Work & Career
Something we have been seeing more of, especially with the advent of AI, is the increasing strategy by companies to outsource the “less critical” aspects of the job to AI or cheaper labour markets. Every boardroom conversation about artificial intelligence and offshoring eventually lands on the exact same slide: Cost per interaction. The math seems undeniable. Why pay $13.50 for a local human representative when an offshore team or an Agentic AI chatbot can handle the query for under $2.00?
The logic is seductive. Artificial intelligence has matured to the point where it can plausibly handle a customer conversation. Labour in lower-cost markets is a fraction of the price at home. So the “less critical” parts of the business, the parts that supposedly do not require deep institutional knowledge, get handed off. First, it was IT and back-office desk functions. That migration has been underway for two decades. What is new, and what should concern anyone who cares about the long-term health of a brand, is that the functions now being outsourced and automated are precisely the ones that have always lived or died on human interaction: customer care, complaints handling, and increasingly parts of HR and recruitment.
Done well, this should be a genuine win. Faster resolution, round-the-clock availability, a happier customer, and a lower cost to serve. The promise is real. The execution, on the current evidence, is mostly failing.
The numbers are not flattering
Start with the headline finding that should give every chief financial officer pause. According to research published by Qualtrics in October 2025, nearly one in five consumers who used AI for customer service got no benefit at all from the experience, a failure rate roughly four times higher than AI use in general. Consumers ranked AI for customer service among the worst applications of the technology for convenience, time savings and usefulness. Only “building an AI assistant” scored lower.
A separate national survey published in December 2025 found that three-quarters of consumers were left frustrated by AI customer service through 2025, reporting more loops, more dead ends, more repeat explanations, and ultimately declining trust. Crucially, customers prioritise outcomes over speed: 68 per cent said getting a complete resolution was the single most important thing in a support interaction. Speed without resolution is not a feature. It is a complaint waiting to happen.
The failure is structural, not cosmetic. Global research from COPC found that the handoff from AI to a human agent is the single most frequent point of failure across markets. In the United States, once an AI interaction failed, customers achieved full resolution only about half the time. The same research surfaced something I would urge every operations leader to internalise: a “transparency dividend.” Customers who were told upfront that they were dealing with AI reported satisfaction rates 34 percentage points higher than those who were not. Honesty, it turns out, is cheaper than pretending.
This is the gap between the metric and reality. The dashboards look healthy. Resolution rate, time to first response, tickets handled per hour, all green. What those aggregate numbers mask is the quality collapse in the interactions that actually matter: the disputes, the fraud claims, the mortgage queries, the moments when a customer is anxious and needs to be understood rather than processed.
Klarna is the case study everyone will be citing for years
You do not need a hypothetical to understand the risk. You have Klarna.
In 2023 and 2024, the Swedish buy-now, pay-later giant became the poster child for AI replacing human work. It said an OpenAI-powered assistant was doing the work of 700 customer service agents, handling roughly two-thirds of all queries, resolving issues in under two minutes against eleven for a human, and on track to add around 40 million dollars in profit. The CEO, Sebastian Siemiatkowski, froze hiring entirely and held the company up as a template for every services business.
By May 2025, he was reversing himself in public. He told Bloomberg the cost-cutting had gone too far and that Klarna had “ended up with lower quality” service. The company began rehiring human agents. His own framing was blunt: the focus on efficiency and cost had reduced quality and eroded customer trust. For a regulated consumer credit business, the inability to handle disputes, fraud, and hardship cases effectively was never a user experience problem. It was a trust and regulatory problem.
Two lessons from Klarna are worth sitting with. First, the volume metrics the AI scored well on masked the deterioration in quality in high-stakes interactions, exactly the failure mode the broader research describes. Second, and this is the part executives never model, the cost of unwinding a bad automation decision, recruiting, onboarding and retraining a human workforce, can exceed the savings that justified the move in the first place. Gartner has separately forecast that by 2027, half of the companies that cut customer service staff because of AI will reverse course and rehire. Klarna simply got there first.
The broader failure-rate context is sobering. Industry research compiled in 2025 and 2026 points to 70 to 85 per cent of AI initiatives falling short of expected outcomes, and an IBM finding that only around one in four AI projects delivers the return it promises. This is not an argument against AI. It is an argument against deploying it as a blunt instrument across interactions of wildly different stakes.
Offshoring carries its own hidden bill
The same cost logic drives the geographic shift, and it carries a similar pattern of hidden costs.
