AI Is 25 Times Cheaper: The Number That Reprices Knowledge Work
On a quarterly spend report, $10,000 that used to flow to a freelancer marketplace now shows up as a few hundred dollars of model usage tied to an expense platform dataset and a handful of vendor names that every CFO now recognizes. The shift looks small in a chart until you translate it into unit economics. Then it feels like a pricing shock that lands in the middle of knowledge work.
Ryan Stevens uses payments data from thousands of firms to track spending from Q3 2021 through Q3 2025, then treats the October 2022 release of ChatGPT as an adoption shock in a difference-in-differences design detailed in a Ramp research paper preprint on arXiv. The result delivers a hard ratio that leaders can apply to real budgets.
Twenty-Five Times Cheaper Changes Everything
Among the most exposed firms, each $1 decline in online labor marketplace spending lines up with about $0.03 of additional spending on AI model providers by Q3 2025, which implies roughly 20–25x cost savings when firms swap outsourced task labor for model usage. That single number deserves more attention than any headline about prompts replacing jobs because it captures the real mechanism that drives adoption inside companies: unit cost.
A 25x gap forces a new kind of budgeting conversation. Contract work typically scales linearly. More output requires more hours, more invoices, and more coordination time. Model usage scales through throughput. A manager pays for tokens and tooling, then pushes volume through a workflow that blends generation, review, and deployment. The spend looks like software spend, yet it often replaces labor spend in categories like copy drafts, first-pass research, support macros, lightweight coding scaffolds, and structured summaries.
Stevens’s dataset shows the spending mix changing alongside that ratio. The share of spending on online labor marketplaces falls sharply over time, while spending on AI model providers rises, reaching 2.85% by Q3 2025. Ramp also summarized the same pattern for a broader audience in its spending analysis, which helps connect the econometrics to what finance teams see in their own general ledger.
The ratio also explains why substitution looks gradual and then speeds up. Early usage often sits in pilots, sandboxes, and individual experimentation. Once a team builds a reliable loop for quality control, the economics pull the work toward AI like gravity. A workflow that turns a $2,000 contractor assignment into $80 of model usage will spread across departments fast. That same math becomes even more compelling when teams pair models with retrieval, templates, and evaluation, because the marginal cost stays low while quality rises.
The bigger point for leaders sits behind the decimal. Three cents on the dollar changes pricing power in every market where language and analysis drive value. It changes how agencies price retainers. It changes how startups staff early functions. It changes how enterprises think about shared services. The labor market reacts to wages and unemployment with a lag. Accounts payable reacts the moment a manager reroutes spend.
The Iceberg Below The Spend Data
The paper captures the visible tip of the iceberg because it measures what firms pay externally through online labor marketplaces and AI model providers in a payments platform. The deeper mass sits underwater inside payroll budgets, internal teams, and embedded contractors that never show up as Upwork or Fiverr line items. In my work with companies adopting generative AI, I keep seeing the same cost curve play out in those internal budgets, where leaders redirect effort rather than cancel a marketplace contract. The savings show up as slower hiring, smaller backfills, shorter project timelines, and fewer outsourced hours in categories that procurement never labeled as “freelance marketplace spend.”
This matters for interpretation. A $1 to $0.03 substitution ratio in a narrow spend category signals a broader capability shift that reaches far beyond the measured slice. The ratio captures direct replacement of purchased task labor. The iceberg includes internal task compression, where one analyst finishes in an afternoon what used to take a week because the model handles the first pass and the human focuses on judgment.
Research on task exposure helps explain why the underwater portion grows quickly. The OpenAI and University of Pennsylvania team behind a recent research paper estimates that about 80% of the U.S. workforce has at least 10% of tasks exposed to LLM capabilities, and about 19% has at least 50% exposure, which frames how wide the potential surface area is. When exposure spreads across roles, substitution can occur through workflow redesign even when the company keeps headcount steady.
Once leaders view the iceberg clearly, the operational priorities sharpen. Finance teams track spend share, yet they also track throughput metrics that reveal internal task compression. HR teams redesign entry roles toward evaluation and domain context. Procurement teams negotiate usage governance and data handling terms with model vendors. Public-sector guidance increasingly treats these capabilities as general-purpose technologies that require management capacity, as emphasized in the OECD’s work on AI adoption in firms.
