When Klarna announced in 2024 that it was replacing its customer service with an AI chatbot, customer satisfaction dropped sharply. Six months later, the company announced a strategic reversal and began rehiring human support staff.1 "Klarna Walks Back AI Overhaul, Rehires Staff After Customer Service Backlash"

What happened? The failure of many AI automation projects is not due to the technology or the AI layer, but to the interface between automation and human intervention. The course correction introduced a hybrid model: simple support cases handled by AI, while more complex and sensitive cases were managed by human agents.

Another example: Air Canada's chatbot promised a customer a discount that did not actually exist in the company's booking terms. The customer sued, won before a consumer protection tribunal, and the airline was forced to pay under protest.2 "Air Canada Must Honour Refund Policy Invented by Its Chatbot", TechHQ

Cost Reduction Over Customer Value Damages Trust

When deploying AI in internal processes such as customer service, the design of the interface between AI and human is the core of success. Nearly one in five customers reported seeing no benefit from using AI-powered services — four times the rate seen with general AI use.3 "AI-Powered Customer Service Fails at Four Times the Rate of Other Tasks", Qualtrics

Most companies deploy AI to cut costs rather than to solve customer problems — and customers perceive this. AI should therefore deliver a clear customer benefit: Qualtrics' Customer Service Report 2026 also shows that customers are more satisfied and have greater trust when they choose a company for its good customer service and good products.

Human in the Loop

In my work on user interfaces for safety-critical systems, the goal was to avoid human operational errors — and where that was not possible, to reduce them and simplify their correction. User interfaces are the window to operational reality in control centres for aviation and rail; the consequences of inadequate design in these sectors can be fatal. Recent examples include deaths linked to self-driving cars, where drivers became inattentive due to missing warnings and over-trusted partially autonomous systems.

Humans require continuous cognitive stimulation to maintain situational awareness. When this is absent for too long, attention drops within minutes and error rates rise. Designing systems that keep humans in the loop requires a holistic perspective. Interface design alone is not sufficient.

These concepts transfer directly to non-safety-critical business processes. The design of AI-supported systems must follow a design process that first examines processes with clear handover and check-in points, and only then addresses the concrete design of interactions and user interfaces.

The ROI of AI-Supported Process Optimisation

Not everything that can be optimised with AI operates cost-efficiently. In the RAND Corporation study,4 "The Root Causes of Failure for Artificial Intelligence Projects" 84% of respondents cited misguided management directives as the reason for the failure of AI projects in companies.

Only 5% of AI initiatives in companies ever reach productive implementation with measurable benefit.5 "The Gen AI Divide: State of AI in Business 2025" The study also shows that approximately 70% of AI budgets are consumed by sales and marketing, because their results are easier to measure and therefore have the greatest visibility in companies — not because they deliver the best outcomes.

Yet interestingly, employee-driven AI adoption in the back office can yield a higher ROI than large-scale, costly initiatives. In the best case, it is possible to increase customer satisfaction and reduce reliance on outsourced functions such as customer service, without reducing headcount.

These are genuinely good news for exhausted, sceptical, and frustrated teams who now face yet another promise that this time AI will "definitely make your life easier".

The Human Role Remains Central

AI is fundamentally changing how we think and make judgements, and the effects of these processes are not entirely positive. Research is actively examining the phenomenon of "cognitive surrender" — the uncritical adoption of AI outputs.6 "Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender", SSRN

When used correctly, AI can extend our capabilities and increase business success — and that short-sighted cost-cutting measures can often produce the opposite effect. Working life is changing fundamentally right now, but hasn't that always been the case? Entire professions have disappeared from our world. What matters is how we shape this transition.