When AI Fails: What Klarna, McDonald's and Ford Teach Us About Automation

Klarna replaced 700 customer service employees with AI — then brought them back. McDonald's tested AI-powered order-taking at more than 100 drive-through locations and shut the pilot down after a wave of viral failures. Ford rehired more than 350 experienced engineers because AI systems couldn't replicate the tacit knowledge their workforce had built up over years. These cases don't prove that AI doesn't work — but they do show where blanket automation hits its limits.

According to Robert Half, 32% of U.S. companies that eliminated roles due to AI have since rehired for similar positions. In an Orgvue survey of over 1,100 executives, 55% acknowledged that AI-related redundancy decisions turned out to be mistakes. Forrester Research predicts that by 2027, roughly half of all AI-attributed headcount reductions will be reversed.

What Drove the Reversals

The pattern is consistent across cases: AI performs well on the routine majority — but fails at the exceptions that actually determine customer satisfaction and output quality.

Klarna's AI handled 2.3 million customer conversations per month — impressive on paper. But in May 2025, CEO Sebastian Siemiatkowski acknowledged that the AI-only experience had damaged quality and customer empathy. Complex requests, emotional situations, and non-standard cases were handled markedly worse than by human agents. McDonald's was tripped up by accents, unusual customizations, and mid-order changes — viral failure videos left a lasting mark on brand perception. Ford found that its engineers carry years of tacit knowledge that is never documented anywhere, and which AI systems therefore cannot replicate.

Which Processes Actually Suit AI

The problem isn't AI — it's the assumption that every task is automatable. For any AI integration and automation to succeed, businesses need an honest process audit first: where do rule-based, well-documented workflows run with clearly defined response paths? That's where automation pays off. Where judgment, empathy, or tacit experience matter, humans remain essential.

• Suitable for AI: standard requests with defined response paths, data entry and processing, routine reporting and analysis
• Not suitable for AI: complaints requiring emotional sensitivity, specialist assessments based on tacit domain knowledge, ethically or legally sensitive decisions
• Hybrid model: AI handles initial processing and routine cases — humans manage exceptions and maintain the customer relationship

Deploying AI primarily as a cost-cutting tool — without first auditing the right processes — risks using it precisely where it causes the most damage.

What This Means for Your Business

The lesson from Klarna, McDonald's, and Ford isn't 'automate or don't automate.' It's: audit your processes first, then automate selectively. A structured technology assessment helps evaluate AI potential realistically — before investments are made and roles are eliminated.

• Identify which processes in your company are truly rule-based and well-documented
• Map where the tacit knowledge of experienced employees makes the real difference — and where knowledge transfer is needed before automation
• Plan for hybrid operation from day one: AI as a tool, humans as decision-makers for exceptions
• Measure AI success not just by efficiency gains, but also by customer satisfaction and quality metrics

Mid-sized companies have a structural advantage here: they are agile enough to course-correct quickly. Starting with a clear AI strategy and automation roadmap from the outset — rather than rushing to deploy — avoids the costly detour that Klarna and others had to make.

AI Integration for Mid-Sized Companies: Our Approach

This article was created with AI assistance and editorially reviewed.