Will AI Replace Your Outsourcing Partners Before Your Employees?

By Victor Hoang · Updated Oct 10, 2026

When executives ask where AI will take over work first, the usual answer points inward: entry-level analysts, junior writers, support reps on the payroll. The better answer is probably sitting in your accounts payable ledger. It's the agency retainer, the offshore support team, the managed-services contract and the BPO line item.

Our view is simple. Work gets outsourced when it can be written down: a scope, a service level, a unit price, a monthly report. That same property is what an AI system needs to do the work. So the contracts you signed precisely because the work was well specified are the ones AI can undercut first, and the rational order of review is vendor spend before internal roles.

This piece lays out why that order makes sense, what the early evidence says, where it breaks down, and a sequence you can run this quarter.

Key Facts

• Klarna said in February 2024 that its AI assistant handled 2.3 million conversations, two-thirds of its customer service chats, in the equivalent work of 700 full-time agents, resolving errands in under 2 minutes versus 11 previously.
• In May 2025 Klarna's CEO said AI support was cheaper but produced "lower quality", and the company began hiring human agents again.
• Teleperformance, one of the largest call-center outsourcers, guided 2026 revenue growth to between 0% and 2% like-for-like and, in February 2026, named an AI customer-operations specialist as CEO.
• Concentrix guided its fourth quarter of fiscal 2026 to a constant-currency revenue decline of 3% to 5%, and its CEO said the company is "aggressively disrupting our own traditional business."
• A Management Science study of a global freelancing platform found a 21% decrease in job posts for automation-prone writing and coding work within eight months of ChatGPT's launch.
• Stanford researchers using ADP payroll data found employment of 22-to-25-year-olds in AI-exposed occupations 19% below where it would be had it kept pace with less-exposed peers, with no evidence of economy-wide displacement.

Why Codified Work Is the Easiest Work to Replace

Think about what has to be true before a company hands work to an outside party. You need to describe it well enough to put in a statement of work. You need to measure it well enough to hold someone to a service level. You need to price it per ticket, per hour, per deliverable. Work that fails those tests stays in-house, because nobody can write the contract.

Now think about what an AI system needs. A clear definition of the task. Examples of good output. A way to score results. Volume, so the cost per unit matters.

Those lists are nearly the same list. Outsourcing is a filter that has already sorted your work by how well specified it is, and the most specified work ended up outside your walls. The knowledge your own employees hold tends to be the opposite: tacit, relational, spread across judgment calls nobody ever documented. That's harder for any tool to take over, and it's why we expect the order of substitution to run opposite to what most org-chart exercises assume.

There's a second reason, and it's about money rather than technology. A vendor contract is a visible, dated, negotiable cost with a renewal clause. An employee is a person you have to manage out, retrain or redeploy, with morale and legal consequences. Even when the technical difficulty is equal, the contract is the easier lever to pull. A CFO looking at a seven-figure support contract with a 90-day notice period has a clean decision to make. A CFO looking at forty employees has a harder one.

What the Early Evidence Shows

None of the evidence below proves our thesis directly, and we'd rather say so. But it all points the same way, and it's drawn from the places where outsourced and contract work is most visible.

The BPO providers are already feeling it. Klarna's February 2024 announcement is the clearest example of a company moving work from outsourced agents to an AI assistant, with the company's own figures showing two-thirds of chats handled and resolution time falling from 11 minutes to under 2. On the provider side, Teleperformance's 2025 annual results showed group revenue of 10,209 million euros, up 1.3% like-for-like, with a 2026 outlook of 0% to 2% growth. In the same release the board replaced its founder-CEO with an executive whose background was designing AI-native customer operations. A vendor changing its leadership profile that way is telling you how it reads its own market.

Concentrix's third-quarter fiscal 2026 results, published in September 2026, reported revenue of $2,453.7 million, down 1.2% as reported, a goodwill impairment of $1,050 million tied to its share price and market capitalization, and a fourth-quarter outlook implying a 3% to 5% constant-currency decline. Its CEO described the company as aggressively disrupting its own traditional business and said half of revenue comes from business won in the last three years. The filing doesn't isolate how much of the pressure is AI versus pricing or offshoring, so read it as direction rather than attribution.

Contract and freelance work moved first. Researchers studying a global freelancing platform found a 21% decrease in job posts for automation-prone writing and coding jobs, relative to manual-intensive ones, within eight months of ChatGPT's introduction. They also found a 17% decrease for image creation after image-generating AI arrived. A separate CESifo working paper on a large online labor platform found that freelancers in the occupations most affected by ChatGPT saw reductions in both employment and earnings, with similar effects after image-generating models were released. These are individuals rather than agencies, but the mechanism is the same: buyers of codified deliverables found a cheaper substitute.