The savings are genuine. Offshoring customer service to destinations such as South Africa or the Philippines can cut operating costs by roughly half, sometimes more. But the 2023 Deloitte Customer Service Excellence study found that 42 per cent of businesses reported a drop in customer satisfaction after offshoring, driven by language barriers, cultural gaps and slower resolution times. Attrition in offshore centres has been reported in the range of 24 to 75 per cent, compared with roughly 12 per cent in the UK, and churn quietly translates into inconsistent training, lost institutional knowledge, and a worse customer experience.
This is the cultural context problem in its sharpest form. When an HR or recruitment function is run by people who do not understand how a candidate’s qualifications, professional bodies or experience translate into the host country’s market, you do not just get a slower process. You get the wrong hires, reject good candidates, and end up with a recruitment pipeline that systematically misreads its own market. The friction is invisible on a cost spreadsheet and corrosive on the balance sheet that matters.
There is a reputational dimension too. A study published in the Journal of Consumer Research, analysing some 35,000 real and simulated cases, found that offshoring provokes significantly more consumer backlash than job cuts driven by automation or domestic restructuring, because customers read shipping jobs abroad as a breach of an unspoken social contract. The brand damage simmers long after the savings are booked.
Tellingly, the market is already correcting. UK research from 2025 found that 34 per cent of customer experience leaders were preparing to move part or all of their operations back to the UK within twelve months, with 73 per cent saying they would reshore if cost were not a factor. The pull factors named were cultural familiarity, customer preference for local support, better staff retention and simpler training. What is making reshoring viable is, ironically, the very technology being blamed for poor service: AI is cutting UK delivery costs from roughly 42 pounds per hour for a human agent to 16 pounds per hour, closing the gap with offshore destinations. Used well, technology is becoming the means by which companies bring quality home rather than the means by which they send it away.
Walk the shop floor, and you see the same trade
The dynamic is not confined to call centres and chatbots. Walk into an understaffed store, and you watch it happen in real time. A customer cannot find a product that is physically in the building, there is no one on the floor to ask, and they leave.
This is not an anecdote; it is measured behaviour. A 2025 survey of 500 US store managers found that 76 per cent had seen an increase in stockouts or empty shelves, and roughly half had cut their workforce in the previous six months. The general manager who ran the research put the consequence plainly: when an item is not where it should be, most customers will not go hunting for help. They simply walk out, and the store never even hears about the lost sale.
The academic work is unambiguous about the cost. Research estimating the impact of understaffing on retail sales and profitability found that aligning staffing with actual traffic patterns, rather than cutting to the bone, delivered around a 6 percent saving in lost sales and a comparable improvement in profitability. Industry analysis in 2025 put customer satisfaction drops in understaffed environments at 18 to 25 percent, correlating with a 12 to 15 per cent revenue decline, and described the doom loop precisely: cost cuts lead to service decline, which leads to revenue decline, which triggers more cost cuts.
That loop is the whole point. The cost reduction that looks so clean on a quarterly deck quietly cannibalises the revenue and profit it was meant to protect.
This is not a case against AI or efficiency. It is a case against doing them carelessly.
The companies getting this right are not the ones refusing to automate or insisting on keeping every job onshore. They are the ones being deliberate about where stakes and capability meet.
A few principles separate the winners from the cautionary tales.
Segment interactions by stakes before automating anything. Informational queries are fair game for full automation. Transactional ones need human review on edge cases. Disputes, fraud, hardship and complex financial decisions need a human in the seat with AI assisting, not the reverse. Klarna’s error was in applying a single automation logic across all three tiers.
Measure the right things. Aggregate resolution rate and handling time will lie to you. Track satisfaction and repeat contact rate on the specific high-value interaction types the AI handles, because that is where quality quietly dies.
Build the human escalation path first, not last. The handoff from machine to human is the most common failure point in the entire industry. If a customer has to battle a chatbot for an hour only to be routed to the agent who should have handled it from the start, you have spent money to manufacture frustration.
Be transparent. The 34-point satisfaction gap between customers who know they are talking to AI and those who do not is one of the cheapest wins available. Tell people.
Invest in cultural training where work is offshored. If teams do not understand the host market’s qualifications, norms, and expectations, no cost savings will offset the cost of contact with a frustrated customer or a misjudged hiring decision.
The strategic point is simple, and it is the one boards keep missing. Cost reduction and efficiency are not the goal. They are inputs to the goal: a durable, profitable relationship with a customer who chooses to come back. When the drive for lower cost degrades the experience, it does not protect margin. It quietly dismantles the revenue the margin sits on.
The race to the bottom has a finish line. It is just not the one anyone meant to cross.
Members Discussion