reversed: customer satisfaction had declined, operational hiccups had surfaced, and CEO Sebastian Siemiatkowski admitted publicly that the company focused too much on efficiency and cost, with the result being lower quality that was not sustainable.
Klarna is now rebuilding with a blended human and AI model, targeting students and remote workers, with a deliberate shift toward handling complex and emotionally charged interactions through humans, while AI manages the routine and the structured. This is precisely the augmentation model that delivers sustainable results: AI that works alongside human agents rather than replacing them.
The Klarna Effect, as it is being called across the industry, has given boards and procurement teams a compelling cautionary reference point. The race to contact centre headcount reduction via AI has slowed noticeably. The question organisations are now asking is not “how many agents can we eliminate?” but “how do we deploy AI in a way that is sustainable, measurable, and compliant?”
Governance and Integration Are the Differentiators in 2026
The enterprises that are seeing genuine AI returns are not the ones that moved fastest. They are the ones that built governance frameworks before scaling. In regulated verticals, the pattern is consistent.
In debt collections, platforms that embedded real-time compliance checks, consent tracking, auditable call logs, and rule-based controls from the outset are reporting recovery rate improvements of 10 to 15% alongside operational cost reductions of 40 to 60%. The global debt collection software market is forecast to grow from $4.8 billion in 2025 to $11.3 billion by 2033, driven largely by compliant-by-design AI tools. In financial services, Deloitte’s 2026 State of AI in the Enterprise report shows that firms with clear AI governance achieved the sharpest efficiency gains across all sectors surveyed.
Gartner’s warning about “agent washing” is relevant context here: only approximately 130 of the thousands of vendors currently describing themselves as agentic AI providers are considered substantive. The majority are rebranding existing chatbots, RPA tools, and IVR systems. Procurement teams in regulated industries are learning to ask harder questions. They want to see audit trails, escalation protocols, and integration depth before they sign. They have been burned once. Many have been burned twice.
So What?
The core problem is not that AI cannot deliver value in enterprise contact centre environments. It is that most AI deployments were architected around cost deflection rather than outcome delivery, governance was treated as a retrofit rather than a foundation, and integration into existing telephony and CRM infrastructure was an afterthought rather than a design principle.
What measurable, sustainable enterprise AI looks like in practice is purpose-built rather than generic, governance-first rather than compliance-bolted-on, designed to augment agents rather than replace them, and integrated deeply into existing workflows from day one so that every interaction is auditable and every outcome is measurable.
At calld.ai, we built for exactly this moment: enterprise AI for contact centres and outbound calling that delivers measurable outcomes within existing telephony and CRM infrastructure, with compliance and auditability at its foundation, not added later.