Everyone had the same prediction: AI arrives, headcount drops. Customer service agents, data entry clerks, the people doing repetitive transactional work – surely these were first. But recent data tells a messier story.
A16z recently published an analysis showing that employment in the Philippines' IT and business processing sector rose from 1.15 million workers in 2016 to 1.9 million in 2025 – straight through every major leap in AI capability. GPT-3, GPT-4, the agent wave, all of it. The industry's trade group is projecting another 70,000 jobs in 2026. In the US, Indeed's job-posting data shows customer service roles growing roughly 10 percentage points faster year-over-year than the broader headline figure, with the gap widening since August 2025.
This isn't what the replacement narrative predicted.
The assumption behind "AI replaces workers" is that if a task can be described, it can be automated. There's truth in that, but the economics are more complicated than the slide decks suggest.
Goldman Sachs ran an internal experiment comparing the all-in cost of an AI call center agent versus a human one. AI runs about $92 per day. A human costs roughly $90. For voice-based interactions, where latency and tone sensitivity make the technical problem significantly harder than text, full substitution doesn't pencil out the way it looks on paper.
Klarna found this out directly. In early 2024, the company announced it had replaced 700 customer service agents with AI and called it a success. By May 2025, it was rehiring. Service quality had dropped, customers were getting repetitive and generic responses, and the internal diagnosis was that they had optimized for cost at the expense of actual experience. "We focused too much on efficiency," the CEO told Bloomberg. "The result was lower quality, and that's not sustainable."
The same substitution logic has been applied to finance and accounting, with the same result. AI can process invoices, flag anomalies, and run reconciliations faster than any human. What it can't do is exercise judgment when the numbers don't add up in a way the model wasn't trained on, explain variance to a board, or adapt when a client's business changes in ways that break existing assumptions.
A16z's David George draws a useful line between substitution – AI takes over a task entirely – and augmentation, where AI makes a skilled person more capable.
Substitution cases tend to be narrow: text-based, well-defined, low-judgment tasks with cheap API costs and high volume. Augmentation cases are broader than most people expect. A16z's analysis of Amazon's ebook market found that after ChatGPT's release, it wasn't just AI-generated content that increased – output from established human authors went up too. The people who were good before the tools arrived got more done after.
In finance: a strong FP&A analyst with good AI tooling can model more scenarios and spend more time on interpretation rather than data assembly. A skilled Senior Accountant can handle more complex reconciliations when AI absorbs the routine pass.
But the productivity gain from augmentation is a multiplier on human expertise. When the expertise isn't there, the multiplier doesn't kick in. You just have an AI tool producing outputs that nobody is equipped to validate or act on.
Finance sits in the middle of this dynamic. Some functions are genuinely automatable: basic data entry, standard reporting, routine categorization. Others require exactly the kind of judgment AI still handles poorly: interpreting ambiguous data, navigating edge cases, and standing behind the numbers in front of leadership.
A few things we're observing in practice: companies that cut finance headcount and assumed AI would absorb the slack are finding that errors accumulate in places nobody is monitoring; AI doesn't flag what it doesn't know to look for. Senior finance leaders are getting pulled back into execution because there aren't enough capable operators in the stack, which isn't an AI problem so much as a staffing structure problem that AI has made more visible. And the talent that's hardest to find – mid-level, hands-on finance professionals with five to seven years of experience who can own a process independently – is also the hardest to replace with AI. These are roles where AI would make the person more effective, not substitute for them.
Per YipitData via a16z, less than 20% of B2B companies are now using just one AI vendor. The trend is toward specialized tools for specific functions rather than one platform that handles everything.
The same logic is playing out in talent. Companies are moving toward specialized operators who can own a specific domain – people who know the function well enough to manage the tools, catch the errors, and be accountable for the output. In finance, that means someone who knows month-end close deeply and can operate independently within it. A generalist who loosely knows accounting and loosely knows the software is less useful, not more.
MAVI places pre-vetted, US-caliber global finance and accounting professionals with companies that need capable people quickly. Most placements happen within five days, with talent sourced globally to deliver US-caliber expertise at significantly lower cost.
The companies we work with aren't replacing their teams with AI. They're making sure they have the right people in place to run the finance function well. If that's a problem you're working on, book a call with us.
The data suggests AI is changing the nature of finance work more than eliminating it. Transactional, rules-based tasks are being automated, but demand for experienced finance professionals who can work alongside AI tools is holding steady or growing. The companies seeing problems are the ones that assumed AI could substitute for experienced judgment, not just routine execution.
High-volume, repetitive work – data entry, standard reporting, routine reconciliation – is seeing the most automation. Roles requiring interpretation, judgment, stakeholder communication, and oversight of automated outputs are seeing increased demand.
AI is a multiplier on the people you have. That means the quality of your team matters more than it used to, not less. The question isn't "how many people do I need" – it's "do I have the right people to actually leverage these tools?"
Klarna replaced 700 customer service agents with AI in early 2024 and initially reported success. By May 2025, it was rehiring after service quality dropped. The CEO attributed it to over-indexing on efficiency at the expense of actual customer experience. AI handled the volume but not the complexity or judgment-dependent situations.