Employees are affected too, but differently. The Stanford payroll study found that the gap is concentrated among early-career workers in occupations where AI substitutes for tasks, that it comes mainly through reduced hiring rather than separations, and that there's no evidence of widespread economy-wide displacement. Read alongside the vendor evidence, the picture is not "contractors yes, employees no." It's that substitution shows up first where work is most repeatable, and the contract is where repeatable work tends to live.

Where This Thesis Breaks Down

A one-sided argument isn't worth much to a CEO, so here's where we'd push back on ourselves.

Quality can swing back. Klarna's CEO said in May 2025 that AI support was cheaper but produced "lower quality", and the company started hiring humans again so customers could reach a person. That's the strongest argument against a blanket cut. The lesson isn't that AI failed. It's that the right answer was a mix: AI for the simple, high-volume contacts and people for the complex or sensitive ones.

Vendors do more than labor. A good partner brings tooling, benchmarks across clients, compliance coverage, surge capacity and someone else to blame when quality dips. If you only price the headcount, you'll undercount what you lose.

Switching has a cost. Moving a process off a vendor means documenting it, measuring it, and standing up monitoring. Those are real projects, and they land on your internal team.

Vendors will adapt, and they have scale. The large providers are rebuilding around AI themselves. Your leverage is that you can ask them to share the savings. Your own customers will ask for a share of AI savings too, a pressure we look at in customers expect AI savings passed on.

Regulated and relationship work stays human longer. Outsourced work that carries legal accountability, or that sits in front of your most valuable customers, deserves a slower path.

A Framework for Sorting Your Vendor Spend

Rather than ask "outsource or automate?" in general, score each contract on two dimensions. How codified is the work, meaning could you hand a new person a one-page spec and get the right output? And what does an undetected error cost you?

The top three rows are where we'd look first. They share two traits: the output can be judged against a standard, and a mistake is caught by a sample check rather than a lawsuit.

An Illustrative Example

To make the math concrete, here is an invented mid-market company. It spends a fixed amount per year on an outsourced tier-1 support team and a content agency, and employs a small internal operations group. Suppose a pilot shows AI handling a meaningful share of the simple support contacts at an acceptable quality score. The company can renegotiate the support contract's volume tier, shrink the agency retainer to review work, and leave the operations group untouched while it studies what that group's tacit work really is.

Under that sequence nobody on the payroll is dismissed, savings arrive at the next renewal date, and the internal team gets time to learn the tools before its own roles are examined. This is not a forecast about any specific company. It's the order of operations we'd defend to a board.

What to Do Next

For a CEO, COO or CFO, here is a sequence that fits one planning cycle.

1. Pull the vendor ledger. List every outsourced contract above a threshold you choose, with owner, renewal date and notice period. Expect to find contracts nobody has re-scoped in years.
2. Score each contract with the table above. Codification and error cost are enough. Don't build a 40-criteria model.
3. Run paired pilots on the top three. Give an AI-assisted internal owner and the vendor the same sample of work, and compare quality, speed and cost per unit using your existing service levels. Our executive decision framework covers the build-versus-buy logic behind that choice.
4. Calendar the notice periods. The decision window is the renewal notice date, not the day you feel ready. Put those dates in front of the finance team now.
5. Negotiate before you terminate. Ask vendors for outcome-based pricing and shared AI savings. A vendor facing a credible in-house alternative has a reason to listen.
6. Only then review internal roles. Use the AI roles analysis for mid-market companies and the AI-augmented org chart to see what is genuinely at risk inside, and plan redeployment first.
7. Decide where savings go. Whether they become margin or growth capacity is a leadership choice worth making explicitly, and we take it up in AI gains: margin or growth. Finance should also set the measurement rules up front, as outlined in how CFOs measure AI investment returns.

One caveat to hold throughout: the vendor-side evidence is early, much of it comes from customer-service outsourcing, and AI capabilities and pricing keep moving. Plan for the direction of travel, and re-score contracts at each renewal.

Learn More

• Executive Decision Framework for AI Workforce Investments
• Which Roles AI Is Eliminating and Creating in Mid-Market Companies
• The Org Chart of AI-Augmented Departments
• How CFOs Should Measure AI Investment Returns
• AI Gains: Margin or Growth
• Customers Expect AI Savings Passed On