New Study: Businesses Are Replacing Freelancers with AI at a 97% Cost Savings

A new paper from Ramp, “Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI”, provides some of the first hard evidence that businesses are replacing freelancers with AI tools — and the cost savings are staggering.

Unlike most AI-and-jobs studies that rely on surveys or job posting data, this one uses actual company spending data from Ramp’s expense management platform, tracking thousands of firms from 2021 to 2025.

The Key Findings

• More than half of businesses that were spending on freelance platforms (Upwork, Fiverr, Toptal, etc.) in 2022 have completely stopped by 2025.
• Freelance marketplace spending as a share of total spend dropped from 0.66% to 0.14% — a nearly 5x decline.
• AI model provider spending (OpenAI, Anthropic) went from 0% to 2.85% of total spend in the same period.
• The companies that spent the most on freelancers shifted to AI the fastest.

The $1-for-$0.03 Ratio

The most striking finding: firms in the highest exposure quartile (those that previously spent the most on freelancers) substitute $1 of freelance labor for just $0.03 of AI spending. That’s roughly a 33x cost savings, or ~97% reduction in cost.

For the middle quartile, the ratio is $1 of labor to $0.30 of AI — still a significant 70% savings. The true rate likely falls somewhere in between.

As the paper puts it: if a firm was spending $100K on freelance labor, they’re now getting comparable output for $3K–$30K in AI API costs.

How They Measured It

The study uses a difference-in-differences approach, treating the launch of ChatGPT in October 2022 as a natural experiment. The dataset includes firms with spending on platforms like Upwork, Fiverr, Toptal, PeoplePerHour, and others, tracked against spending on OpenAI and Anthropic.

The “exposure” variable is clever: firms that had a higher share of their total spend going to freelance platforms pre-ChatGPT were treated as more “exposed” to AI disruption. The logic is that these firms had the most to gain from exploring AI alternatives, and the data confirms they adopted faster and more aggressively.

Firm and quarter fixed effects control for company-specific trends and seasonal patterns.

Important Caveats

The paper is careful to note several limitations:

• This doesn’t mean all labor gets replaced. The micro-level substitution patterns don’t tell us about aggregate labor market outcomes. Demand for workers who deploy and maintain AI systems could be growing faster than displacement.
• Hidden costs aren’t captured. The $0.03 ratio doesn’t include infrastructure costs for self-hosting models, or increased engineering headcount to build and maintain AI capabilities.
• Only OpenAI and Anthropic spending is tracked. Google, Meta, and Microsoft AI spending couldn’t be cleanly isolated in the data, meaning total AI spend is likely underestimated.
• The study only covers freelance marketplace work — writing, coding, design, and other tasks well-suited for automation. It doesn’t speak to full-time employment substitution.

What This Means

This is one of the clearest empirical signals yet that AI isn’t just changing how work gets done — it’s changing who does it. The freelance marketplace was the canary in the coal mine: tasks that are well-scoped, project-based, and digital-first were the easiest to hand to AI.

The 97% cost savings figure will make headlines, but the more important insight may be the speed: the highest-spending companies moved first and moved fastest. If your business model depends on freelance labor arbitrage, the clock is ticking.