The iceberg metaphor also clarifies why public debate often feels behind. A marketplace invoice disappearing creates a clean story. A team that quietly ships twice as much with the same headcount creates a subtle story that still drives real labor demand shifts downstream.
A Labor Market Repriced In Real Time
A 20–25x unit cost advantage pushes a repricing of work that touches every layer of the labor market, from gig platforms to entry-level pipelines to professional services. Work that maps cleanly to predictable language output faces direct demand pressure, while complementary work that involves evaluation, domain nuance, and integration gains value.
The most immediate effects show up where tasks are modular and buyers already treat labor as on-demand. A recent platform strategy working paper reports a meaningful decline in job posts for automation-prone categories, including a 21% drop in job posts for automation-prone work in the analysis presented in platform demand shifts. Those movements align with the Stevens finding because they share the same driver: buyers can buy output quality at a far lower unit price.
The next layer hits early-career roles that historically served as training grounds for higher-skill judgment. Stanford researchers using ADP payroll data report that early-career workers ages 22–25 in highly exposed occupations experienced a 16% relative employment decline after the widespread adoption of generative AI, even after controlling for firm-level shocks, as described in their research. That finding fits the iceberg dynamic. Firms gain the first-pass productivity from models, then reserve the remaining human work for people who already hold domain context, trust, and accountability.
Over time, the labor market will adapt through new task bundles. Companies will hire fewer people for pure drafting and more people for evaluation, customer nuance, and system building. Education and training will shift toward model supervision, data stewardship, and applied domain reasoning. Wage premia will flow toward roles that combine judgment with tool mastery, and toward roles that create proprietary feedback loops that raise quality. The same exposure research that highlights breadth also hints at productivity upside when tasks run faster at similar quality.
The Stevens ratio brings discipline to all of this. Leaders do not need to guess whether substitution exists. A three-cent-on-the-dollar signature in real payments data shows it already happening at scale in a measurable slice of the economy. The iceberg view suggests the larger change sits inside internal workflows, where the ledger records outcomes in slower hiring and higher throughput rather than clean vendor swaps.
Conclusion
The most important labor-market statistic in the generative AI era may be a ratio hidden in payments data: $1 of contracted online labor giving way to about $0.03 of model spend among the most exposed firms by Q3 2025 in Stevens’s estimate. That ratio explains why adoption persists even when quality debates continue, because buyers follow unit economics when they can protect output standards through review and governance.
The broader implications extend beyond freelancer marketplaces. The spending shift in the paper captures a visible sliver of substitution. The larger iceberg includes internal task compression, delayed hiring, and redesigned junior roles that reshape career ladders. Companies that treat this as a workforce redesign opportunity, with clear governance and strong training, will convert three-cent inputs into premium outputs. Everyone else will watch their cost structure get repriced by competitors who already did the math.
Key Take-Away
Generative AI reprices knowledge work by turning tasks that once required costly outsourced labor into low-cost model-driven workflows, reshaping hiring, budgeting, and productivity across industries. Share on XImage credit: AlphaTradeZone/pexels
Dr. Gleb Tsipursky, called the “Office Whisperer” by The New York Times, helps tech-forward leaders stop overpaying for AI while boosting engagement and innovation. He serves as the CEO of the AI consultancy Disaster Avoidance Experts. Dr. Gleb wrote seven best-selling books, and his forthcoming book with Georgetown University Press is The Psychology of Generative AI Adoption (2026). His most recent best-seller is ChatGPT for Leaders and Content Creators: Unlocking the Potential of Generative AI (Intentional Insights, 2023). His cutting-edge thought leadership was featured in over 650 articles and 550 interviews in Harvard Business Review, Inc. Magazine, USA Today, CBS News, Fox News, Time, Business Insider, Fortune, The New York Times, and elsewhere. His writing was translated into Chinese, Spanish, Russian, Polish, Korean, French, Vietnamese, German, and other languages. His expertise comes from over 20 years of consulting, coaching, and speaking and training for Fortune 500 companies from Aflac to Xerox. It also comes from over 15 years in academia as a behavioral scientist, with 8 years as a lecturer at UNC-Chapel Hill and 7 years as a professor at Ohio State. A proud Ukrainian American, Dr. Gleb lives in Columbus, Ohio.
