Jan Carlzon's 'moments of truth' concept—brief interactions that define customer perception—now operates at machine speed. Support leaders must redesign which moments AI handles alone, which require human judgment, and which demand both.
By OperatorRadar Editorial
In 1987, SAS Airlines CEO Jan Carlzon introduced 'moments of truth'—the idea that every customer interaction, no matter how brief, shapes overall perception of a brand. He calculated that SAS handled 50,000 such moments daily. Each one was an opportunity to reinforce or damage trust. The concept became foundational to customer experience strategy: train staff, empower frontline teams, measure outcomes, repeat. The underlying assumption was human judgment at scale.
Carlzon's framework emerged during the service economy's maturation. Airlines, hotels, and financial institutions competed on reliability and courtesy. Technology existed (phone systems, reservation databases) but remained a tool in human hands. Support leaders focused on hiring, training, and incentive alignment—making sure the person answering the phone or handling a complaint had both authority and motivation to resolve it. Moments of truth were inherently local, synchronous, and dependent on individual competence and empathy.
Carlzon meant that customer loyalty is built or lost in small, discrete interactions—not in advertising or pricing alone. A gate agent's tone, a billing representative's willingness to waive a fee, a technician's follow-up call—these moments accumulate into a reputation. He argued that organizations should invest in frontline capability and decision-making authority because those moments happen too fast and too frequently for central control. The insight was about distributed responsibility and the economics of trust.
AI compresses moments of truth into milliseconds and multiplies them exponentially. A chatbot handles 500 support interactions per day per instance. An AI-powered routing system makes triage decisions before a human sees the ticket. Predictive systems flag churn risk before a customer complains. The moments still exist—but they now occur at machine speed, at scale, and often without human awareness. This creates three new operator problems: (1) AI systems make judgment calls that used to require human discretion; (2) customers expect instant resolution, collapsing the time available for empathy or explanation; (3) the failure mode shifts from 'one bad agent' to 'one bad model affecting thousands simultaneously.'
The core insight holds: small interactions compound into reputation. A single AI response that dismisses a customer's concern, misroutes a complex issue, or fails to escalate appropriately still damages trust—now at scale. Customers still judge brands on whether they feel heard and whether problems actually get solved. Frontline capability still matters; it's just that the 'frontline' now includes model behavior, prompt design, and escalation logic. Organizations that treat AI support as a cost center rather than a trust-building system still lose customers. The economics of distributed decision-making remain valid—you cannot centralize every judgment call.
The assumption that moments of truth are primarily about individual agent empathy and training no longer holds. You cannot train your way out of a poorly designed AI system. Hiring better support staff does not fix a chatbot that lacks context or escalation paths. The synchronous, human-paced interaction model no longer applies; customers now expect asynchronous, instant, and multi-channel moments of truth. The idea that one empowered person can resolve most moments is challenged by complexity—many issues now require coordination across systems, data, and expertise that no single agent (human or AI) possesses. Finally, the assumption that moments of truth are primarily positive or negative is outdated; they are now probabilistic and require continuous measurement and adjustment.
As a support leader, you must explicitly categorize your moments of truth by three criteria: (1) Judgment required—does this decision need human discretion, or can rules and data suffice? (2) Speed required—does the customer expect an instant response, or is a thoughtful delay acceptable? (3) Consequence—if the AI gets this wrong, what is the damage? Use this matrix to decide: AI-only (low judgment, high speed, low consequence); AI-with-human-review (medium judgment, medium speed, medium consequence); human-first-with-AI-assist (high judgment, low speed, high consequence). Then instrument each category: measure resolution rate, customer satisfaction, escalation rate, and time-to-resolution. Most support leaders over-automate high-judgment moments and under-automate low-judgment ones. The AI-era moment of truth is not 'did the AI sound nice?' but 'did the system route this to the right expertise, provide context, and know when to escalate?'
Take your three most common support issues. For each, ask: (1) What judgment does resolving this require? (2) How fast does the customer need an answer? (3) What happens if we get it wrong? (4) Is our current AI system designed to handle this moment of truth, or are we forcing it to? (5) If we're forcing it, what would we need to change—the AI, the process, or the expectation? Write your answers down. These are your redesign priorities.
Jan Carlzon
“Carlzon introduced the concept that every customer interaction is an opportunity to reinforce or damage brand perception. He famously stated that SAS handled approximately 50,000 such moments daily, each lasting 15 seconds on average.”
Jan Carlzon
Rethinking the Business: SAS Airlines Case Study
“Carlzon's operational philosophy emphasized empowering frontline staff to make decisions quickly and in the customer's interest, reducing the need for central approval and enabling faster resolution of moments of truth.”
OperatorRadar Research
AI Support Systems and Escalation Patterns
“Modern support leaders report that AI systems handle 60-80% of first-contact interactions but escalate 15-25% of those to human review, suggesting that judgment-heavy moments of truth remain difficult to fully automate without quality loss.”
Your moments of truth are unique to your product, customer base, and constraints. OperatorRadar's support design service helps you map, measure, and optimize them. Let's audit your current system and build a roadmap.
Request implementationNeed a custom implementation?
Have Ekofi Lyrae design and implement AI agents and automations.