Three years ago the consensus was that generative AI would hollow out offshore services within a decade. Routine cognitive work was the first thing large language models did well, and routine cognitive work is what much of the outsourcing industry was built on. The logic was clean. The numbers have not cooperated.
The Philippines, the second largest outsourcing market in the world, expects its IT and business process sector to reach $42.3 billion in 2026, up from $40 billion in 2025, with headcount rising to 1.96 million full-time workers from 1.9 million. That is not an industry being automated out of existence. It is an industry still adding people in the third year of the technology that was supposed to replace it.
The honest version is more interesting than either the doom forecast or the victory lap. AI did not kill offshoring. It changed what gets sent, who does it and what buyers should pay for.
AI took the task, not the job
The most rigorous evidence comes from online labor marketplaces, where task-level data is visible in a way it never is inside a corporate contract. World Bank research covering more than 2,700 skill areas found that after ChatGPT’s launch, cross-border contracts in the most AI-exposed skills fell 34 percent, and the value of that outsourcing fell 43 percent.
Then look at what happened to the workers who stayed. Average contract value rose 41 percent among those who continued, and the work shifted toward more complex, more interactive tasks. The floor of the market got automated and the ceiling got taller.
The gap between capability and realized savings shows up inside companies too. A Federal Reserve Bank of Atlanta study surveying nearly 6,000 executives across the United States, the United Kingdom, Germany and Australia found that around 70 percent of firms actively use AI, yet more than 80 percent report no impact on either employment or productivity over the past three years. Adoption is close to universal. Measurable results are not, because someone still has to redesign the workflow, check the output and own the exception.
Do not spin the slowdown
Executives should be skeptical of anyone in this industry claiming AI changed nothing. The Philippine industry association recently cut its 2028 targets from $59 billion in revenue and 2.5 million workers to a best case of $50.5 billion and 2.14 million, with a downside scenario of $43.3 billion and 1.85 million. Growth continues. The slope flattened, and the mix moved toward higher-value work.
For company leaders, the useful signal in that revision is not the headline number. It is that the cheapest, most repetitive tier of offshore work is genuinely shrinking, while demand for people who can operate alongside these systems is not. If your sourcing strategy was built entirely on hourly arbitrage for simple tasks, that strategy has a shelf life. If it was built on access to skills you cannot hire locally at a sane price, it just got stronger.
A better test than domestic, nearshore or offshore
Most companies approach the location question backwards. They pick a region first, usually on cost, then try to force roles into it. The better sequence is to sort the work first, then let each category tell you where it belongs.
Start with the cost of being wrong. Work where an error is cheap and quickly caught tolerates distance, asynchronous handoffs and heavy automation. Work where an error is expensive, hard to detect or lands in front of a customer needs to sit close to the people who can catch it, whatever the time zone math says about savings.
Then sort by how much real-time collaboration the role actually requires, not how much it inherited. A role that lives in a queue and produces a deliverable does not need overlapping hours. A role that spends its day in decisions, escalations and half-formed problems does, and that requirement is why nearshore hiring has grown so quickly in the last two years. Three or four hours of genuine overlap is often the difference between a contributor and a bottleneck.
Finally, ask whether the process is documented well enough to hand over. If a role cannot be explained to a capable new hire down the hall, moving it 8,000 miles away will not fix it, and adding AI to it will only produce errors faster.
What this means for the next planning cycle
The companies handling this well are not choosing between AI and people. They are automating the predictable middle of a process, keeping human judgment at both ends, and placing that judgment wherever the talent actually is.
The prediction that AI would end offshoring assumed the value of global talent was low cost per hour. It was never only that. It was access to skills that are scarce and expensive in one market and available in another, and no model has automated that away.
